One of the great mistakes of the financial crisis was mortgage underwriters and investors thinking they would be OK because their models told them what a bad outcome looked like and that they would be OK under that outcome.

The problem, of course, was the models accounted for “normal” stress and not an extreme stress.

I fear the same thing is happening this year in the prop cat market. Underwriters are ignoring the elevated risk of a historically active season and continuing to price off the cat models.

You would price the odds of who wins the Presidential election differently before the debate vs. afterwards in response to new information.

Well, we have lots of information that suggests it won’t be a normal hurricane season, so why are reinsurers pricing cat treaties like it is a normal year?

When The Facts Change, I Change My Mind

But reinsurers don’t!

While cat models are a very useful tool (and continue to get better), they are not a substitute for one’s better judgment.

The models tell you what is expected to happen over the long run. If you don’t have any special insight into what will happen this year, then, sure, using the long run is a great default choice.

And, of course, there is a fine line between speculating on outcomes using incomplete information and applying “better judgment”.

But it seems perfectly reasonable to operate under the following premise:
“My default is to use the vendor models. However, when there is an accumulation of evidence suggesting we are in a multi standard deviation environment, I will incorporate that knowledge and deviate from the default.”

When you approach it this way, you are not speculating. You are making an informed decision. In many cases, this decision acts as a prudent hedge.

For example, if you took this approach as a mortgage trader in 2006, you would have missed out on gains in 06 but avoided a lot of pain in 07, 08, & 09.

Note, equity investors tend to make this mistake too. I would argue all the time with my internal risk people that my risk was far lower when a stock had fallen from 1.0 BV to .75 BV than from 1.5X to 1.25X.

Yet, the risk models tended to treat them the same (equity risk models tend to overlook valuation floors). Human judgment at extremes can be much more valuable than a model.

As a rough rule of thumb, I would say you’re probably looking to overrule the models once every ten or twenty years.

It isn’t something to undertake lightly, but neither is ignoring warning signs because it’s more convenient to follow protocol.

The Forecasts

Since I’m screaming fire in the movie theater, it’s only reasonable that I highlight why I am so concerned.

Let’s begin with the seasonal forecasts. The Colorado State forecast is the highest they’ve ever put out, calling for 23 named storms, including 11 hurricanes and 5 major hurricanes.

They have never previously called for more than 20 storms and 10 hurricanes. There have been some other years they also predicted five majors.

Only two years this century have produced higher actual results than the current year’s forecast (2005 and 2020). They each produced 7 major hurricanes, a total of 14 (2020) or 15 (2005) hurricanes, and 28 (2005) or 30 (2020) named storms.

So the 2024 forecast is calling for something just short of the two worst actual years. FYI, the forecast for 2005 & 2020 at the time was for 8 or 9 hurricanes with 4 majors, so this forecast is beyond those years’ expectations.

For those who pay close attention, you will instantly recognize 2005 as a terrible year for insured losses (Katrina + Rita + Wilma).

You may not recall 2020 being so bad. That’s because there were few landfalls and most of them came late in the season and away from population centers.

I will have more to say on landfalls below, but note the forecasts are silent on how many US landfalls to expect. If 2024 is similar to 2020, insurers will breath a large sigh of relief.

For thoroughness, there have been four other years this century where we have had 5 or more major hurricanes2004, 2008, 2010, and 2017.

Three of those were pretty bad insured loss years and the fourth (see below) had lots of activity but few losses, so a forecast of five majors is concerning.

I will also point out a couple of other years where there were also forecasts of increased activity with mixed results.

In 2010, the forecast was for 18 storms, 10 hurricanes, and 5 major storms. We ended up at 19, 12, and 5, so very close!

Fortunately, all the major hurricanes that year headed out to sea. Another reminder than an active season may not have many landfalls.

The previous highest forecast was in 2022. That year CSU predicted 20 storms, 10 hurricanes, with 5 major, but we only got 14, 8, and 2. Of course, one of those two was Hurricane Ian.

So if you bet on 2022 as an active hurricane year, you would have lost, but if you bet there would be high insured losses, you were absolutely correct.

One other note before moving on from the forecasts. CSU lists analog years for this season and they are chilling – 2020, 2010, 2005, 1998, 1926, and 1878.

I can’t say much about 1878 (sorry!) but 1926 was the Great Miami Hurricane. 1998 had a cat 5 though not necessarily a bad insured loss year. 2005, 2010, and 2020 I covered above – two of those are the most active years on record and the third was pretty high as well.

I did look back at prior years with above average forecasts, and the analog years back then were not nearly this forboding.

Hurricane Beryl Lessons

Beryl was a massive storm for so early in the season and a particularly scary harbinger of what may be to come.

It was the earliest we have ever had a Cat 5. The prior record? Emily in … 2005! Not the ideal parallel to invoke.

In fact, Emily followed a nearly identical path, though Beryl was slightly stronger (165 vs. 160 mph winds). Besides the hook at the end into Texas, it’s practically a carbon copy.

But that’s not all the bad news! There have only been three other cat 4 or 5 in July since accurate records began – Emily in 2005, also … Dennis right before it in 2005(!), and well before that – wait for it – the 1926 Great Miami Hurricane.

So, for those who might have been a little skeptical about how much to trust the forecasts above, since they can be spotty at times, how are you feeling now?

We’ve already matched (and surpassed) the start of two of the worst years on record, both of which were included by CSU as analogs for 2024. Now do I have your attention?

Of course, an ominous start doesn’t mean the next four months will continue to emulate those terrible years, but it absolutely increases the likelihood that it will.

Steering Currents

I mentioned earlier I would address landfalls. What is the most important thing to know about landalls?

Steering currents. Hurricanes, by nature, want to recurve and stay in the ocean. For there to be a landfall, something needs to get in the way to prevent that.

The most likely thing to get in the way is called the Bermuda High. The Bermuda High is a ridge that sits in the Atlantic that can block storms from going north.

In years like 2010, the Bermuda High is out in the Atlantic leaving plenty of room for storms to reach the edge of it and turn north (the right most arrow above).

However, in years like 2005, the High extends westward and storms get forced into the East Coast or the Gulf.

Guess where the High was sitting for Beryl?

That’s right, it was far to the west leaving Beryl no choice but to head towards the Gulf of Mexico. This is very similar to what Dennis and Emily did to start 2005.

Now, nothing says the High will stay in one place all summer. However, if it continues to be close to the East Coast, an active year will result in landfalls much closer to 2005 than 2020.

Adjusting Pricing

So am I suggesting reinsurers should price cat treaties assuming a 2005 like loss year? Absolutely not.

Am I even suggesting put 50% on 2005 and 50% on 2020? No. The predictive power of the forecasts isn’t high enough to have that level of confidence.

However, would I price to the long term averages? Hell, no! That is not a prudent way to manage risk when the evidence suggests a strong tilt to an above average year.

I know there is the ability to “turn on” the “warm weather” button to generate increased activity in the model and I’m sure some carriers use this.

However, it’s not really sufficient as it doesn’t explicitly model active seasons.

My suggestion would be to replicate the atmospheric conditions of 2005 and 2020 and run 10,000 scenarios for those hurricane seasons.

This gets around the “single outcome” bias for similar seasons. People forget 2005 could have been a lot worse…Rita could have demolished Houston. On the other hand, Katrina could have missed the loop current and not bombed out leading to a much less severe outcome.

Similarly, the steering currents in some of those simulations would result in fewer landfalls in 05 and more in 20.

By looking at a distribution of active season outcomes, one can model AALs and see how they compare to the “normal” AAL. You can also observe how much fatter the tails are in active seasons.

Let’s say hypothetically, the AAL for 05 & 20 is 35% higher than most years but the 1 in 250 is 65% higher. Because there isn’t a 100% guarantee we are repeating those years, you would need to haircut those numbers.

There is obviously some judgment called for, as well as a need to respect market forces, but if the above were correct, maybe you’d raise pricing by 25%???

I’m not trying to arrive at a precise number, but the point is the answer isn’t zero!

Managing Clients

One challenge with my advice is implementation. It’s much harder for a primary commercial insurer to tell it’s real estate client that prices are up 25% this year because of the hurricane forecasts.

The primaries are probably going to have to eat the gross risk and manage their exposure on the back end.

That would lead to increased demand for reinsurance which should let reinsurers charge more. Margins will get squeezed to some extent, but better than not buying and taking a larger net loss.

For reinsurers, while they’ll be happy to sell more at a higher rate on line, they will need more retro which, in turn, will become more expensive.

If it helps, I’d think of it a bit like managing live cat risk. You have to decide whether it’s worth it to give up some margin to hedge out some risk.

Frequency

One more important point. While I suspect many people would default to thinking they should buy more up the tower, I don’t think that’s the main takeaway.

I am far more concerned about frequency. Look again at the forecasts. The number of storms forecast is what is most elevated. The most damaging seasons are the 2005 and 2017 type years where we get multiple big landfalls.

If I were a primary, I’d be buying extra reinstatements (and reinstatement premium protection) and lowering my retention. The capital risk is often higher from taking three retention losses than blowing through the top on one event.

As a reinsurer, I’d be buying more aggregate retro. If that’s not available, I’d issue more regional cat bonds.

If you’re selling retro, I don’t know what to tell you other than…good luck! I guess attach as high as you can to avoid the frequency risk.

Also, if you’re a homeowner and you’re reading these ridiculous news articles telling you to self insure, no, this is not the year for that, especially if you live in Florida, Louisiana, or Texas. Buy as much as you can.

What’s The Worse Way To Be Wrong?

I’m sure some readers are still unconvinced and think they need to slavishly follow the models. After all, the forecasts haven’t always correlated well with actual outcomes, as many are quick to point out.

OK, well let me leave you with this question…

If you’re going to be wrong relative to your assumptions, would you rather be wrong by:

a) buying too much protection that you didn’t need in retrospect
OR
b) not having enough and having much larger losses than expected because you ignored the elevated risk?

If you choose a), your worst case is your earnings are a little lower, but still above average (since this suggests a low cat year).

If you chose b), you are perhaps looking at raising defensive capital, not being able to participate in a better market next year, or even facing a ratings downgrade because of your cavalier risk management.

For me, it’s a no brainer to buy the extra hedges and raise the price of any reinsurance or retro you sell.

The long term winners in the industry are those who outperform the industry during bad cat years, and are thus able to play offense into the hardening market.

You can’t play offense next year, if you don’t play some defense this year.

4 thoughts on “Prop Cat Is Significantly Underpriced This Year”

  1. If Beryl had a come a little earlier – before 1.6 – we might have seen a somewhat different outcome. Otherwise, pricing almost always follows experience, not forecasts.

  2. Very thoughtful piece Ian. The introduction of transactional catastrophe reinsurance capacity (ILS funds, cat bonds, etc.) has significantly weakened the traditional relationship-based reinsurance model. In the old days, as you no doubt know, the idea was that reinsurers would “trade through” volatility with their clients. In so doing, large losses are spread through time. That model has been disrupted by the fact that you can call your broker and buy spot capacity. In this new world, you are absolutely right! It is consistent with the increasingly short time horizon of everything in finance!

    1. Absolutely. If there were similar concern about the Pacific basin, I wouldn’t suggest this as the Japanese still maintain long term relationships.

      But in the US, I think it has to be looked at differently for the reasons you explained.

Comments are closed.