They rated every deal, including the ones structured by cows. But were they fraudulent? Until last Tuesday, when Standard & Poor’s parent company McGraw Hill paid $1.3 billion to settle a US Department of Justice lawsuit, there existed the tantalising possibility that this question about S&P’s structured finance ratings business might be answered in court.
In the absence of any admission of fraud by S&P (a key concession that the rating agency wrung from the DoJ) we can only look at the evidence and try to judge for ourselves. Fortunately, data visualisation technology now makes life easier for us. When I was first researching the Devil’s Derivatives in December 2008, S&P published a spreadsheet on its website (it was later removed). Entitled ‘Cash_Hybrid_CDO_Actions’, the spreadsheet listed 800 collateralised debt obligation vehicles, including the rating history of the 4,000 securities or tranches issued between them.
In total, this issuance amounted to about $460 billion, or nearly 40 per cent of the CDOs that S&P rated. By December 2008, the portfolio was already plunging in value, and subsequent releases by the rating agency would confirm that by 2011, most of it was worthless. What I have done is update the spreadsheet, and apply a simple rule that values each tranche according to its subsequent rating, where ‘D’ or default gives a value of zero.1)I mapped each rating to a number between zero and 20, and adjusted the value of each tranche when it was downgraded. Up to the end of 2008, S&P’s spreadsheet provides detailed information on the timing of downgrades. After that, full rating histories are unavailable for most deals and some interpolation was necessary, utilising publications by S&P listing which CDOs had been downgraded to ‘D’ by 2011. This interpolation explains the downwards step in the curve at the start of 2010.
In this visualisation, the purple line plots the cumulative issuance of AAA tranches over time, adjusted for subsequent rating. If you mouse over the curves, you will see the individual CDO events, from issuance to downgrade, to default. The curve is striking, and I wish I had included the chart in my book.
So where is the fraud? Yes, over a trillion dollars worth of CDOs rated AAA by S&P lost most of their value over a two year period. That caused great problems for financial institutions exposed to those CDOs and the governments that bailed them out, as well as other investors such as pension funds. But maybe S&P just made a mistake.
To go further, take a look at the second chart – in particular the green dashed line. In December 2005, S&P published a document describing something called CDO Evaluator 3.0 (CDOE 3). This was a model for rating CDOs, and it gave issuers a tool to help them meet S&P’s thresholds for a given rating. Although CDOE 3 was not the sole determinant of S&P’s ultimate rating for a CDO, it was the most important one, because it was the model that its clients used. If S&P didn’t deliver on what CDOE 3 promised, clients could simply shop around for a better rating from another agency.
In my visualisation, the green line uses CDOE 3’s default rate predictions for AAA tranches to calculate an expected value for the AAA deals in S&P’s 2008 spreadsheet 2)The appendix to S&P’s CDO Evaluator technical document contains credit curves, or the expected default rate for a given number of years after the date of issue, and I applied these default rates to the deals in S&P’s spreadsheet. To be more generous to S&P, one could include the impact of rating migrations as well as default rates, but CDOE 3 does not allow for CDO tranche migrations. See how after the end of 2007, this line stays horizontal with a barely detectable downward slope, contrasting with the actual value (measured in terms of S&P’s own reduced ratings for the CDOs). Now we can ask the question: what did S&P know about the problems in CDOE 3 and when did it know?
Unsurprisingly, this question was at the heart of the DoJ’s February 2013 complaint against S&P, along with similar questions about the rating agency’s treatment of residential mortgage-backed securities linked to subprime (which were key ingredients of the CDOs). The DoJ attempted to answer the question using a trove of S&P internal emails. It starts out listing examples where analysts fretted about changes to CDOE 3 to make it more ‘business friendly’.
The emails get especially interesting in early 2007 as S&P staff realised that the rating agency’s clients, the CDO issuers, had assembled vast warehouses of subprime loans and RMBS deals that were beginning to go sour. With indicators like the Markit ABX index plunging, the issuers would soon report mark-to-market losses on their warehouses unless they could shift them into CDOs where the losses could be hidden – behind S&P’s credit ratings. Increasingly, CDOs would have to invest in each other for this to work.
The DoJ quotes a March 2007 memo from senior S&P executive David Tesher warning his colleagues about this rush to close deals. Around that time his subordinates begin performing Talking Heads skits and talking about cows structuring deals. A Bethany McLean article in Fortune makes them squirm further. Looking at the chart once again, you can see what happened: S&P continued to rate CDOs, with about $140 billion being issued between March and December 2007 when the market slammed shut for good.
According to the DoJ, this is what fraud looks like: knowing you are doing something wrong but continuing to do it because of the money at stake. The complaint details what these incentives were: S&P charged its clients fees of $500,000-$750,000 for individual CDO ratings, resulting in global CDO revenues of $203 million in 2007 alone. Rather than stop the gravy train, S&P assigned its senior executives to deflect blame in the op-ed pages of the Wall Street Journal.
That isn’t to say that S&P was alone. Moody’s behaved similarly, but escaped prosecution, something which enabled S&P to claim that it was singled out because it dared to downgrade the United States (S&P agreed to retract the claim as part of its settlement). Of course, we’ll never know what would have happened in court, but the data tell an incredible story that deserves not to be forgotten.
References
1. | ↑ | I mapped each rating to a number between zero and 20, and adjusted the value of each tranche when it was downgraded. Up to the end of 2008, S&P’s spreadsheet provides detailed information on the timing of downgrades. After that, full rating histories are unavailable for most deals and some interpolation was necessary, utilising publications by S&P listing which CDOs had been downgraded to ‘D’ by 2011. This interpolation explains the downwards step in the curve at the start of 2010. |
2. | ↑ | The appendix to S&P’s CDO Evaluator technical document contains credit curves, or the expected default rate for a given number of years after the date of issue, and I applied these default rates to the deals in S&P’s spreadsheet. To be more generous to S&P, one could include the impact of rating migrations as well as default rates, but CDOE 3 does not allow for CDO tranche migrations |
Comments are closed.
A $1.375 billion fine seems trivial in view of the direct loss of $460 billion. Did all of the rest of losses just disappear? Somehow, I’ve got to believe that there were some winners in these deals that have the culpability of knowing that there were serious flaws in these CDO’s well before others discovered them and they took advantage of them. Taking your example of the “money machine” from Devil’s Derivatives, it seems to me that no matter how much diversity that you build into a transaction, you cannot eliminate the effects of putting in sub-prime mortgages. Further, the pricing models that are used for such things are fundamentally flawed at least in the respect that no amount of explanation or “proving” of a model will eliminate the mistrust of that model. That’s the subjective part of the model that is really not quantifiable.
It seems to me that this subject has been addressed rather directly by Emanual Derman in Models.Behaving.Badly. (I got that book on your recommendation. Thank you.) He said on p.109 that while the numbers in QED are accurate to 10 significant digits, financial models are not accurate to even one. And he said that more to the point, who can define accuracy in a financial model since so many of the model inputs are subjective in nature. I might question such a brash statement in the face of the continuing widespread use of financial models to price securities, except for the fact that Mr. Derman was chief quant at Goldman for a number of years, not to mention his previous background in quantum physics.
That said, it seems to me that former Chief Justice Louis Brandeis had it nailed with how to deal with all of this. He said: “Sunlight is said to be the best of disinfectants; electric light the most efficient policeman.”
I worked for S&P for 7 years in the 2000’s Their business model allows them to be paid for producing a rating and then they sell the rating. Yes they make money on both ends of a deal. I warked on the retail side in plan participant products and not in ratings and I was by no means an exeeutive or senior manager however even to my low status eyes this is this inherent conflict of interst has never been addressed to my knowledge. by any regulator.