Deriving Expected Principal Loss

Why Expected Principal Loss Needs Two Inputs

A cat bond’s expected principal loss depends on both how much principal is written down in a loss scenario and how likely that scenario is. A severe event can produce a large write-down but contribute little to the expectation if it is rare. Conversely, a more frequent, moderate write-down can make a material contribution. Keeping scenario severity separate from scenario probability is therefore central to a sound calculation.

The calculations here use a one-year horizon and a bond exposed to at most one qualifying event in that year. The event-loss measure $X$ is the measure used by the bond’s trigger and write-down terms. The result is an expected principal write-down, not a coupon, spread, or investor return calculation.

Definition: Let $C$ be the bond’s principal before the event, $L_A$ the attachment point, and $L_E$ the exhaustion point, with $0 \leq L_A < L_E$. For event loss measure $X$, the principal write-down $W$ is $$ W = h(X) = \begin{cases} 0, & X \leq L_A,\\ C\dfrac{X-L_A}{L_E-L_A}, & L_A < X < L_E,\\ C, & X \geq L_E. \end{cases} $$ The write-down fraction is $W/C$, when $C>0$. The formula assumes a linear write-down between attachment and exhaustion, no write-down below attachment, and full principal exhaustion at or above $L_E$. If the contract specifies different write-down terms, use those terms instead.

From Scenario Write-Downs to an Expectation

For a set of discrete event-loss scenarios, multiply each scenario’s write-down by its probability and add the products. The probabilities must describe the distribution relevant to the quantity being calculated. In particular, probabilities conditional on an event occurring sum to one among event scenarios; they are not the same as annual probabilities of those scenarios.

E[W\mid E] = \sum_j P(X=x_j\mid E)\,h(x_j)

Here $E$ denotes the occurrence of a qualifying event. The expression $E[W\mid E]$ is the expected write-down conditional on an event. If the annual probability of a qualifying event is $q=P(E)$, and $W=0$ when no such event occurs, then the one-year expected principal loss is

E[W] = P(E)E[W\mid E] + P(E^c)\cdot 0 = qE[W\mid E].

This identity separates the event’s annual probability from the average severity given that it occurs. It applies to the stated one-event model; it does not by itself model the effect of multiple events or the contract’s treatment of successive events.

Worked Example: Discrete Event Severities and Annual Expected Loss

A hypothetical cat bond has principal $C=\$40$ million, attaches at $L_A=\$100$ million, and is exhausted at $L_E=\$300$ million. Conditional on a qualifying event, the event-loss measure $X$ is $\$80$ million with probability 0.20, $\$160$ million with probability 0.30, $\$260$ million with probability 0.30, and $\$400$ million with probability 0.20. The annual probability of a qualifying event is $q=0.0400$. Find the expected write-down conditional on an event and the one-year expected principal loss.

Method. First apply the tranche formula to each event-loss value, then probability-weight the write-downs. Finally, multiply the conditional expectation by the annual event probability.

At $X=\$80$ million, the loss is below attachment, so $W=\$0$.

At $X=\$160$ million, the write-down fraction is $(160-100)/(300-100)=60/200=0.30$. Thus $W=0.30(\$40\text{ million})=\$12$ million.

At $X=\$260$ million, the write-down fraction is $(260-100)/200=160/200=0.80$. Thus $W=0.80(\$40\text{ million})=\$32$ million.

At $X=\$400$ million, the loss exceeds exhaustion, so $W=\$40$ million.

The conditional expected write-down is

E[W\mid E] =0.20(0)+0.30(12)+0.30(32)+0.20(40) =0+3.60+9.60+8.00 =\$21.20\text{ million}.

The one-year expected principal loss is

E[W]=0.0400(\$21.20\text{ million}) =\$0.848\text{ million}.

Check. The conditional mean, $\$21.20$ million, is between the smallest and largest possible write-downs, $\$0$ and $\$40$ million. The annual expectation is smaller because it includes the 0.9600 probability of no qualifying event, when the write-down is zero. A common mistake is to report $\$21.20$ million as the annual expectation; that omits the event probability.

Continuous Severity Distributions

If $X$, conditional on an event, has a continuous distribution with density $f_{X\mid E}(x)$, the conditional expected write-down is the probability-weighted area under the write-down function:

E[W\mid E] = \int_{-\infty}^{\infty} h(x)f_{X\mid E}(x)\,dx.

The integral only needs to include values where the conditional density is positive. The write-down function is zero below attachment, increases linearly between attachment and exhaustion, and is constant at principal above exhaustion. Thus, when the distribution has density over those regions, it can be useful to split the integral at $L_A$ and $L_E$. If the distribution has point masses as well as a density, include both the integral over the continuous part and the probability-weighted contributions of the point masses.

Why the expectation formula holds: For a discrete severity distribution, $W=h(X)$ takes the value $h(x_j)$ whenever $X=x_j$. By the definition of expectation, each possible value is multiplied by its probability and the products are summed. For a continuous distribution, the corresponding expectation is the integral of the function $h(x)$ against the probability density of $X$. In either case the write-down is evaluated first; the probabilities then weight those write-down amounts.

Worked Example: Integrating a Uniform Severity Distribution

A hypothetical bond has principal $C=\$30$ million, attachment $L_A=\$100$ million, and exhaustion $L_E=\$200$ million. Conditional on a qualifying event, $X$ is uniformly distributed from $\$50$ million to $\$250$ million. The annual probability of a qualifying event is $q=0.0200$. Calculate the conditional expected write-down and the one-year expected principal loss.

Method. Use the uniform density, $1/200$ per million dollars, and integrate the piecewise write-down function across the severity range. Below $\$100$ million the write-down is zero; from $\$100$ million to $\$200$ million it ramps up; above $\$200$ million it is the full principal.

Using amounts in millions of dollars, the conditional expected write-down is

E[W\mid E] =\frac{30}{200} \left[ \int_{50}^{100}0\,dx +\int_{100}^{200}\frac{x-100}{100}\,dx +\int_{200}^{250}1\,dx \right].

The first integral is $0$. For the ramp-up region,

\int_{100}^{200}\frac{x-100}{100}\,dx =\left.\frac{(x-100)^2}{200}\right|_{100}^{200} =\frac{100^2}{200}-0 =50.

For the full-write-down region, $\int_{200}^{250}1\,dx=50$. Therefore,

E[W\mid E] =\frac{30}{200}(0+50+50) =15\text{ million dollars}.

Multiplying by the annual event probability gives

E[W]=0.0200(\$15\text{ million})=\$0.300\text{ million}.

Check. The conditional expectation is half the bond principal. This is consistent with the uniform distribution: the ramp region contributes an average write-down of half the principal over its range, while the full-write-down region contributes the maximum; together their probability-weighted areas produce a mean fraction of 0.50. The annual result is lower because an event occurs in only 0.0200 of years on average.

Annual Scenarios and Probability of Any Loss

A scenario table can include the no-event outcome directly. In that case, the probability of each scenario is an annual probability, all probabilities sum to one, and the expected write-down is the sum of annual scenario probability times scenario write-down. Do not multiply by a separate event probability again: it is already reflected in the table.

Worked Example: Reading an Annual Scenario Table

A hypothetical cat bond has principal $C=\$20$ million, attachment $L_A=\$100$ million, and exhaustion $L_E=\$300$ million. Its one-year scenario model is:

Annual outcome	Probability	Event-loss measure $X$
No qualifying event	0.94	—
Scenario A	0.03	$\$150$ million
Scenario B	0.02	$\$300$ million
Scenario C	0.01	$\$500$ million

Find the annual probability of a principal write-down, the expected write-down conditional on a write-down, and the one-year expected principal loss.

Method. Calculate each scenario write-down, then weight directly by its annual probability. The probability of any write-down is the sum of the probabilities of scenarios with $X>L_A$.

Scenario A gives a fraction $(150-100)/(300-100)=0.25$, so its write-down is $0.25(\$20\text{ million})=\$5$ million. Scenario B is at exhaustion, so its write-down is $\$20$ million. Scenario C exceeds exhaustion, so its write-down is also $\$20$ million.

The annual probability of a write-down is $0.03+0.02+0.01=0.06$. The annual expected principal loss is

E[W]=0.94(0)+0.03(5)+0.02(20)+0.01(20) =0+0.15+0.40+0.20 =\$0.75\text{ million}.

Conditional on a write-down, the expected amount is

E[W\mid W>0] =\frac{\$0.75\text{ million}}{0.06} =\$12.50\text{ million}.

Check. The conditional mean lies between the possible positive write-downs, $\$5$ million and $\$20$ million. Multiplying $\$12.50$ million by 0.06 recovers the annual expectation of $\$0.75$ million. The probability of a write-down, 0.0600, is not the expected write-down: one is a probability and the other is a dollar amount.

Common Probability-Weighting Traps

The correct calculation depends on identifying what each probability conditions on and what horizon it covers. A scenario table may give probabilities conditional on an event, annual probabilities, or probabilities conditional on a trigger being met. These are not interchangeable.

Using conditional scenario probabilities as annual probabilities. In the first example, the probabilities 0.20, 0.30, 0.30, and 0.20 describe severity given an event. Their weighted write-down, $\$21.20$ million, is conditional severity. The annual expected loss is $0.0400(\$21.20\text{ million})=\$0.848$ million.
Multiplying twice by the event probability. If a table already lists annual probabilities, include its no-event row and sum annual probability times write-down. Do not multiply that total by an event probability a second time.
Averaging event losses before applying the tranche formula. In general, $E[h(X)]$ is not $h(E[X])$, because the write-down function has flat regions and a linear region. Apply the tranche terms to each scenario or integrate the write-down function over the distribution.
Confusing probability of loss with expected loss severity. $P(W>0)$ gives the chance of a positive write-down. It does not state the expected dollar amount or the expected fraction of principal lost.

Key takeaway: For each scenario, determine its write-down under the contract, then weight that amount using probabilities defined for the correct horizon and conditioning event. For a one-event annual model with no write-down when there is no event, annual expected principal loss equals annual event probability times expected write-down conditional on an event.

Check Your Understanding

For each problem, distinguish the stated probability basis from the write-down severity. Amounts are hypothetical.

A bond has $C=\$10$ million, $L_A=\$40$ million, and $L_E=\$140$ million. Given an event, $X$ is $\$20$ million with probability 0.25, $\$90$ million with probability 0.50, and $\$180$ million with probability 0.25. If the annual event probability is 0.0300, find the conditional and annual expected write-downs. (Answer: conditional expected write-down $\$5.00$ million; annual expected principal loss $\$0.15$ million.)
A bond has principal $\$12$ million, attaches at $\$50$ million, and is exhausted at $\$150$ million. Given an event, $X$ is uniform from $\$0$ million to $\$200$ million. The annual event probability is 0.0400. Find the conditional and annual expected write-downs. (Answer: conditional expected write-down $\$6.00$ million; annual expected principal loss $\$0.24$ million.)
A bond has $C=\$25$ million, $L_A=\$100$ million, and $L_E=\$300$ million. Annual outcomes are no event with probability 0.90, $X=\$120$ million with probability 0.04, $X=\$180$ million with probability 0.03, and $X=\$300$ million with probability 0.03. Find the annual probability of a write-down, the expected write-down conditional on a write-down, and annual expected principal loss. (Answer: probability 0.1000; conditional expected write-down $\$11.50$ million; annual expected principal loss $\$1.15$ million.)
A bond has principal $\$8$ million, attachment $\$20$ million, and exhaustion $\$100$ million. Given an event, $X$ is $\$0$ million with probability 0.20, $\$40$ million with probability 0.30, and $\$100$ million with probability 0.50. The annual event probability is 0.0500. Find the conditional and annual expected write-downs. (Answer: conditional expected write-down $\$4.60$ million; annual expected principal loss $\$0.23$ million.)
A bond has $C=\$10$ million, attaches at $\$150$ million, and is exhausted at $\$250$ million. Given an event, $X$ is uniform from $\$100$ million to $\$300$ million. The annual event probability is 0.0100. Find the conditional and annual expected write-downs. (Answer: conditional expected write-down $\$5.00$ million; annual expected principal loss $\$0.05$ million.)

What You'll Learn