Exposure Rating from Loss Exceedance Curves

Why Exposure Rating Uses a Loss Exceedance Curve

When credible subject-specific claim experience is limited, an exposure rating estimates a reinsurance layer’s expected loss from a benchmark severity distribution and measures of the subject exposure. The central calculation is geometric: the area under the benchmark loss exceedance curve across the layer gives the expected payment per covered occurrence. Exposure adjustments then translate that benchmark indication to the subject risk.

The result depends on what the curve represents. In this tutorial, the benchmark curve describes the severity of a single covered occurrence, conditional on an occurrence taking place. It is not an annual aggregate loss curve. Expected occurrence counts are applied separately to convert expected loss per occurrence into an annual expected loss.

Definition: Let $X$ be the ground-up loss from one covered occurrence, conditional on an occurrence taking place. Its survival function is $S_X(x)=P(X>x)$. A loss exceedance curve (LEC) plots or tabulates $S_X(x)$ against loss amount $x$. For a layer attaching at $A$ and having limit $L$, the payment from one occurrence is $Y=\min((X-A)_+,L)$, where $(z)_+=\max(z,0)$. The expected layer loss per occurrence is $E[Y]$.

A curve may be supplied as a set of points rather than a formula. To calculate areas between points, an actuary must state how the curve is interpolated. Unless another method is specified, the examples below use straight-line interpolation between adjacent points. The benchmark curve is illustrative and assumes no loss above $500,000.

Loss amount $x$	$S_X(x)$, probability loss exceeds $x$
$0	1.00
$100,000	0.60
$200,000	0.30
$300,000	0.10
$400,000	0.03
$500,000	0.00

Layer Loss Is the Area Under the Curve

For any nonnegative loss $X$, the layer payment can be written as the amount of the layer exposed to the loss. At each loss level $x$ between attachment and exhaustion, the layer pays for that increment exactly when the loss exceeds $x$. Taking expectations gives the survival-integral formula:

E[Y]=E[\min((X-A)_+,L)] =\int_A^{A+L}P(X>x)\,dx =\int_A^{A+L}S_X(x)\,dx.

This identity is justified by writing the payment as an integral of indicator functions:

Y=\int_A^{A+L}\mathbf{1}_{\{X>x\}}\,dx.

If $X\le A$, all indicators are zero. If $A<X<A+L$, the indicators are one from $A$ up to $X$, so the integral is $X-A$. If $X\ge A+L$, they are one throughout the layer, so the integral is $L$. Taking expectations and interchanging the nonnegative integral and expectation yields the formula above. This result applies to a nonnegative ground-up severity and a layer whose payment is capped at $L$; it does not, by itself, account for annual aggregate limits, reinstatements, or dependence among multiple occurrences.

With straight-line interpolation, the area over each interval is a trapezoid. If the exceedance probabilities at the ends of an interval of width $w$ are $p_1$ and $p_2$, its contribution is $w(p_1+p_2)/2$. Add partial trapezoids as needed when attachment or exhaustion falls between listed curve points.

Worked Example: Integrating a Benchmark Layer

A subject exposure is expected to produce 4 covered occurrences during the year. Estimate annual expected loss to a $100,000 limit excess of $200,000 using the benchmark curve above. Assume the subject severity follows the benchmark curve without a scale adjustment, occurrences are counted on the same coverage basis as the curve, and expected layer losses are additive across occurrences.

Method. Integrate the curve from $200,000 to $300,000, then multiply the per-occurrence expected loss by the expected number of occurrences.

At $200,000 the exceedance probability is 0.30; at $300,000 it is 0.10. The interval is a single trapezoid with width $100,000:

E[Y]=\$100{,}000\left(\frac{0.30+0.10}{2}\right) =\$100{,}000(0.20) =\$20{,}000

Thus each occurrence contributes an expected $20,000 to the layer. For 4 expected occurrences:

E[\text{annual layer loss}]=4(\$20{,}000)=\$80{,}000.

Check. The per-occurrence expected payment, $20,000, is between $0 and the $100,000 limit; the annual result is four times that amount. A common mistake is to use the exceedance probability at attachment, 0.30, times the full limit, giving $30,000 per occurrence. That treats every loss above attachment as if it exhausts the layer. Integrating the declining curve correctly gives $20,000 per occurrence and $80,000 annually.

Adjusting the Benchmark to Subject Exposure

An exposure adjustment changes the benchmark severity scale to reflect how subject losses compare with benchmark losses. Define $s>0$ so that subject severity $X_s=sX$, where $X$ follows the benchmark curve. A value $s=1$ leaves the benchmark unchanged; $s>1$ scales losses upward and $s<1$ scales them downward. This is a severity scale assumption, not a frequency adjustment.

Because $P(X_s>x)=P(X>x/s)=S_X(x/s)$, the subject expected payment to a layer attaching at $A$ with limit $L$ is:

E[\min((X_s-A)_+,L)] =\int_A^{A+L}S_X(x/s)\,dx =s\int_{A/s}^{(A+L)/s}S_X(z)\,dz.

The second expression follows by substituting $z=x/s$, so $dx=s\,dz$. It gives a practical calculation: divide both attachment and exhaustion by $s$, find the corresponding area under the benchmark curve, and multiply that area by $s$. Equivalently, transform the subject layer endpoints to benchmark loss amounts before reading the curve. The method assumes the entire severity distribution is scaled proportionally; it is not suitable if the subject differs from the benchmark in shape, coverage terms, or tail behavior in a way that a single scale factor cannot represent.

Worked Example: Applying a Severity Scale Factor

For one expected covered occurrence, estimate expected loss to a $150,000 limit excess of $150,000. Subject severity is $X_s=1.5X$, where $X$ has the benchmark curve above. Assume the scale applies to the full ground-up severity and the curve is linearly interpolated.

Method. Divide the subject layer endpoints by 1.5. The subject layer runs from $150,000 to $300,000, which corresponds to the benchmark interval from $100,000 to $200,000:

\frac{\$150{,}000}{1.5}=\$100{,}000,\qquad \frac{\$300{,}000}{1.5}=\$200{,}000.

The benchmark exceedance probabilities at these endpoints are 0.60 and 0.30. The benchmark area is:

\$100{,}000\left(\frac{0.60+0.30}{2}\right) =\$100{,}000(0.45) =\$45{,}000.

Scale that area by 1.5 to obtain the expected subject layer payment per occurrence:

1.5(\$45{,}000)=\$67{,}500.

Check. The layer limit is $150,000, so the expected payment of $67,500 is within the allowable range. A common mistake is to read the unscaled benchmark area, $45,000, as the subject expected payment. The benchmark interval must be scaled back into subject dollars, so the correct result is $67,500.

Exposure Mix: Calculate by Class Before Adding

A portfolio often contains exposure classes with different occurrence counts, severity scales, or layer terms. For class $j$, let $N_j$ be its expected number of covered occurrences and $s_j$ its severity scale. Under the assumption that each class uses its specified scaled benchmark curve, the total expected layer loss is the sum of class-specific expected losses:

E[\text{total layer loss}] =\sum_j N_j s_j \int_{A_j/s_j}^{(A_j+L_j)/s_j}S_X(z)\,dz.

This calculation uses linearity of expectation: expected payments from distinct occurrences or classes add, whether or not those payments are independent. Dependence can matter for the distribution of total loss and capital measures, but not for this sum of expected payments. The formula still requires the frequency and severity inputs to refer to compatible occurrence definitions and coverage terms.

Worked Example: Combining Two Exposure Classes

A hypothetical portfolio has two classes of covered occurrences. Class A has 3 expected occurrences, severity scale $s_A=1$, and a $100,000 limit excess of $100,000. Class B has 2 expected occurrences, severity scale $s_B=2$, and a $200,000 limit excess of $200,000. Estimate total annual expected layer loss using the benchmark curve. Assume the stated frequencies apply to the classes separately and their expected layer losses can be added.

Class A. Its layer corresponds to the benchmark interval from $100,000 to $200,000. The endpoint exceedance probabilities are 0.60 and 0.30, so:

E[Y_A\text{ per occurrence}] =\$100{,}000\left(\frac{0.60+0.30}{2}\right) =\$45{,}000.

For 3 expected occurrences, Class A contributes $3(\$45{,}000)=\$135{,}000$.

Class B. Divide its subject attachment and exhaustion by its scale factor of 2. The subject layer from $200,000 to $400,000 maps to the benchmark interval from $100,000 to $200,000. Its benchmark area is $45,000, as calculated above; multiplying by 2 gives expected loss of $90,000 per occurrence. For 2 expected occurrences, Class B contributes $2(\$90{,}000)=\$180{,}000$.

Total and check.

E[\text{total layer loss}] =\$135{,}000+\$180{,}000 =\$315{,}000.

Each class’s expected payment per occurrence is no greater than its own limit, and the total equals the sum of the two class indications. A common mistake is to use the same benchmark interval for both classes without applying Class B’s scale factor; that would understate its per-occurrence expected loss as $45,000 rather than $90,000.

How Attachment, Limit, and Mix Affect the Indication

For a fixed severity curve and scale, raising the attachment while holding the limit fixed generally reduces expected layer loss: the integral shifts to a higher range where exceedance probabilities are no larger. Increasing the limit while holding attachment fixed generally increases expected loss, because it adds a nonnegative area above the former exhaustion point. These are monotonicity statements, not a claim that the changes are equal or proportional. The effect depends on the shape of the curve across the relevant range.

Exposure mix also matters. A class with greater expected frequency contributes more expected layer loss, and a class with a larger severity scale may move more loss into the layer. For a fixed layer, a single portfolio-average scale can fail to reproduce the sum of class-specific indications because the curve need not be linear in the scale factor. Calculate each materially distinct class separately when its exposure characteristics differ.

Change or assumption	Effect on the exposure indication
Higher attachment, same limit	Usually lowers expected layer loss by removing lower-layer area.
Higher limit, same attachment	Raises or leaves unchanged expected layer loss by adding area.
Higher severity scale	Changes both the benchmark interval represented by the subject layer and the dollar scale of the area.
More expected occurrences	Raises annual expected loss in direct proportion, if severity and terms are unchanged.
Different exposure classes	Can change the result; estimate each class on its own terms and add.

Interpolation, Tail Assumptions, and Common Traps

An LEC is only as useful as its definition and supporting assumptions. Confirm that its probabilities are conditional on a covered occurrence, that loss amounts use the same currency and valuation basis as the contract, and that the curve reflects the same coverage boundaries. If a layer endpoint falls between tabulated points, interpolation is a modeling choice; straight-line interpolation gives trapezoidal area, while another stipulated interpolation rule may give a different result.

A finite table may not specify the curve above its final point. Do not silently assume zero exceedance beyond that point unless the data or problem states that the severity is bounded there. A tail extrapolation, such as a fitted severity model, must be stated and justified. Similarly, an exposure scale factor should not be used as a substitute for an adjustment to expected frequency, and a frequency multiplier should not be mistaken for severity scaling.

Using an exceedance probability as the layer payment. An endpoint probability is a height on the curve, not an expected dollar amount. Integrate across the entire layer.
Multiplying the attachment exceedance probability by the full limit. This assumes every loss that reaches the layer exhausts it and usually overstates the expected payment.
Scaling only the final answer or only the endpoints. Under $X_s=sX$, the benchmark integration endpoints are divided by $s$, and the benchmark area is multiplied by $s$.
Combining unlike classes too early. Apply class-specific frequencies, scales, and layer terms before summing expected losses.
Confusing occurrence and aggregate curves. The curve here is conditional severity for one occurrence. Annual expected loss requires an occurrence count; aggregate covers require an aggregate-loss distribution.

Key takeaway: Exposure rating converts a benchmark occurrence LEC into a layer expected loss by integrating exceedance probabilities over the layer, applying any stated severity scale consistently, and multiplying by the appropriate expected occurrence count. State the curve basis, interpolation, scale, and tail assumptions so the indication can be interpreted and checked.

Check Your Understanding

Use the benchmark curve and straight-line interpolation stated above. Unless a question says otherwise, assume the indicated expected occurrence count is on the curve’s coverage basis.

For one occurrence with no scale adjustment, estimate expected loss to a $50,000 limit excess of $50,000. (Answer: $35,000 per occurrence.)
For one occurrence with no scale adjustment, estimate expected loss to a $100,000 limit excess of $250,000. (Answer: $11,625 per occurrence.)
Subject severity is $X_s=2X$. There are 1.5 expected occurrences, and the layer is a $200,000 limit excess of $200,000. Find annual expected layer loss. (Answer: $135,000.)
Class A has 2 expected occurrences, scale 1, and a $100,000 limit excess of $200,000. Class B has 1 expected occurrence, scale 1.5, and a $150,000 limit excess of $150,000. Find combined annual expected layer loss. (Answer: $107,500.)
With one expected occurrence and no scale adjustment, compare a $100,000 limit excess of $100,000 with a $100,000 limit excess of $200,000. By how much does expected layer loss decrease when attachment rises? (Answer: It decreases from $45,000 to $20,000, a decrease of $25,000.)

What You'll Learn