Event Loss Distributions For Cat Bonds

Why Event Loss Distributions Matter

A catastrophe bond layer responds when a covered event loss crosses specified thresholds. To assess how often that might happen, an analyst needs more than the layer’s attachment point and limit: the analyst also needs a probability model for event losses. An occurrence exceedance-probability (OEP) curve summarizes the annual probability that at least one event in a year exceeds each loss amount. Comparing the curve with a layer’s thresholds gives the annual probabilities of reaching attachment and exhausting the layer.

The loss measure must match the contract being analyzed. Depending on the transaction, it could be an indemnity loss, an industry loss, a parametric index, or a modeled-loss measure. The calculations below treat $X$ as the event loss measure to which the layer responds. They do not assume that this measure is the same as an insurer’s ultimate net loss or a physical measure such as wind speed.

Definition: For a specified annual event-loss measure $X$, an OEP curve gives $p_{\mathrm{OEP}}(x)=P(\text{at least one event in the year has loss greater than }x)$, for each threshold $x$. For a layer attaching at $A$ with limit $L$, the exhaustion threshold is $E=A+L$. Ignoring probability mass exactly at a threshold, the annual probabilities of a loss exceeding attachment and exceeding exhaustion are $p_{\mathrm{OEP}}(A)$ and $p_{\mathrm{OEP}}(E)$, respectively.

“Reached” can be used informally to include a loss exactly equal to attachment, even though a layer generally has no positive payment until the attachment is crossed. Similarly, a loss exactly at the exhaustion point may fully exhaust the layer under the contract. An OEP curve is conventionally expressed using strict exceedance, $X>x$. If the model has nonzero probability at exactly a threshold, use the contract’s exact boundary convention and include or exclude that probability accordingly. For continuous loss distributions, the distinction has probability zero.

Reading an OEP Curve Against Layer Thresholds

An OEP curve is an annual probability curve, not simply a list of event severities. For example, if a curve gives a 3% annual probability of exceeding a particular loss, then it estimates the chance that at least one event in a year exceeds that loss. It does not say that 3% of individual events exceed it, and it does not by itself give the probability that the annual sum of all event losses exceeds it.

For a layer attaching at $A$ and exhausting at $E=A+L$, use the curve at both thresholds. The probability of an annual occurrence that crosses attachment but does not exhaust the layer is the difference between the two exceedance probabilities, provided the boundary conventions are consistent. The conditional probability of exhaustion given that attachment was crossed is the exhaustion probability divided by the attachment probability, when the latter is positive.

P(\text{attachment crossed})=p_{\mathrm{OEP}}(A),\qquad P(\text{exhaustion crossed})=p_{\mathrm{OEP}}(E)

P(\text{attachment crossed but exhaustion not crossed}) =p_{\mathrm{OEP}}(A)-p_{\mathrm{OEP}}(E)

Worked Example: Reading Attachment and Exhaustion From a Curve

A hypothetical cat bond has an attachment point of $200 million and a limit of $200 million, so its exhaustion point is $400 million. A model reports annual OEP values of 5.0% at $200 million and 1.5% at $400 million. Estimate the annual probabilities of crossing attachment, crossing exhaustion, and crossing attachment without exhausting the layer. Also calculate the probability of exhaustion conditional on crossing attachment.

Method. Read the annual exceedance probability at each threshold. Subtract the two probabilities for the attachment-only range, then divide the exhaustion probability by the attachment probability for the conditional probability.

The exhaustion threshold is $E=A+L=\$200\text{ million}+\$200\text{ million}=\$400\text{ million}$. Therefore:

P(\text{attachment crossed})=0.050,\qquad P(\text{exhaustion crossed})=0.015

The probability of crossing attachment but not exhaustion is:

$$ 0.050-0.015=0.035 $$

The conditional probability of exhaustion, given that attachment is crossed, is:

\frac{0.015}{0.050}=0.300

Thus the annual attachment-crossing probability is 5.0%, the exhaustion-crossing probability is 1.5%, and the attachment-only probability is 3.5%. Conditional on crossing attachment, the chance of also crossing exhaustion is 30.0%. Check: the attachment-crossing probability partitions into 3.5% that do not exhaust and 1.5% that do, totaling 5.0%. The common mistake avoided is treating the 1.5% exhaustion probability as conditional rather than annual.

From Event Frequency to Annual OEP

Sometimes the input is not an annual OEP curve but an event frequency and a severity distribution. A standard model assumes that the number of events in a year is Poisson with mean $\lambda$, and that event severities are independent and identically distributed, independent of the event count. Here $\lambda$ is the expected number of events per year, and $S_X(x)=P(X>x)$ is the probability that one event exceeds threshold $x$.

Under these assumptions, the number of events exceeding $x$ is also Poisson, with mean $\lambda S_X(x)$. An annual exceedance occurs if this count is at least one. The probability of no exceedances is the probability that a Poisson random variable with mean $\lambda S_X(x)$ equals zero.

Theorem: If the annual event count is Poisson with mean $\lambda$, and independent event severities have exceedance probability $S_X(x)$, then

p_{\mathrm{OEP}}(x)=1-\exp\left(-\lambda S_X(x)\right)

This result is specific to the stated frequency-severity model. It does not apply automatically when event counts are not Poisson, severities depend on the number of events, or the input probabilities refer to annual totals rather than individual events.

Find the event exceedance probability.
At each layer threshold $x$, determine $S_X(x)$, the probability that one event exceeds $x$.

Calculate the exceedance-event rate.
Multiply the annual event rate by the severity tail probability: $\lambda S_X(x)$.

Convert the rate to an annual probability.
Use $1-\exp(-\lambda S_X(x))$, the probability of at least one exceedance during the year.

Compare the layer thresholds.
Calculate the annual OEP at attachment and exhaustion; their difference is the probability of crossing attachment but not exhaustion.

Worked Example: Poisson Event Frequency and Severity

A hypothetical catastrophe model assumes a Poisson event count with mean $\lambda=0.40$ events per year. For the contract’s event-loss measure, the probability that one event exceeds a $100 million attachment point is 0.25, and the probability it exceeds the $250 million exhaustion point is 0.08. Calculate the annual probabilities of crossing attachment and exhaustion, and the probability of crossing attachment without exhaustion.

Method. Apply the Poisson OEP formula separately at each threshold. The event exceedance rates are $\lambda S_X(x)$.

At attachment, the annual rate of events exceeding $100 million is:

$$ 0.40(0.25)=0.10 $$

Thus the annual attachment-crossing probability is:

1-\exp(-0.10)=1-0.904837=0.095163

At exhaustion, the exceedance-event rate is:

$$ 0.40(0.08)=0.032 $$

The annual exhaustion-crossing probability is:

1-\exp(-0.032)=1-0.968507=0.031493

The probability of crossing attachment without crossing exhaustion is:

$$ 0.095163-0.031493=0.063670 $$

To four decimal places, the probabilities are 0.0952, 0.0315, and 0.0637, respectively. Check: 0.0637 plus 0.0315 is 0.0952 after rounding. The calculation avoids the common error of using $\lambda S_X(x)$ itself as an annual probability; that product is an exceedance-event rate, not the probability of at least one exceedance.

Using Discrete Event-Loss Scenarios

A scenario set can represent event losses by categories, each with an annual occurrence rate. If each category follows an independent Poisson count, add the rates of all categories that exceed a threshold, then convert the total rate to an annual OEP. This makes the threshold comparison explicit and avoids treating a scenario’s annual rate as if it were already an annual probability.

The scenario rates should represent mutually distinct event categories if they are added in this way. If scenarios overlap, or if their weights are not occurrence rates, adding them may double count events or produce an invalid rate. The modeling convention and the time unit must be checked before calculation.

Worked Example: Scenario Rates for a Cat Bond Layer

A hypothetical event catalog has four loss categories for one event: $80 million at an annual rate of 0.20, $140 million at 0.12, $220 million at 0.06, and $360 million at 0.02. Assume independent Poisson counts by category. A layer attaches at $150 million and has a $150 million limit, so it exhausts at $300 million. Estimate the annual probabilities of crossing attachment, crossing exhaustion, and crossing attachment without exhaustion.

Method. Select the scenario categories above each threshold, sum their annual rates, and use the Poisson probability of at least one qualifying event.

The exhaustion point is $150+150=\$300$ million. The scenarios exceeding the $150 million attachment are $220 million and $360 million, so their combined annual rate is:

$$ 0.06+0.02=0.08 $$

The annual probability of at least one event exceeding attachment is:

1-\exp(-0.08)=1-0.923116=0.076884

Only the $360 million category exceeds the $300 million exhaustion point. Its annual rate is 0.02, giving:

1-\exp(-0.02)=1-0.980199=0.019801

Therefore, the probability of crossing attachment without crossing exhaustion is:

$$ 0.076884-0.019801=0.057083 $$

To four decimal places, the attachment-crossing probability is 0.0769, the exhaustion-crossing probability is 0.0198, and the attachment-only probability is 0.0571. Check: the probabilities of attachment-only and exhaustion sum to 0.0769 after rounding. The common mistake avoided is using the rate 0.08 as the annual probability, rather than converting it with the Poisson formula.

Interpretation, Assumptions, and Exam Traps

An OEP curve describes whether at least one event exceeds a threshold during a year. It does not sum the losses from multiple events. That distinction matters when a layer responds to a single event: the event-level threshold and annual OEP are directly relevant. If a contract instead responds to an annual aggregate, the relevant probability model must account for the sum of losses; an OEP curve alone is not enough.

The model’s basis also matters. The loss curve must correspond to the contract’s covered event definition, loss metric, and time horizon. A curve for gross industry loss cannot automatically be used for an indemnity trigger based on a cedent’s net loss. Likewise, a curve reported per event cannot be treated as an annual probability without an event-frequency model.

Do not confuse event rates with annual probabilities. Under a Poisson model, a rate $r$ converts to probability $1-\exp(-r)$.
Do not confuse OEP with annual aggregate exceedance. OEP asks whether any one event exceeds a threshold; annual aggregate probability concerns the sum across events.
Check what the curve measures. The modeled loss must be on the same basis as the layer’s contractual trigger.
Respect threshold conventions. If losses can equal a threshold with positive probability, confirm whether the contract uses $>$ or $\geq$.
Keep rates and probabilities on the same time basis. An annual event rate cannot be combined directly with a monthly or multi-year severity probability without adjustment.

For a continuous loss distribution, an OEP probability at the attachment point estimates how often the layer begins to respond, while the probability at exhaustion estimates how often the layer is fully consumed by an occurrence. These are useful frequency summaries, but they do not by themselves determine the expected amount of principal loss. That requires information about the sizes of losses between and beyond the thresholds.

Key takeaway: Match the loss curve to the contract, identify attachment and exhaustion, and use annual OEP values at both thresholds. If starting from Poisson event frequencies and event severities, first calculate the exceedance-event rate and then convert it to an annual probability with $1-\exp(-r)$.

Check Your Understanding

For each problem, use the stated convention that threshold probabilities refer to strict exceedance, and round probabilities to four decimal places where needed.

A hypothetical layer attaches at $120 million and has a $180 million limit. Its annual OEP curve gives 0.072 at $120 million and 0.018 at $300 million. Find the attachment-crossing probability, the exhaustion-crossing probability, and the probability of crossing attachment without exhaustion. (Answer: 0.0720, 0.0180, and 0.0540.)
Annual event counts are Poisson with mean 0.30. A single event has probability 0.20 of exceeding attachment and 0.05 of exceeding exhaustion. Calculate both annual OEP probabilities and the probability of attachment without exhaustion. (Answer: 0.0582, 0.0149, and 0.0433.)
A scenario catalog gives event losses of $90 million, $180 million, $280 million, and $450 million, with respective annual Poisson rates 0.15, 0.10, 0.04, and 0.01. A layer attaches at $200 million and exhausts at $400 million. Find the annual attachment-crossing and exhaustion-crossing probabilities, then their difference. (Answer: 0.0488, 0.0100, and 0.0388.)
A model explicitly specifies linear interpolation in annual OEP probability between its tabulated points: 0.12 at $100 million, 0.06 at $200 million, and 0.02 at $300 million. A hypothetical layer attaches at $150 million and exhausts at $300 million. Estimate the attachment-crossing probability, exhaustion-crossing probability, and attachment-only probability. (Answer: 0.0900, 0.0200, and 0.0700.)
An annual OEP curve gives 0.080 at a layer’s attachment and 0.025 at its exhaustion. Calculate the probability of crossing attachment but not exhaustion, and the conditional probability of exhaustion given attachment crossing. (Answer: 0.0550 and 0.3125.)

What You'll Learn