Selecting an Expected Loss Ratio

Why the Expected Loss Ratio Matters

The Bornhuetter-Ferguson method uses an expected ultimate loss amount as a prior estimate of the losses for an accident period. When that amount is based on premium, the expected loss ratio (ELR) converts the premium measure into expected losses. Selecting the ELR is therefore not merely choosing a convenient percentage: it requires deciding which assumptions best represent the risks and time period being reserved.

Possible indications include a prospective pricing estimate, adjusted historical experience, or an external benchmark. They need not agree. A sound selection explains the relevance and reliability of each indication, and does not treat unlike data as directly comparable.

Definition: An expected loss ratio is expected ultimate losses divided by the corresponding premium measure. If $P$ denotes that premium and $r$ is the selected ELR, the expected ultimate losses are $P r$. The premium and loss amounts must refer to compatible risks, coverage, time periods, and valuation bases for this multiplication to be meaningful.

The ELR concerns losses, not necessarily all insurance costs. Unless the problem explicitly defines a different ratio, do not include expenses or profit in the numerator. In Bornhuetter-Ferguson, the expected ultimate loss amount is combined with an estimate of the proportion unreported or unpaid; selecting the ELR does not itself determine that development proportion.

Sources of an ELR Indication

A pricing-based indication starts with prospective assumptions. For example, if expected claim frequency per exposure is $f$ and expected severity per claim is $s$, expected loss cost per exposure is $fs$, provided the severity assumption applies to the claims counted by the frequency assumption. Dividing that loss cost by the corresponding expected premium per exposure gives a pricing ELR.

Historical experience can also inform the selection, but an observed or estimated historical ratio may need adjustment before it represents the prospective period. Relevant adjustments can include completion to ultimate and differential trends in losses and premium. The historical data also need enough credibility to be informative. An external benchmark can provide context, particularly when internal experience is limited, but only if the benchmark's risk and measurement basis are reasonably comparable.

Worked Example: Building a Pricing-Based Indication

A hypothetical insurer is pricing a small equipment-protection product. Its assumptions are an expected frequency of $0.08$ claims per exposure, an expected severity of $\$5{,}000$ per claim, and expected premium of $\$640$ per exposure. Assume claim severity has the same expected value for each claim counted, and that claim count and severities satisfy the usual frequency-severity expectation relationship. The calculation is undiscounted; no mortality model or parametric severity distribution is assumed. Find the pricing ELR and the expected losses for 2,500 exposures.

Step 1: Expected loss cost per exposure. Multiply expected frequency by expected severity:

0.08 \times \$5{,}000 = \$400.

Thus the expected loss cost is $\$400$ per exposure.

Step 2: Divide by premium per exposure.

r = \frac{\$400}{\$640}=0.625=62.5\%.

Step 3: Scale to the stated exposure count. Expected premium is $2{,}500 \times \$640=\$1{,}600{,}000$, and expected losses are $2{,}500 \times \$400=\$1{,}000{,}000$. Checking the ratio directly gives $\$1{,}000{,}000/\$1{,}600{,}000=62.5\%$.

The method is a prospective pricing indication. It does not establish that the product will actually produce this ratio; it reports the implication of the stated assumptions.

Adjusting Historical Experience

Historical losses may be immature: reported or paid losses at the evaluation date can be less than their eventual amount. If the available loss estimate is reported at a fraction $q$ of ultimate, a simple completion estimate divides reported losses by $q$, where $0<q\leq 1$. This is appropriate only when $q$ is a suitable estimate for the data and loss basis being used.

If historical losses and premium are then brought to a prospective level, adjust each on its own basis. For example, with annual loss trend $t_L$, annual premium trend $t_P$, and $n$ years between the historical and prospective levels, the adjusted ratio is:

r_{\text{adjusted}} = \frac{L_{\text{historical}}(1+t_L)^n} {P_{\text{historical}}(1+t_P)^n} = \frac{L_{\text{historical}}}{P_{\text{historical}}} \left(\frac{1+t_L}{1+t_P}\right)^n.

This expression assumes the stated annual trends apply over the specified period and that the loss and premium amounts are being moved to the intended common level. If losses and premium are already on that level, do not apply the trend again. If the loss trend differs from premium trend, trending only one side will generally change the ratio incorrectly.

Worked Example: Developing and Trending Historical Experience

A hypothetical portfolio has $\$1{,}200{,}000$ of reported losses and $\$2{,}400{,}000$ of historical earned premium. The reported losses are estimated to represent 90% of ultimate. The historical period is two years before the prospective level; annual loss trend is 5% and annual premium trend is 3%. Treat the completion percentage and trend rates as supplied assumptions. The calculation is undiscounted, uses no mortality assumption, and does not assume a parametric loss distribution. Estimate the adjusted historical ELR.

Step 1: Estimate ultimate losses.

L_{\text{historical}} = \frac{\$1{,}200{,}000}{0.90} = \$1{,}333{,}333.33\ldots

Step 2: Calculate the historical ratio.

\frac{\$1{,}333{,}333.33\ldots}{\$2{,}400{,}000} = 0.555555\ldots = 55.5555\ldots\%.

Step 3: Apply the relative trend for two years. The loss trend factor is $1.05^2=1.1025$, and the premium trend factor is $1.03^2=1.0609$. The relative factor is $1.1025/1.0609=1.039212\ldots$. Therefore:

r_{\text{adjusted}} = 0.555555\ldots \times 1.039212\ldots = 0.577340\ldots = 57.73\%.

Rounded to two decimal places, the adjusted historical indication is 57.73%. As a check, the separately trended amounts are $\$1{,}333{,}333.33\ldots \times 1.1025=\$1{,}470{,}000$ and $\$2{,}400{,}000 \times 1.0609=\$2{,}546{,}160$. Their ratio is approximately $57.73\%$, consistent with the relative-trend calculation.

Credibility and Selecting Among Indications

Credibility is the degree to which an indication is relied upon in a selection. A large, stable, relevant body of experience may support greater reliance on experience than a small or volatile body. An external benchmark or pricing assumption may receive more weight when the internal data are sparse, immature, or not representative of the prospective business. Credibility is not determined by the numerical size of an ELR alone, and a weight should have a stated rationale.

A common practical selection is a weighted average of an experience indication and a benchmark:

r_{\text{selected}}=Zr_E+(1-Z)r_B,\qquad 0\leq Z\leq 1,

where $r_E$ is the experience indication, $r_B$ is the benchmark, and $Z$ is the selected weight on experience. This formula defines an arithmetic blend; by itself, it does not claim that either input is unbiased or that the selected value is statistically optimal.

The arithmetic justification is direct: the selected ratio allocates share $Z$ to one indication and share $1-Z$ to the other, and those shares sum to one. If $r_E$ and $r_B$ are additionally modeled as unbiased random estimates of the same target ratio, then linearity of expectation gives $E[r_{\text{selected}}]=Z E[r_E]+(1-Z)E[r_B]$, which equals the target. Without those assumptions, that expectation claim does not follow. In practice, a credibility weight is a modeling or judgment choice, not automatically a proof of unbiasedness.

Worked Example: Blending Experience and a Benchmark

For a hypothetical reserving portfolio, an adjusted experience indication is 58%, and a comparable external benchmark is 64%. The actuary assigns 75% weight to experience because it is judged more representative, while recognizing that the experience is not fully credible. The applicable premium for the accident period is $\$3{,}200{,}000$. Treat both indications as already stated on the intended basis; no further trend, development, interest, mortality, or distribution assumption is needed for this arithmetic. Find the selected ELR and expected ultimate losses.

Step 1: Blend the indications.

r_{\text{selected}} = 0.75(0.58)+0.25(0.64) = 0.435+0.160 = 0.595 = 59.5\%.

Step 2: Apply the selected ratio to premium.

\$3{,}200{,}000 \times 0.595 = \$1{,}904{,}000.

The selected expected ultimate loss amount is $\$1{,}904{,}000$. As a check, this lies between the two indications: $58\%<59.5\%<64\%$. The method is an explicitly weighted arithmetic blend. It avoids the mistake of treating the blend as a statistical expectation without stating a probabilistic model for the inputs.

Consistency Checks Before Selection

Before assigning weights, examine whether each indication measures the same underlying quantity. A ratio built from one type of premium cannot automatically be applied to another. Likewise, differences in coverage, limits, deductibles, claim definitions, exposure mix, or valuation date can make two ratios incomparable. The next step in a reserving analysis may require closer reconciliation of exposure, premium, and loss bases; here, the selection should at least identify any material mismatch rather than conceal it in the weight.

Consider these questions for each candidate indication:

Is it prospective? Identify whether losses and premium are at historical or future levels and whether required development or trend adjustments have already been made.
Is the experience sufficiently informative? Review volume, volatility, maturity, and whether recent experience is distorted by unusual events or operational changes.
Is the benchmark comparable? A market or broader portfolio ratio may be useful context, but differences in risk mix or coverage can limit its relevance.
Is the premium denominator appropriate? Confirm that the premium measure represents the same intended business basis as the losses before multiplying it by an ELR.
Are adjustments applied once? If an indication is already trended or developed, avoid repeating that adjustment.

Key takeaway: An ELR selection is a reasoned prospective assumption, not simply the latest observed loss ratio. State the source of each indication, make necessary development and trend adjustments, assess credibility, and explain why the selected basis matches the premium to which it will be applied.

Common Selection Trap

A frequent error is to develop historical losses to ultimate and then trend those losses, while leaving the premium unchanged even though premium trend was also specified. This confuses a loss trend with the change in the loss ratio. When both losses and premium move, use their relative trend factor.

For the historical example above, using the historical ratio of 55.5555% and multiplying by the loss trend alone gives $55.5555\%\times1.1025=61.25\%$. That is the wrong prospective ratio when premium also trends by 3% annually. The correction is to divide by the premium trend factor as well:

55.5555\ldots\% \times \frac{1.1025}{1.0609} = 57.73\%.

Another error is to select a blend weight solely because it produces a convenient answer. A weight should reflect the relative relevance and reliability of the indications. If the benchmark and experience describe different coverage or exposure bases, a neat weighted average does not resolve that incompatibility.

Check Your Understanding

For each problem, identify the needed adjustment or selection and check your result before moving on.

A hypothetical product has expected frequency $0.05$ claims per exposure, expected severity $\$8{,}000$, and expected premium $\$500$ per exposure. Assuming the severity mean applies to the counted claims, what is the pricing ELR? (Answer: 80.00%)
A hypothetical portfolio has reported losses of $\$720{,}000$, estimated at 80% of ultimate, and historical earned premium of $\$1{,}500{,}000$. The prospective level is one year later; annual loss trend is 4% and premium trend is 2%. What is the adjusted historical ELR? (Answer: 61.18%)
An experience indication is 55%, a benchmark is 70%, and the selected weight on experience is 40%. What is the blended ELR? (Answer: 64.00%)
A historical ratio of 50% is at a level three years before the prospective period. Annual loss trend is 6% and annual premium trend is 2%. What is the adjusted ratio, to two decimal places? (Answer: 56.12%)
A hypothetical accident period has applicable premium of $\$4{,}000{,}000$. The experience indication is 62%, the benchmark is 54%, and the selected weight on experience is 60%. Find the selected ELR and expected ultimate losses. (Answer: 58.80%; $\$2{,}352{,}000$)

A disciplined selection keeps the numerical calculation and the actuarial judgment distinct: the calculation shows what each indication implies, while the judgment explains which indication deserves reliance for the reserving purpose. This approach is aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

What You'll Learn