(Crossposted from FT Alphaville.)
“The question is,” said Alice, “whether you can make words mean so many different things.”
In a recent Alphaville post, I made the claim that if the monolines had been required to mark the credit risk that they had taken to market, they would not have played such a prominent role in the financial crisis. Here I want to provide some support for that claim.
There will be several threads to this narrative. We begin with credit spreads.
What’s in a credit spread?
A credit spread is the compensation the taker of credit risk receives for risk. It is well-known that this includes more than just compensation for default risk. Citi research, for instance, produced this illustration recently, showing the default and non-default components of generic BBB credit spreads over time:
They use the term ‘risk premium’ for the non-default component: in reality this component is a mix of compensation for liquidity risk, funding risk, and other factors.
Notice how this non-default component varies over time. What this means is that a holder of credit risk who is marking to market suffers some P&L volatility that is unrelated to default risk (as well as some that is).
The consequences of marking credit to market
If you have to mark a credit risk position to market, then:
- You have to fund losses caused by credit spread volatility;
- You have to support the risk of credit spread volatility with some equity; and
- The risk of the position includes the risk of movements in the non-default component in the credit spread.
A non-mark-to-market holder of the same risk does not have these issues. Depending on their precise accounting standard they may have earnings volatility resulting from changes in perceptions of default, but they won’t have volatility resulting from non-default factors, and thus they don’t need as much equity to support the same position. The need for less equity means that a non-MTM credit risk taker will require a lower return than one who has to mark-to-market.
The history of historic cost
Long ago, the existence of multiple ways of accounting for financial instruments made sense. There were liquid (or at least semi-liquid) securities, and these were marked; there were totally illiquid loans, and these were accounted for based on historic cost (with a reserve being taken if the loan was judged to be impaired). Banks had a buy-and-hold strategy in the loan book, so recognising P&L on an historic cost basis made sense, while marking to market was natural for the flow-based trading book. Insurance companies had approaches* that were similar to historic cost: essentially they recognised premiums as they were paid, and reserved for claims that had been incurred but not yet presented.
Over the 1990s, these boundaries became blurred. Credit default swaps and securitisation liquidified banking book credit risk, and some institutions adopted originate-to-distribute strategies, while others were able to take credit risk in unfunded form by writing credit protection.
This meant that an arbitrage became available whereby the same risk could be taken by both mark-to-market players (by buying a trading book security) and non-MTM players (by writing credit protection which did not have to be marked or making a loan).
So how exactly did you take unfunded credit risk without having to mark it?
Several methods were developed to allow insurance companies to take unfunded credit risk without having to mark it to market.
- In the transformer approach, the insurance company would write a contract of insurance to a SPV which then wrote a CDS. Provided that neither the insurer nor the CDS buyer consolidated the SPV, this provided a compound contract that at one end looked like and was accounted for as insurance, and at the other end looked like and was accounted for as a credit default swap.
- In the wrap approach, the insurance company provided a financial guarantee contract on a bond. If the guarantor was AAA-rated (which the large monolines were pre-crisis), this essentially split the bond into a funding component provided by the buyer of the wrapped bond and a risk component, provided by the insurer.
Insurance companies took credit risk in other ways, too, of course, including some mark-to-market ones; we will come back to this shortly.
Why was taking credit risk in unfunded non-MTM form attractive to some insurers?
The insurance business model is, roughly: take risk by writing insurance, receive premiums, invest the premiums, and pay claims when presented. It works well when the value of invested premiums is larger than that of the presented claims. Given this model, some insurers found credit risk attractive: due to the non-default components of the credit spread, it seemed as if they could get paid more to take credit risk than defaults would cost them, and the structuring technology described above allowed them to do this without having to worry about intermediate earnings volatility caused by having to mark to market. The only question in this business model was ‘do you expect ultimate default losses to bigger or smaller than the value of invested premiums?’
Insurance risk models vary significantly from firm to firm, but what they share is a desire to estimate the capital required to support the risk of unexpectedly large claims. In other words, they assumed that the key risk was the risk of bigger-than-expected claims; something that is perfectly reasonable given the insurance accounting model.
Credit risk taking, then, was potentially attractive to insurers for three reasons:
- It could be made to look like a business model they were familiar with (take premiums, invest them, pay claims);
- It could be accounted for as insurance; and
- The capital required to support some forms of credit risk taking, such as writing protection on asset backed securities, was rather small according to their models.
Was insurance accounting a gateway drug?
It is certainly not that case that most credit risk taken by insurers pre-crisis was non-MTM. AIG, for instance, used fair value accounting on most of the contracts it wrote. However I believe that the availability of the non-MTM model in the early 2000s acted as a kind of ‘gateway drug’, getting some insurers into credit risk taking. Without it, the capital required to support credit risk taking would have been higher, and thus the business would have seemed less attractive. Moreover, the earnings volatility potentially created by having to mark to market** would have at least have given pause for thought at a much earlier stage.
To be fair to insurance accounting standards setters, it is hard to see what they could have done differently. Financial guarantee accounting makes some kind of sense where the guarantor is writing a wrap on an entire municipal bond, and there are no reasonable proxies available. The transformer structure, where a transaction is accounted for as insurance at one end and marked to market as CDS at the other, is less defensible. Arguably, though, the transformer SPV is a major part of the issue, and the rules governing when such things are consolidated have been tightened up (as have the details of financial guarantee accounting). One does wonder, though, what the relevant supervisors had in mind given their evident comfort with the types of practice described here.
Conservation of P&L volatility
Many laws in physics say that in any interaction, some property such as momentum or charge is conserved: the total amount of it in the system remains the same. In a certain sense, moving credit risk from an MTM to a non-MTM player violates conservation of risk. A non-MTM party sees less risk in the deal than an MTM party as the volatility of the non-default-related component of the spread has disappeared. Early 2000s structures such as the one we have described facilitated this, and thus allowed the non-default component of the spread to be monetized.
The introduction of CVA made this situation somewhat better. Non-MTM parties do not get the full benefit of their accounting if they have to post collateral based on the mark of the position – at very least they have to fund the collateral, and that is a drain on their liquidity. So many of monoline trades were done without collateral agreements. This in turn meant that once CVA charges were imposed, some of the volatility of the non-default component of the credit spread reappeared as CVA volatility. This risk hadn’t disappeared after all. Perhaps that is the real lesson: if your trade seems to make risk disappear, there’s something wrong with it.
*Obviously a one sentence account glosses over many complexities and jurisdictional differences.
**Of course, being able to use a model to mark means that much of this volatility can be avoided, at least while the asset credit protection has been written on is not obviously impaired.