Texas changed one word and cut $2.4 billion from its storm fund

My renewal came in June. I read the number twice.

Nothing about my house had changed. The bill had.

You've had the same letter. Or you've had the escrow notice, which is worse, because it comes from your mortgage servicer and the payment you thought was fixed for thirty years goes up anyway. The average escrow shortfall this year runs about $2,157. Over twelve months that's roughly $180 more each month.

So I went looking for where the money goes.

Some of it ends up in a place most people have never heard of.

Thirty-three states run an insurer of last resort. In California it's the FAIR Plan, set up by statute in 1968 for the handful of properties nobody else would write. It was meant to be a few thousand policies… a shelf you reach for when the market says no.

It now carries more than 600,000 policies.

Its wildfire exposure reached $700 billion in March. That is 234% higher than September 2022. The policy count is up 151% over the same stretch.

The FAIR Plan is not a state agency. No tax money goes into it. It runs cash in, cash out. When the cash runs low it can order every private insurer writing property coverage in California to send money. Those insurers can then recover part of that bill from their own customers.

That is not a hypothetical. In February 2025, after the Palisades and Eaton fires, the FAIR Plan asked for a $1 billion assessment on its member insurers. It was the first such call in about thirty years. Regulators let those insurers recoup half of it from policyholders, at $11 to $176 a year for two years.

So a homeowner in Sacramento with no fire exposure paid toward a house in Pacific Palisades. She never signed anything. The surcharge showed up on her renewal.

The insurer of last resort became the insurer of first resort, and nobody voted for it.

The fires get the coverage. What sits underneath them does not.

Because here is what I did not expect to find. While that exposure was more than doubling, the price the capital markets charge to carry catastrophe risk was falling. Not holding. Falling, and fast, and the sellers keep finding more buyers than they asked for.

Two lines going opposite directions. That's the whole issue.

This connects directly to the question of who actually stands behind a state fire pool. The answer, increasingly, is bond investors.

In November the California FAIR Plan came to the catastrophe bond market for the first time. It asked for $250 million of wildfire cover through a deal called Golden Bear Re. Investors wanted more. The target went to $500 million, then to $750 million, and it closed there in December as the largest pure wildfire catastrophe bond ever sold.

The price moved too. Early guidance offered buyers a spread of 10.5% to 11.5%. It priced at 9.75%, about 11% below the midpoint of where it started. Investors took 4.35 times the expected loss on the notes.

Ten weeks later the FAIR Plan came back for $200 million more. That deal closed at $400 million in February. And these notes sat lower in the tower, with an expected loss of 2.65% against 2.24% on the first one.

More risk. Buyers took 3.59 times expected loss.

So in ten weeks, for the same sponsor and the same peril in the same state, the price of wildfire risk fell by about 17%… while the risk being sold went up. That is the anomaly in one transaction.

And here's where it spreads.

A catastrophe bond is a simple trade. You lend money to an insurer. You collect a fat coupon. If the named disaster happens, you lose your principal and the insurer keeps it to pay claims. The coupon is supposed to compensate you for that.

The second quarter of 2026 was the largest quarter the market has ever had. Sponsors sold $11.3 billion of new risk across 48 deals. The first half came to almost $18 billion, beating last year's record. The market outstanding is about $65 billion.

Now the price. The average deal in the quarter paid 2.29 times its expected loss, down from 3.14 a year before. The spread over expected loss fell to 3.74%. Sixty of the seventy tranches priced below the midpoint of their initial guidance.

Sponsors are not complaining. They are issuing more because it got cheap.

A quiet stretch of weather is not the same thing as less risk. But the market prices it that way.

Swiss Re gave the reason in its half-year report. Spreads fell to levels last seen before Hurricane Ian, on strong investor demand and a calm first six months. So the price of the next disaster is being set by the absence of the last one.

Meanwhile, on the Gulf coast, the same pressure produced a different answer.

The Texas Windstorm Insurance Association covers wind and hail for coastal property owners the private market turns away. It insures 286,278 policies and about $127 billion of property. It is the third largest residual plan in the country.

For decades Texas law required it to hold enough money to pay every claim from a one-in-a-hundred-year hurricane. For 2025 that meant a funding level of $6.227 billion.

In June 2025 the legislature changed the word "hundred" to "fifty." House Bill 3689 cut the requirement to a one-in-fifty-year storm. The board set the new funding level at $4.305 billion for 2026 and voted in February to go no higher.

Total resources for this hurricane season come to roughly $3.8 billion, against $6.2 billion last year. In exchange, coastal rates were frozen.

The hurricanes were not consulted. And if the money runs short, Texas has the same tool California used: an assessment on insurers, then a surcharge on everybody else.

Three more things worth keeping track of this week.

Refining margins have come loose from the price of oil. The three-two-one crack spread, which is what a refiner earns turning crude into fuel, ran at 70% to 75% of the value of a barrel of crude in late June. At the start of this year it was around 27%. Crude has fallen and diesel has not followed it down, because the world is short of the thing you actually put in a truck rather than the thing you pump out of the ground. Winter heating demand starts building in October.

On January 1 the Netherlands moved about €550 billion of pension assets into a new defined-contribution system, with another €900 billion scheduled for 2027. Dutch funds hold more than €1.5 trillion between them, and under the old rules they had to match long-dated promises with long-dated bonds. That obligation is gone. De Nederlandsche Bank estimates the sector may shed €100 billion to €150 billion of long-maturity bonds and swaps. Those funds have been the patient buyer at the thirty-year point of the euro curve for twenty years. Watch what the long end does without them.

The container ship orderbook hit a record 11.8 million containers in March, roughly 34% of the fleet already on the water. Freight rates keep sliding. Drewry's Shanghai to Rotterdam leg finished late July near $4,677 for a forty-foot box, down again on the week, with carriers discounting into what should be peak season. Owners are ordering anyway. We'll see.

Before dawn on August 24, 1992, a hurricane came ashore at Homestead, Florida, about twenty-five miles south of Miami. Sustained winds of 165 miles an hour. The strongest storm to hit the United States in almost sixty years.

Ahead of it, one veteran of the insurance business had put a number on paper. A storm of that strength on that coast would cost insurers four to five billion dollars.

The industry's own generous worst case was about eight billion. That was already twice the largest insured loss the country had ever paid.

Andrew cost about fifteen and a half billion. Roughly three times the estimate.

Seven American insurance companies and one foreign company failed. Others stayed open only because a parent wired money in. Florida called a special legislative session the following May to stop carriers from leaving the state.

The reason the number was so wrong is worth sitting with. Insurers priced hurricanes the way you'd price fender benders. They looked at what had happened, took an average, added a margin.

No storm of that size had ever hit Miami-Dade in their records. So the worst case they used described a hurricane that had never occurred in the data. Andrew was that hurricane.

Out of that failure came catastrophe modelling. Physics-based storm simulation, engineering damage curves, thousands of synthetic hurricanes run against actual buildings. A small firm in Boston called AIR Worldwide published an estimate of thirteen billion while adjusters were still driving south. Nobody wanted to believe it. It was the closest number anyone had.

Every price on disaster risk today rests on machinery built because the last price was wrong by a factor of three.

Thirty-four years on, those models set how much capital regulators demand, how much money state hurricane pools must hold, and the expected-loss figure printed on every disaster bond sold to investors. Three firms do most of it.

And they are rebuilt after each surprise. The 2004 and 2005 hurricane seasons changed the assumptions. So did Northridge. So did every event that came in above the curve.

Which means the number in front of you always describes the disasters we have already had.

I have no idea when the next one arrives, or where. The models are far better than what existed in 1992. They are still fitted to a past.

A price is only as good as its worst imagined day. We'll see.