Paul Kedrosky sent in a link to a fabulous article in the NY Times Magazine this last weekend by Michael Lewis (author of Liar’s Poker, Moneyball and many other books that delve into the fascinating overlap of numbers and society) about the development of the market for so-called “catastrophe bonds.” The article is entitled In Nature’s Casino, and focuses on the work of hedge fund manager John Seo in developing risk models for what are, by definition, one-of-a-kind events.
This is a story that hits a number of Radar themes: the value of building prediction models based on data that’s already out there but largely ignored, the way that financial markets progress via “hacks” designed to solve hard problems, and the way that global problems are increasingly being tackled by private enterprise via “collective” action (of which insurance is one of the models, albeit one ignored by most thinkers about Web 2.0 and collective intelligence.) And as you know from the announcement of our Money:Tech conference and the recent Release 2.0 issue on the subject, the connections between Web 2.0 and financial markets is very much on our mind.
The bottom line:
insurance company could function only if it was able to control its
exposure to loss. Geico sells auto insurance to more than seven
million Americans. No individual car accident can be foreseen,
obviously, but the total number of accidents over a large population
is amazingly predictable. The company knows from past experience what
percentage of the drivers it insures will file claims and how much
those claims will cost. The logic of catastrophe is very different:
either no one is affected or vast numbers of people are. After an
earthquake flattens Tokyo, a Japanese earthquake insurer is in deep
trouble: millions of customers file claims. If there were a great
number of rich cities scattered across the planet that might plausibly
be destroyed by an earthquake, the insurer could spread its exposure
to the losses by selling earthquake insurance to all of them. The
losses it suffered in Tokyo would be offset by the gains it made from
the cities not destroyed by an earthquake. But the financial risk from
earthquakes — and hurricanes — is highly concentrated in a few places.
What’s more, the scale of such occurrences could, in fact, bankrupt the insurance industry. This issue first became clear in 1992, when Hurricane Andrew validated the work of an insurance company researcher named Karen Clark. In 1985, Clark had published a paper entitled “A Formal Approach to Catastrophe Risk Assessment in Management”:
To better judge the potential cost of catastrophe, Clark gathered very
long-term historical data on hurricanes. “There was all this data that
wasn’t being used,” she says. “You could take it, and take all the
science that also wasn’t being used, and you could package it in a
model that could spit out numbers companies could use to make
decisions. It just seemed like such an obvious thing to do.” She
combined the long-term hurricane record with new data on property
exposure — building-replacement costs by ZIP code, engineering
reports, local building codes, etc. — and wound up with a crude but
powerful tool, both for judging the probability of a catastrophe
striking any one area and for predicting the losses it might inflict.
Then she wrote her paper about it.
The attention Clark’s paper attracted was mostly polite. Two years
later, she visited Lloyd’s — pregnant with her first child, hauling a
Stone Age laptop — and gave a speech to actual risk-takers. In
nature’s casino, they had set themselves up as the house, and yet they
didn’t know the odds. They assumed that even the worst catastrophe
could generate no more than a few billion dollars in losses, but her
model was generating insured losses of more than $30 billion for a
single storm — and these losses were far more likely to occur than
they had been in the previous few decades. She projected catastrophic
storms from the distant past onto the present-day population and
storms from the more recent past onto richer and more populated areas
than they had actually hit. (If you reran today the hurricane that
struck Miami in 1926, for instance, it would take out not the few
hundred million dollars of property it destroyed at the time but $60
billion to $100 billion.) “But,” she says, “from their point of view,
all of this was just in this computer.”
She spoke for 45 minutes but had no sense that she had been heard.
That is, until 1992, when Hurricane Andrew validated her models. By then, Clark had her own forecasting firm, Applied Insurance Research, and she was the only one who came close to forecasting the true cost of Andrew’s devastation in Florida, which “exceeded all the insurance premiums ever collected in Dade County.”
But once the industry realized that Clark was right, there was still no solution to the problem:
The companies’ models disagreed here and there, but on one point they
spoke with a single voice: four natural perils had outgrown the
insurers’ ability to insure them — U.S. hurricane, California
earthquake, European winter storm and Japanese earthquake. The
insurance industry was prepared to lose $30 billion in a single event,
once every 10 years. The models showed that a sole hurricane in
Florida wouldn’t have to work too hard to create $100 billion in
losses. There were concentrations of wealth in the world that defied
the logic of insurance. And most of them were in America.
The more John Seo looked into the insurance industry, the more it
seemed to be teetering at the edge of ruin. This had happened once
before, in 1842, when the city of Hamburg burned to the ground and
bankrupted the entire German insurance industry many times over. Out
of the ashes was born a new industry, called reinsurance. The point of
reinsurance was to take on the risk that the insurance industry
couldn’t dilute through diversification — say, the risk of an entire
city burning to the ground or being wiped off the map by a storm. The
old insurance companies would still sell policies to the individual
residents of Hamburg. But they would turn around and hand some of the
premiums they collected to Cologne Re (short for reinsurance) in
exchange for taking on losses over a certain amount. Cologne Re would
protect itself by diversifying at a higher level — by selling
catastrophic fire insurance to lots of other towns.
But by their very nature, the big catastrophic risks of the early 21st
century couldn’t be diversified away. Wealth had become far too
concentrated in a handful of extraordinarily treacherous places. The
only way to handle them was to spread them widely, and the only way to
do that was to get them out of the insurance industry and onto Wall
Street. Today, the global stock markets are estimated at $59 trillion.
A 1 percent drop in the markets — not an unusual event — causes $590
billion in losses. The losses caused by even the biggest natural
disaster would be a drop in the bucket to the broader capital markets.
“If you could take a Magnitude 8 earthquake and distribute its shock
across the planet, no one would feel it,” Seo says. “The same
principle applies here.” That’s where catastrophe bonds came in: they
were the ideal mechanism for dissipating the potential losses to State
Farm, Allstate and the other insurers by extending them to the broader
The big challenge, and the one that John Seo has tackled, is pricing. Before catastrophe bonds can work, they need to be priced appropriately, so that, despite the scale and uniqueness of catastrophe, providers of the bonds can raise enough money to actually spread the risk far enough to make “the financial consequences of catastrophe … into something they have never been: boringly normal.”
It’s a fascinating tale of how financial traders developing new instruments are, at their best, hackers working on really hard problems. Well worth reading.