Skip to main content

S&P Gets In Trouble For Giving AAA Rating To Obviously Terrible Thing

  • Author:
  • Updated:

Oh man, CPDOs. CPDOs! Why was I not aware? This Australian court decision is like 3,000 pages long but it is riveting; if you built a CPDO, email me, I will buy you a drink and you can tell me all about it. My God it's so beautiful.

The story is that ABN Amro invented a structured-credit monstrosity called a constant proportion debt obligation, got it rated AAA by S&P, and sold it to some people; it ended up in the hands of some Australian regional councils, and then it chewed their hands off. As well it might have! It was monstrous. Anyway the councils sued S&P (and others) and today they won their lawsuit, which is bad news for S&P, though they kind of deserved it.

To simplify enormously the CPDO deal was:

  • You are a ten-year pool of money
  • You make a levered investment in some 5-year investment grade credit indices
  • Every six months you roll that investment into the next 5-year index
  • If credit has widened, you have lost money and therefore lever your investment more, to try to make your money back
  • If credit has tightened, you have made money and therefore ratchet down your leverage, hoping to get out in one piece
  • If you keep winning you take more money off the table until you end up with your money in Treasuries for the remainder of the 10 years
  • If you keep losing you run out of money and just give up, with your investors losing everything

This is obviously a martingale gambling strategy and the analogy is made extensively in the opinion but don't worry about that now. Worry about the purity of the ratings arb here. It is breathtaking. Here is the core trick of it: S&P rated structured credit products based solely on the probability that they would pay off less than 100% of their principal plus interest, and not at all based on the expected loss if that happened. A triple-A rating required a <0.728% probability of defaulting. What this means is that:

  • (1) A bond with a 99.3% chance of paying off 100 cents on the dollar, and a 0.7% chance of paying off 60 cents on the dollar, is AAA.
  • (2) A bond with a 99.2% chance of paying off 100 cents on the dollar, and a 0.8% chance of paying off 60 cents on the dollar, is AA+.
  • (3) A bond with a 99.3% chance of paying off 100 cents on the dollar, and a 0.7% chance of paying off zero, is AAA.

Which would you rather have, between (2) and (3)? The expected value of (3) is noticeably lower than the expected value of (2),1 but its rating is higher. This means that (3) is rated AAA but should have a higher spread than an AA+ instrument, through sheer arithmetic. This makes it a much better product to sell, though not necessarily a good product to buy. Item (3) is the CPDO.

It's easy to conflate these things with CDOs because, shared letters! Bad ratings! Shoddy derivatives! But CDOs are basically an intuitive idea with real economic reasoning behind them. You got some stuff. Some of it is good, some is bad, you can't tell which is which. You put the stuff in a box and you say to Guy A "you get the best stuff" and Guy B "you get the worst stuff." Guy B is actually taking more risk than if he just had the whole box; Guy A is actually taking less risk. CDOs got a boost because people and, especially, ratings agencies2 misjudged both the quantity of bad stuff and the shielding Guy A got from it, but the basic concept is fundamental to, like, everything; that is the difference between debt and equity.

But the CPDO marketing pitch was basically "we have constructed a magic trick, would you like some?" They exist as a pure creature of rating-agency gaming. Wonderfully one of the main structurers of this thing was named Paul Poet and paragraph 86 of the opinion contains an email that he wrote. He starts martingale:

The taking on of leverage when you underperform is similar to the following casino strategy:
1. Turn up with £128, you risk £1 on a 50:50 bet.
2. If you win you’re done.
3. If you lose, double your bet (you can try this 7 times).
4. Repeat.

But to me the telling sentence is: "As in the casino strategy the final IRRs are very stable, there are 2 outcomes: you either hit your target with a high probability or walk away with nothing with a low probability." Having only two outcomes - all or nothing - is not necessarily an ideal economic proposition, but it perfectly describes the perfect ratings arbitrage. Any chance of a 40% recovery, say, would be wasteful: it's economic value for investors that does not enhance the rating, and the rating is all you want. This thing was built with the single-minded purpose of gaming the agencies, creating literally the worst thing that could still be rated AAA.

You can't blame S&P for that: they are always transparent that their rating is not an investment recommendation but rather a statistical judgment about likelihood of default; the fact that they ignore loss-on-default is public. Stupid, but public. So the fact that ABN built this thing around tricking the S&P system is not really a reason for S&P to be liable to anyone who lost all their money on it: the core trick of the CPDO, although it tricked S&P in some real sense, should not really have tricked investors.

But that was just the core trick! The other tricks make up the bulk of the 8,000-page opinion and they are ... tricky. I do not understand all of them. In my defense it's not clear anyone does; here is Peter Tchir:

CPDO had a brief moment in the spotlight. It let you take the CDX indices, leverage them as much as you wanted, and get a much better rating than the index itself. It was a Frankenstein of a rating relying on “rolls and steep curves”, “mean reversion”, “short term default probabilities”, and “portfolio insurance” to justify the rating. It didn’t make sense to me from the first day ...

And here is Judge Jayne Jagot at the beginning of her 17,000-page opinion:

The CPDO was described in part of the evidence as a “grotesquely complicated” instrument. This is accurate.

But she ultimately concludes that S&P "did not have a reasonable basis" for its AAA rating. Among the tricks worth noting are that this thing was built on the bones of another thing that had a roughly opposite strategy. This thing, the CPDO, would sell more protection when protection was expensive, and buy it back when it was cheap - a buy-low-sell-high strategy that would pay off if credit bounced around normal levels, but that would be a problem if credit blew up or moved in one direction. The other thing did the opposite, basically buying high and selling low, so mean reversion hurt it. S&P more or less modeled the CPDO - "buy low sell high profit from mean reversion" - based on its models for the other thing. Here's an expert witness (paragraph 869):

By using the same model to price the inverse product, it was basically using a model that would obviously give too-high ratings... it's not clear to me exactly how they got their model, but I believe that it is derived, perhaps via ABN or not, from S&P's CPPI [the other thing] model. But, in any case, they used a similar process for how credit spreads might behave in the model, which is to assume mean reversion. And I'm saying that that, in my view, is inappropriate for a product that makes its money, at least claims to make its money, from mean reversion.

That's a good trick: build a thing, get S&P to model it based on conservative stressy assumptions, and then, when they give it a bad rating, point out "well the inverse thing must be really safe then, no?" No! Thing and Anti-Thing can both be risky. But if you model Anti-Thing based on assumptions that make Thing look bad, it will tend to look good.3

The rest of the story is about the marketing to the local councils and whose fault it is - between the councils, S&P, and various intermediaries - that the councils bought such a dumb thing. The judge is not so into the usual "this was all disclosed / they should have read the documents more carefully" line of reasoning; here is an amazing paragraph (2471):

The idea that the councils should have read and re-read the documents with which they were provided “until they could tell whether or not they understood the investment” is no answer at all. The council officers knew that they could not understand the investment. Reading and re-reading would have done nothing but confirm what they already knew.

I'm sure that's true. Nonetheless, for the benefit of Australian local councils of the future, let's have a quick lesson on reading marketing materials. Here is a bit of ABN Amro's marketing materials:

That's a lot of words but all they say really is "DON'T BUY THIS." How can you tell? A few quick guidelines:

  • "designed with a high likelihood of cashing-in to a risk-free investment" = should not be AAA; AAA should mean "99+% likelihood of being a risk-free investment."
  • "A highly rated bank will hold the note proceeds" = your concern should be getting your money back; if the best they can say is "your money will be burned in a secure and hygienic environment" then that's bad.
  • "high degree of transparency and liquidity in ... the index portfolio" = you want transparency in the thing you buy, not the underlying thing; if they're advertising transparency in the underlying thing it's because they've put nineteen layers of opacity between you and it.
  • "correlation is unnecessary for the calculation" = if a financial product is advertised to you as having the benefit of easy calculation, you are in terrible danger.
  • "The extent of the portfolio's unfunded leverage is a source of risk" = you think?

I dunno. Another thing quoted in the opinion (para. 53) is "the Grove Report," which some of the investors read and which concludes that CPDOs are for investors:

  • Looking for very long-dated and high-return structured credit product;
  • Needing a strong principal rating and capital guarantee, including rated income;
  • Able to invest for a 10-year period – while intended liquidity arrangements through making a market appear reasonable, there is no guarantee of liquidity in all market conditions.
  • Able to cope with a high level of potential volatility in NAV, especially as this will be a primary driver of secondary market pricing.

I say unto you that "potential volatility in NAV" means "you can lose a lot of money" (AS DOES "HIGH-RETURN"); the combination of those words with "strong principal rating" means "investors looking to game ratings agencies." That's what ABN Amro was looking to do, and there's little doubt that that's what a lot of its investors were looking for too: they wanted a AAA rating to go with much-riskier-than-AAA-type returns. Some surely knew that those two things were only achievable with, well, much-riskier-than-AAA-type risks: they knew that they, too, were gaming the agencies; they wanted a safe rating, not safety.

The Australian councils probably didn't know that: they thought they were getting not only a AAA rating but a AAA bond, except with a BBB yield. That's a strange thing to think. Yes, no amount of document-reading was likely to correct it, but surely common sense would tell you that those things don't go together?

S&P Misled Towns With AAA Rembrandt Rating, Court Rules [Bloomberg]
Bathurst Regional Council v Local Government Financial Services Pty Ltd (No 5) [2012] FCA 1200 (5 November 2012)
AAAs, alternative universes, and hindsight [FTAV]
The T Report: A Paucity of Change, Greece & CPDO [TFMA]

1.$99.68 versus $99.30 in the simple one-period case, but it's not one-period, S&P's assumptions were off, etc. etc., so the difference is bigger than that.

2.And capital regulators!

3.Another trick - and this isn't a trick so much as just a thing that baffles me - is that, while this is a product that takes highly levered long credit exposure, the model showed that it was safer at higher long-term credit spreads. This baffled even the ABN Amro structurers; the judge notes:

Again, it might be recalled that when dealing with a product where higher spreads acted as a stress and not a benefit Mr Drexler [of ABN Amro] had described to S&P the spread history of the actual indices referenced by the CPDO, the CDX and iTraxx, as “the iTraxx 5yr is currently at 30, averages 35, and ...never has exceeded 50 (the CDX hit 77 for one day in May 2005 – a week pre and post May 17, it was sub-60, and has averaged 52.5 over the past 2.5 years)”.

Anyway, S&P used a long-term value of 80 for the credit spread - which made the product look better than would a lower spread, despite the fact that ABN Amro was apparently arguing for a lower spread.

[Update: Also Felix Salmon has a good rundown of the rest of the tricks:

  • S&P used the wrong model input for starting spread.
  • S&P used the wrong model input for volatility.
  • S&P used the wrong model input for average spread.
  • S&P completely ignored ratings migration.

This is borne out by the opinion; they do not seem to have gotten a lot right.]