Expected Loss Assumptions and Reserve Sensitivity

Why Expected Loss Assumptions Matter

A reserve estimate can change because observed claims change, because development assumptions change, or because the expected losses used as a prior change. This tutorial isolates the third source. It examines how a change in an a priori expected loss ratio or expected loss cost affects an expected-loss estimate and a Bornhuetter-Ferguson (BF) estimate, and why the impact depends on how much of the experience is still unreported.

The distinction matters most for immature experience. A BF estimate uses the prior to estimate the part of ultimate losses that has not yet emerged, while retaining observed claims for the part that has emerged. The prior therefore does not generally affect the entire BF ultimate estimate.

Definition: Let $E$ denote a priori expected ultimate losses for an accident period, on a basis consistent with the data being estimated. Under an expected loss ratio approach, $E=P r$, where $P$ is the relevant earned premium and $r$ is the expected loss ratio. Under an expected loss cost approach, $E=N c$, where $N$ is the number of relevant exposure units and $c$ is the expected loss cost per exposure unit.

The premium, exposure count, expected loss assumption, and observed claims must refer to compatible periods and loss bases. For example, an expected ultimate loss amount should not be compared with observed losses that include different coverage, claim adjustment expense treatment, or trend basis without reconciling those differences.

Bornhuetter-Ferguson and the Unreported Proportion

Let $O$ be the observed paid or incurred losses at the evaluation date, and let $C$ be the cumulative development factor from that evaluation date to ultimate. In the standard BF setup, $C$ is positive and reflects the expected relationship between losses observed to date and ultimate losses. The estimated emerged proportion is $1/C$, so the estimated unreported proportion is $q=1-1/C$. For the usual case $C \geq 1$, $q$ lies between zero and one.

Definition: The BF ultimate estimate is the observed losses to date plus a priori expected ultimate losses multiplied by the estimated unreported proportion:

\widehat{U}_{\mathrm{BF}}=O+qE =O+\left(1-\frac{1}{C}\right)E.

For paid data, $\widehat{U}_{\mathrm{BF}}-O$ estimates total unpaid losses. For incurred data, it estimates IBNR in the basic BF formulation; it is not, by itself, the total unpaid reserve including case reserves.

This form also follows from a blend of the chain-ladder ultimate estimate and the prior. The chain-ladder estimate is $CO$. Weight it by the estimated emerged proportion $1/C$, and weight $E$ by the unreported proportion $q$:

\frac{1}{C}(CO)+\left(1-\frac{1}{C}\right)E =O+\left(1-\frac{1}{C}\right)E.

The observed losses contribute directly, while the prior contributes only for the share estimated to remain unreported. Holding $O$ and $C$ fixed, a change $\Delta E$ in expected ultimate losses therefore changes the BF ultimate estimate by $q\Delta E$. If the estimate sought is the BF IBNR or unpaid component, the same change applies because the observed amount is held fixed.

\Delta \widehat{U}_{\mathrm{BF}}=q\Delta E.

For an expected loss ratio change $\Delta r$ with premium $P$ fixed, this becomes $qP\Delta r$. For an expected loss cost change $\Delta c$ with exposure count $N$ fixed, it becomes $qN\Delta c$. These are one-way sensitivities: they assume observed losses, development factors, and exposure or premium bases do not also change.

Worked Example: Revising an Expected Loss Ratio on Paid Data

A hypothetical accident period has earned premium of $12,000,000. Its selected a priori expected loss ratio is 68%. Paid losses to date are $1,500,000, and the selected paid cumulative development factor is 4.0. Calculate the BF ultimate and unpaid estimate. Then recalculate using an expected loss ratio of 62%, and quantify the sensitivity.

Method and setup. This is a paid BF estimate. The expected ultimate losses at the original ratio are:

$$ E=12{,}000{,}000(0.68)=8{,}160{,}000. $$

The factor implies an emerged proportion of $1/4=0.25$ and an unreported proportion of $q=1-0.25=0.75$. The prior contribution is $0.75(8{,}160{,}000)=6{,}120{,}000$. Thus:

\widehat{U}_{\mathrm{BF}} =1{,}500{,}000+6{,}120{,}000 =7{,}620{,}000.

Subtract paid losses to obtain the unpaid estimate: $7{,}620{,}000-1{,}500{,}000=6{,}120{,}000$. This equals the prior contribution, as it should under the paid BF formula.

At 62%, expected ultimate losses are $12{,}000{,}000(0.62)=7{,}440{,}000$. The revised prior contribution is $0.75(7{,}440{,}000)=5{,}580{,}000$, so the revised ultimate is $1{,}500{,}000+5{,}580{,}000=7{,}080{,}000$, and the unpaid estimate is $5,580,000.

The ratio fell by 0.06, so the expected ultimate fell by $12{,}000{,}000(0.06)=$720{,}000$. The BF reserve falls by $0.75($720{,}000)=$540{,}000$, from $6,120,000 to $5,580,000. The calculation checks. The common mistake is to use $1/C=0.25$ as the unreported proportion; it is the emerged proportion. Doing so would give an incorrect unpaid estimate of $0.25($8{,}160{,}000)=$2{,}040{,}000$.

Maturity Determines How Much the Prior Can Move the Estimate

Since $q=1-1/C$, the BF estimate is more sensitive to its expected-loss prior when the selected development factor implies a larger unreported share. As experience develops and $C$ approaches 1, $q$ approaches 0, and the prior has less effect. This statement holds with the observed amount and selected development factor fixed; in practice, a valuation update can change both.

Worked Example: Comparing Sensitivity at Two Maturities

A hypothetical portfolio has $20,000,000 of earned premium and $7,000,000 of incurred losses to date. The expected loss ratio is revised from 55% to 60%. Compare the impact on the incurred BF ultimate and IBNR at $C=5.0$ and $C=1.25$, holding the observed incurred losses fixed in both calculations.

Method. The prior expected ultimate changes from $20{,}000{,}000(0.55)=$11{,}000{,}000$ to $20{,}000{,}000(0.60)=$12{,}000{,}000$, a $1,000,000 increase.

At $C=5.0$, $q=1-1/5=0.80$. At the 55% ratio, the BF ultimate is $7{,}000{,}000+0.80(11{,}000{,}000)=$15{,}800{,}000$, so BF IBNR is $15{,}800{,}000-7{,}000{,}000=$8{,}800{,}000$. At 60%, the ultimate is $7{,}000{,}000+0.80(12{,}000{,}000)=$16{,}600{,}000$, and IBNR is $9,600,000. The increase is $0.80($1{,}000{,}000)=$800{,}000$.

At $C=1.25$, $q=1-1/1.25=0.20$. At 55%, the ultimate is $7{,}000{,}000+0.20(11{,}000{,}000)=$9{,}200{,}000$, and IBNR is $2,200,000. At 60%, the ultimate is $7{,}000{,}000+0.20(12{,}000{,}000)=$9{,}400{,}000$, and IBNR is $2,400,000. The increase is $0.20($1{,}000{,}000)=$200{,}000$.

The same $1,000,000 change in expected ultimate losses moves the estimate by $800,000 at the larger unreported share and by $200,000 at the smaller one. The arithmetic checks against $q\Delta E$ in each case. The common mistake is to treat a change in the prior as a dollar-for-dollar change in the BF estimate, ignoring the maturity weight.

Expected Loss Cost and Exposure-Based Sensitivity

The same reasoning applies when the prior is built from expected loss cost per exposure rather than an expected loss ratio. The key is to calculate the change in expected ultimate losses on the same exposure base used for the accident period. If exposures are fixed, a per-exposure change is multiplied by the exposure count before applying the BF unreported proportion.

Worked Example: Revising Expected Loss Cost

A hypothetical book has 4,000 exposure units. Its expected loss cost is $350 per unit. Incurred losses to date are $420,000, and the selected incurred cumulative development factor is 3.5. Calculate BF ultimate and IBNR. Then show the result if the selected loss cost is revised to $380 per unit. Report money to the nearest cent.

Method. At $350 per exposure, the expected ultimate losses are $4{,}000($350)=$1{,}400{,}000$. The unreported proportion is $q=1-1/3.5=1-2/7=5/7$. The BF IBNR is:

\frac{5}{7}(1{,}400{,}000)=$1{,}000{,}000.

Therefore, BF ultimate is $420{,}000+1{,}000{,}000=$1{,}420{,}000$. At the revised $380 loss cost, expected ultimate losses are $4{,}000($380)=$1{,}520{,}000$. The revised IBNR is $(5/7)(1{,}520{,}000)=$1{,}085{,}714.29$, and revised ultimate is $420{,}000+1{,}085{,}714.29=$1{,}505{,}714.29$.

The $30 increase per exposure changes expected ultimate losses by $4{,}000($30)=$120{,}000$. Applying the unreported share gives an IBNR increase of $(5/7)($120{,}000)=$85{,}714.29$. This agrees with the difference between the two IBNR estimates. The common mistake is to apply the $30 directly to the reserve without multiplying by exposures and then by the unreported proportion.

Assumptions, Interpretation, and Common Traps

A sensitivity calculation is not a substitute for selecting a defensible prior. It answers a conditional question: how much would this estimate change if the expected loss assumption changed while the other inputs remained fixed? If the revision also reflects different exposure volume, trend, claim handling, coverage mix, or development behavior, those effects may require separate calculations. This tutorial is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Build the prior on the appropriate basis.
Use $E=Pr$ for a loss-ratio prior or $E=Nc$ for a loss-cost prior, with compatible premium or exposure measures.

Determine the unreported share.
For cumulative development factor $C$, calculate $q=1-1/C$. Do not confuse this with the emerged share $1/C$.

Apply the BF prior contribution.
Compute $qE$, add it to observed losses for the BF ultimate, and interpret the difference from observed losses according to whether the data are paid or incurred.

Measure the change consistently.
With observations and development fixed, use $\Delta \widehat{U}_{\mathrm{BF}}=q\Delta E$. State clearly which inputs are being held constant.

An expected-loss estimate without a BF emergence adjustment uses the prior directly as the ultimate estimate, so a change in its prior changes that estimate by the full $\Delta E$. By contrast, the BF estimate uses the prior only for the unreported share. This difference is why the same revised loss ratio or loss cost can have materially different effects under the two methods. Neither method eliminates uncertainty: the sensitivity is conditional on the selected prior, development factor, and exposure basis.

Key takeaway: For a BF estimate with observed losses and a cumulative development factor held fixed, the effect of changing the expected ultimate loss assumption is the change in that assumption multiplied by the estimated unreported proportion, $1-1/C$. Verify whether the requested quantity is ultimate, IBNR, or total unpaid before interpreting the result.

Check Your Understanding

For each problem, hold all stated inputs other than the one being revised fixed.

A hypothetical period has $8,000,000 of earned premium, an expected loss ratio of 62%, incurred losses of $1,100,000, and $C=2.5$. Calculate expected ultimate losses, BF IBNR, and BF ultimate. (Answer: expected ultimate losses $4,960,000; BF IBNR $2,976,000; BF ultimate $4,076,000.)
A hypothetical portfolio has 2,500 exposure units, expected loss cost of $240 per unit, paid losses of $180,000, and paid $C=4$. Calculate the BF unpaid estimate and ultimate. (Answer: unpaid estimate $450,000; ultimate $630,000.)
Earned premium is $15,000,000 and incurred $C=3$. If the expected loss ratio increases from 48% to 53%, by how much does BF IBNR increase? (Answer: $500,000.)
A hypothetical period has 1,800 exposure units, expected loss cost of $520 per unit, incurred losses of $310,000, and $C=1.5$. Calculate expected ultimate losses and BF IBNR. (Answer: expected ultimate losses $936,000; BF IBNR $312,000.)
Two periods each have a $600,000 increase in expected ultimate losses, with observations fixed. One has $C=6$; the other has $C=1.2$. Calculate the BF ultimate change for each and explain which is more sensitive. (Answer: $C=6$, $500,000; $C=1.2$, $100,000; the $C=6$ period is more sensitive because its unreported share is $5/6$, compared with $1/6$.)

What You'll Learn