The multiple describes the average year. The principal at risk describes the year that matters, and the two are linked by the number of years of spread a loss takes to repay.
A catastrophe bond that pays a 7.0% spread over its collateral yield, with a 3.0% annual probability of attachment and 2.0% of exhaustion, has an expected loss of 2.50% and a multiple of 2.8x. The multiple looks comfortable, but one total loss costs 100 points, which is 14.3 years of spread. The chance of at least one total loss in those 14 years is 24.6%.
| Item | Value |
|---|---|
| Principal | 100 |
| Collateral yield (money market, paid to the investor) | 4.0% |
| Spread paid by the sponsor | 7.0% |
| Annual probability of attachment (first loss) | 3.0% |
| Annual probability of exhaustion (total loss) | 2.0% |
The probabilities are those a catastrophe model would report for the layer: the chance that modelled losses reach the bottom of the layer and the chance that they reach the top. The investor is paid the collateral yield and the spread, and loses principal in proportion to the share of the layer that is eaten.
If loss grows roughly in proportion across the layer, the expected loss is the average of the two probabilities. A model gives the exact figure from its loss curve; the average is a first check.
Expected loss = (3.0% + 2.0%) ÷ 2 = 2.50%
Multiple = spread ÷ expected loss = 7.0% ÷ 2.50% = 2.8x
Expected net spread = 7.0% − 2.50% = 4.50%
Excel: =(PoA+PoE)/2, =Spread/EL
The expected return is 4.0% + 4.50% = 8.50% a year, 4.50 points above cash. That is an average over many years, not a return an investor in one bond will see in any given year.
The investor earns 7.0 points a year more than cash. A total loss gives up 100 points of principal, so it takes 100 ÷ 7.0 = 14.3 years of spread to earn it back. Expected loss is only 2.50 points a year, so the average year still pays; but the years in which the bond pays are ordinary and the year in which it does not is not.
| Years | First loss (attachment) | Total loss (exhaustion) |
|---|---|---|
| 3 | 8.7% | 5.9% |
| 5 | 14.1% | 9.6% |
| 10 | 26.3% | 18.3% |
| 14 | 34.7% | 24.6% |
| 20 | 45.6% | 33.2% |
Probability = 1 − (1 − p)n. Over the 14.3 years of spread needed to repay one total loss, the probability of meeting at least one is 24.6%, and the probability of meeting at least some loss, at 3.0% a year, is 34.7%. Held for 14 years, a single bond meets a total loss for about one holder in four and some loss for about one in three.
If a model update raises the expected loss by 40% to 3.50%, the multiple falls from 2.8x to 2.0x and the expected net spread from 4.50% to 3.50%. The coupon has not changed; the bond has become worse.
Twenty identical bonds with independent exhaustion events at 5% weights have the same expected loss, 2.50%. A total loss on one costs 5 points, 0.7 years of spread, and the chance that at least one of the twenty is exhausted in a year is 33.2%, with two or more at 6.0%. The probability of some loss rises; the size of each falls.
The spread divided by the expected loss. A 7.0% spread over a 2.50% expected loss is a multiple of 2.8x.
The principal divided by the annual spread: 100 divided by 7.0 = 14.3 years.
One minus the annual survival probability to the power of the years. At a 2.0% annual exhaustion probability it is 24.6% over 14 years and 18.3% over ten.
The free companion files for The Catastrophe Bond Investor rebuild a fictional UCITS cat bond fund: three bonds, their coupons, expected losses and multiples, a partial loss on an aggregate bond, the portfolio’s VaR and TVaR and the fund’s results. The single bond in this article uses invented probabilities and is not in them. No account and no email address.
Not on Amazon: 24 paperbacks by Julian R. Sterling on insurance and retirement, printed on demand and shipped worldwide. Browse the bookshop →
Reading guide: credit, private credit and distressed debt → · All 1908 articles →