Over the past several years deploying capital in the insurance capital markets, I’ve found that most coverage rarely goes beyond surface-level headlines, outcomes, and broad narratives. 

This newsletter takes a different approach - a perspective from inside the market - the pricing cycles, the capital flows, what’s moving markets, and what’s behind the headlines.

Last month's issue, Same fact sheet. Different trade, started from a question a CIO put to me: are catastrophe bonds and specialty insurance the same trade? On a fact sheet they look it - both insurance-linked, both uncorrelated, both pitched as portfolio diversifiers. 

I argued that they’re nothing alike under the surface. One is a risk premium you harvest; the other is a business you back. They behave differently in different markets and they fail in different ways.

It ended on a line that a few readers wrote back about - “the diligence doesn't transfer”. The questions that tell you a catastrophe bond is well-priced are not the questions that tell you an underwriting business is sound.

Which leaves the obvious follow-up hanging - what tools do I use to evaluate insurance-linked investments more generally, and how are they different from the tools most allocators are familiar with?

The metrics most allocators are familiar with do seem to apply to insurance-linked investments if you squint. That’s the trap!

Pull up the fact sheet of an insurance-linked strategy and you'll find something that exists nowhere else in asset management: two languages on the same page.

The first language you likely already speak. Annualised return, volatility, Sharpe ratio, max drawdown, percentage of positive months. Numbers computed from past performance - together they describe how the fund has behaved.

The second language is likely less familiar. Expected loss, value-at-risk at the one-in-two-hundred downside, specific loss scenarios — what the portfolio would lose if a storm the size of the 1926 Great Miami hurricane struck again. This isn't a marketing flourish: insurance-linked strategies are generally sold with an independent modelling firm's risk analysis stating the probability of loss and the size of that loss.

The first language describes the investment's past. The second describes its possibilities. Allocators who aren't familiar with the insurance capital markets tend to be fluent in the first and skim the second - the opposite of what's required. 

Suppose you're paid 7% a year to take a risk that, on average, wipes out half your capital once every fifteen years.

Is this a good trade? Across all outcomes it earns about 3.5% a year over cash - perfectly rational, if you believe that to be adequate compensation for the risk.

Now watch what the standard toolkit says. In any stretch of years where the event doesn't show up, the strategy earns 7% like clockwork, with almost no volatility.

The Sharpe ratio is extraordinary - the best your consultant’s seen. Then the one-in-fifteen year lands…and the same arithmetic now scores the same trade as uninvestable.

Same trade, same price, same risk - two confident answers, both wrong, because the information that matters - the one-in-fifteen year that produces the minus 50% - lives in the bad year, and the sample either contains the bad year or it doesn't.

Insurers settled this long ago, because they had to. An underwriter can't wait for a track record to reveal the risk; by then the risk has happened to them. So the industry measures risk before taking it.

Three questions do the work.

What am I paid? The premium, against the modelled loss, not against last year's outcome.

What can this do to me? Not the average year, the bad one. Regulators make this literal: in Europe an insurer must hold capital against the one-in-two-hundred-year loss. Not as a metaphor, as a number, computed and held against.

When the loss came, did it match the estimate? The industry calls this actual-versus-expected, and it's the closest thing insurance has to a lie detector. A strategy that beats its model in calm years tells you nothing. A strategy whose losses match its model in the loss years tells you the machine works.

If this framework sounds exotic, it isn't; most allocators already run it in one corner of the portfolio. A bond desk doesn't evaluate credit by the volatility of last year's returns. It asks what the spread pays over the expected loss from default.

Investment-grade spreads have historically run at many multiples of realised default losses, and the entire credit literature is an argument about what that gap compensates for.

So the lens transfers…what doesn't transfer is the timing of the loss.

A corporate bond defaults in a recession, precisely when the rest of your portfolio is having its own bad year, which is why credit must pay extra for when it fails, not just whether it fails.

A hurricane doesn’t check the unemployment rate before deciding to wipe out Miami. Two instruments with the identical expected loss are not the same trade if one fails in bad times and the other fails at random.

(The limit here is clearly operational risk - the plumbing and the people. In the autumn of 2008, catastrophe bonds did not prove immune: not because a storm hit, but because a handful of bonds had parked their collateral with Lehman Brothers, and because holders needing cash sell what they can).

Moving from the backward numbers to the forward ones isn’t a free upgrade. You’re swapping risk in one kind of measurement for another. The one-in-fifteen might really be one-in-eight. The discipline just moves from interrogating a return stream to interrogating a model.

In private insurance structures, positions are often valued by the model rather than by a market. A smooth line of monthly returns may mean the risk is calm, or it may mean nobody has asked the question yet.

After a major event, capital can sit segregated for years while the true loss develops. The price not moving is not the same as the risk not moving. Which is why the diligence question changes shape. 

The wrong question - "show me your returns".

The right question - “show me your actual-versus-expected returns”.

The idea of measuring risk as the chance of being destroyed, rather than the bumpiness of the ride, was worked out by a Swedish actuary in 1903. Half a century before modern portfolio theory defined risk as variance, insurance had already defined it as the probability of ruin: can this kill me?

So when an allocator points a Sharpe ratio at an insurance strategy, the irony runs deeper than a mismatched tool. The older, correct instrument was built by this industry, for this risk, and the rest of finance then went off and measured something else.

What were you paid, against what priced risk?

What can a bad year do to you?

When the bad year came, did it match the estimate?

The standing invitation holds. If you’ve been sitting on a question — about a product, a structure, an idea — send it over.

Until then.

Charlie Evans, CAIA
Chief Investment Officer, Zephyros Investments