In partnership with

On a Friday morning in late 1980, a fire started in a restaurant at the MGM Grand Hotel on the Vegas strip. The fire ultimately killed 85 people (around 5,000 were successfully evacuated). Like any disaster, it also produced a cloud of uncertain legal liabilities, and it turned out that MGM had $30m of liability insurance, which they knew would be quickly exhausted. Which led to a strange deal: they bought insurance after the fire, buying another $170m of retroactive coverage for $35m. At that point, MGM's incentive was to settle claims quickly (so they wouldn't get a constant stream of bad press) even if it was costly (it wasn't their money). The insurers, on the other hand, wanted a drawn-out settlement process (to maximize how much they'd earn from investing the premium MGM had paid) and cheap settlements (it was their money). Legal action ensued. This lawsuit was so complicated, and had so many participants, that the judge ordered both sides to contribute to a fund that would build a courtroom large enough to hear the case.¹

The usual model for insurance is to start with a large pool of capital, write policies that produce premiums upfront and lead to more or less predictable losses later on. The capital plus premiums get invested in income-producing assets, so the insurer earns money based on a mix of how well they handicap the odds and how much they can earn between when they get the premium and when they pay claims.² But you can mix and match the sequence of events; marine insurance has the concept of a lost or not lost clause, which provides coverage of disasters that happened before the policy was signed, but that the insurer didn't know about. Of course, given the information asymmetry, this has some serious adverse selection—the shipping company knows a lot more about how lost or not lost a given ship is, and insurers know this, too. Not that this is the only risk they run: they also have to worry about moral hazard, i.e. that if the cost of an accident is lower, the incentive to prevent it is correspondingly lower, too.

Both of these considerations, timing and moral hazard, play an important role in another kind of insurance that's getting talked about a bit right now. Credit default swaps were basically set up as an insurance policy on debt: pay a counterparty a fixed amount (usually quoted as basis points per year) and, in the event that a given credit instrument defaults, the CDS issuer buys it back at 100 cents on the dollar. Since then, they've evolved to a cash-settled model where there's an auction that determines the gap between par and the recovery value for the defaulted debt.³ So, someone who buys a bond and then buys CDS coverage on it has basically paid extra fees in order to create a synthetic low-risk bond. Which seems like a pointless exercise, but is also actually the origin of the credit default swap: in 1994, JPMorgan had given Exxon a large line of credit, and was worried about the risk of a punitive-damages judgment from the Valdez disaster, and that JPMorgan would have to hold more regulatory capital as a result. They could have cut the line of credit. On the other hand, Exxon was almost certainly going to survive, and had been a JPMorgan client for at least 75 years. (You can stop by the Morgan Library and read the whole deal file for their 1919 preferred offering if you want.) A credit default swap was a way for JPMorgan to separate the risk from the relationship.

But, of course, CDS contracts are also a convenient way to short credit, which makes them a popular instrument for speculation and hedging. On the hedging side, more liquid CDS contracts can end up being a rough hedge for a less liquid debt instrument—so someone who is offering a company a line of credit might hedge by buying more actively-traded credit default swaps on their bonds. And, as famously happened during the financial crisis, someone who wants to make a levered bet on defaults can conveniently do so through CDS.⁴

CDS prices are a good barometer of financial stress in that that's what they respond to, but there are cases where they're noisy. In the Exxon case, JPMorgan's willingness to pay was partly a function of how much it valued a longstanding customer relationship; its counterparties would have to fight over how to split the value of that relationship between themselves and JPM. And in a case where a company is not only borrowing heavily but likely to increase its borrowing over time, lenders will look at it the same way: if they get more nervous than they were before, but also think there's a good chance that the company will be doing twice as much business with them a few years from now, they want to express their nervousness by hedging, while telling the company how much faith they have in it. So, CDS prices in a high-growth, capital-intensive field like AI infrastructure are a bet on current risk potentially scaled up by potential future reward.

CDS prices are still a useful shorthand for how risky a company is, and they're worth paying attention to. But they're also an illustration of how financial supply chains work. Putting together a credit deal is a skill; managing a credit portfolio is also a skill. These are correlated—knowing some bond basics really comes in handy for both—but in a complex financial system, there will typically be different firms that excel at one or the other. And one of the ways the good risk-holders make their money is providing liquidity to good loan-originators so they can keep doing what they do best. It's a pretty elegant system, though it does produce some noisy results.

In The Diff, we’ve talked about insurance, credit, and noisy metrics:

• Late last year, in a piece about whether or not agents would disrupt marketplaces, we asked how the financing mix for AI might shift between equity and debt ($). Similarly, we’ve looked at how AI players are looking for different balance sheets for financing ($).

• We’ve covered parametric insurance ($).

• And: America’s financial system is viral ($), and places that do a lot of business with the US tend to slowly absorb US financing and strategy norms.

Speak naturally. Send without fixing.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. No rough transcription you have to clean up. Actual polished text, ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

Start flowing free

Join the discussion!

1  The 80s really were a time of feverish innovation in financing.

2  If there's a lag here—a policy is a lot simpler if the relevant problem hasn't come up yet—then insurers will tend to look artificially profitable if they're growing fast. This naturally creates a perverse incentive, where a growing company can compete on price but might turn out to be taking bad risks that it isn't capitalized well enough to survive, so insurance regulators have to go to great lengths, and a lot of trial and error, to get the balance right. Make the rules too tight, and insurance companies just leave. Make them too loose, and companies can't pay all of their claims.

3  This is just close enough to the previous mechanism that a) if you're just looking at CDS prices, and not trading them, you don't really have to care, and b) if you are actively trading them, this mechanism can result in some exciting trades.

4  On the other side of the trade, one feature of the late stages of the housing bubble was that there wasn't enough borrowing to meet lenders' demand for particular kinds of risk, but there were speculators and hedgers who wanted to make a direct bet on defaults. Banks helped these two meet in the middle through a synthetic CDO. Consider a regular collateralized debt obligation—a big pile of mortgages (themselves wrapped in mortgage-backed securities) that issues some junior debt that takes the first hit in a default in order to create higher-rated slices with a low probability of default. If these CDOs aren't being produced fast enough to meet customer demand, you can make a fake one by sticking together a portfolio of treasury bonds with the same maturity as the CDO, and then selling credit default swaps on the mortgage-backed securities in the original CDO, and tranching them out the same way. Your return profile is the same as buying the CDO, but you don't need to buy the CDO. So, from the perspective of a lender who was itching to take more subprime-backed structured product risk, hedge funds like John Paulson were in the business of creating a synthetic suburb, with abstract, purely financial households. So they were doing something quite helpful, by absorbing demand that would otherwise have created a bigger housing glut after the crisis.