What one parameter told me about the entire cat bond market

I was observing the ILS market yesterday — specifically, the six catastrophe bonds that priced out of Bermuda in recent weeks — and I ran a small experiment I can't stop thinking about.

Every cat bond discloses two numbers when it comes to market: the modelled expected loss (EL), computed by the deal's risk modeller, and the spread investors actually accepted. Between those two numbers sits the entire psychology of the market. So I asked the simplest possible question:

Can one parameter explain how the market converts risk into price?

The tool is thirty years old and criminally underused outside actuarial circles: the Wang transform. Take the EL, convert it to a z-score, shift it by a constant λ — the "market price of risk" — and convert back:

spread = Φ( Φ⁻¹(EL) + λ )

Three lines of arithmetic. I calibrated λ on eight recent tranches spanning expected losses from 1.1% to 7.2% — hurricane, earthquake, wildfire, multi-peril; indemnity, index and ILW triggers.

Here is what fell out.

1️⃣ One number explained eight deals to within 14%.

A single fitted λ reproduced eight different prices — different sponsors, perils and trigger types — with a mean error of 14%. Think about what that means: the cat bond market prices, to first order, off one input (the modelled EL) through one distortion. Whoever computes the more credible EL is holding the pen on the price. Every argument about model quality is, ultimately, an argument about money.

2️⃣ The fitted λ was 0.356 — and that is the market cycle in a single number.

The textbook convention for property-cat sits around 0.45. The market is currently paying well below that per unit of tail risk. You don't need the renewal reports to know we're in a soft market — the calibration tells you — but the reports agree: property-cat rates down double digits at the July renewals. Tracked over time, λ is a one-parameter thermometer of the hard/soft cycle — and arguably a cleaner one than rate-on-line indices, because it controls for the riskiness of what's actually being issued.

3️⃣ The residuals are where the story lives.

Where the one-parameter curve fails is more interesting than where it fits:

→ The California wildfire bond priced ~3 points ABOVE its EL-equivalent curve — a 3.1x multiple when everything else pays ~2x. That is the market openly charging a model-distrust premium on the peril it trusts least.

→ The earthquake aggregate tranches all priced BELOW the curve. Mature models, plus quake diversifies a hurricane-heavy book: a diversifier discount, visible to the naked eye.

→ The riskiest tranche (EL 7.2%) compressed to a 1.7x multiple. Junior cat risk stops behaving like a remote lottery ticket and starts behaving like equity.

Is any of this "arbitrage"? No — and the reason is worth stating precisely. Cat risk is an incomplete market: no replicating portfolio, no borrow to short a bond. There is no unique "correct" price for the curve to estimate — only the observed one. The residuals are relative value, not free money. The wildfire premium might be perfectly rational if the true EL exceeds the vendor's number.

But that cuts the other way too, and it's the thought I'd leave fellow practitioners with: if the market prices off modelled EL, and openly pays three extra points where it distrusts the models, then the durable edge in this market is not trading skill — it is modelling skill. The distrust premium is a standing invitation: demonstrate a wildfire model that survives out-of-sample validation, and the market has already told you, in the price, what that is worth per year.

The honest caveats, because analysis without them is marketing: eight tranches is a small sample; the 14% error is in-sample; I priced from disclosed EL only — no layer shape, no seasonality, no view on whether the vendor ELs are themselves right. The proper next step is the boring one: a bigger sample, calibrate on the past, predict the next issuance before looking, fit λ per peril. If the wildfire premium survives that, it's a factor. If not, it was noise wearing a story.

Still — not bad for three lines of arithmetic and an afternoon of watching the market.

Deal data from the public Artemis.bm deal directory. Views mine; nothing here is investment advice.

A question for the actuaries and financial engineers here: do you read the wildfire premium as rational epistemic pricing — the market correctly charging for model uncertainty — or as a behavioural overshoot after a bad loss year? Your answer determines whether you'd sell it.