By Tom Adams, an attorney and former monoline executive
Felix Salmon at Reuters and Steve Gandel at Curious Capitalist used some of the analysis in Michael Lewis’s The Big Short as an opportunity to attempt to exonerate Goldman Sachs for the charge of deliberately constructing CDOs to go bad for their own profit. In particular, Mr. Gandel notes that the data of AK Barnett-Hart, the now famous Harvard student who wrote a paper on CDOs, shows that the CDOs on which Goldman Sachs was the banker performed better than most of the CDOs offered by other bankers. As a result, he concludes that that Goldman might not be bad as some people have claimed when it comes to their CDOs. He goes so far as to suggest that this data may even exonerate Goldman of some of the vampire squid mythology.
The problem is Ms. Barnett-Hart’s data offers no conclusive proofs of CDO performace.
First, her data only went through the end of 2008, as Mr. Gandel acknowledges. Very few CDOs defaulted by this point (though some did hit events of termination) even though the collateral underlying the deals has already plummeted in value.
A shortcoming here is that her thesis, which criticizes rating agency ratings, nevertheless also uses them to assess how well or badly a particular CDO fared. For a variety of reasons, including the complexity of the deals and perhaps pressures from outside forces, the rating agencies took a remarkably long time to downgrade many of the CDOs backed by mortgage collateral and this process continued throughout 2009. As a stand-in for defaults, Ms. Barnett-Hart relied on significant downgrades to the CDO bonds as a measure of implied default. This, too, has some problems as a measure, since the rating agencies were, in many cases, wildly divergent. I performed extensive analysis on the CDOs in AIG Financial Product’s portfolio and was surprised to find numerous instances where one rating agency would have the ratings of AIG’s CDO at a A or AA level while the other had it at CCC levels, even at the beginning of this year, when I was performing my analysis.
Another issue I found when examining AIG’s portfolio is that very little data was available for Goldman Sachs Abacus program, the synthetic CDOs that were the subject of the December 2009 New York Times article Mr. Gandel cites. I have also been unable to find transaction-level detail about these deals from other public sources.
If the Abacus transactions (25 in total) are missing from Ms. Barnett-Hart’s data set for these deals, as it was for my own analysis, then it is certainly possible that her analysis of Goldman was incomplete. If they are not present in her data set, that means the impact of Goldman creating bad CDOs to bet against them was not addressed in any way by her analysis.
This discussion prompted me to go back and look at the CDOs in the Maiden Lane III portfolio for AIG, as we previously discussed here at Naked Capitalism. I was able to track down the CUSIPS, transaction parties and ratings, as of January 2010, for all of the ABS CDOs in AIG’s portfolio and the Maiden Lane III transaction. Goldman Sachs was the banker for the largest percentage of CDOs in this portfolio, contributing just over $18 billion worth of bonds to the AIG cause. Merrill Lynch was the banker for the second largest amount, with $12.4 billion. I went back and looked at the current Moody’s ratings (Moody’s was generally more conservative than S&P), as of January, and learned that of the top four bankers Goldman, Merrill, Deutsche Bank and UBS), who were responsible for over $44 billion of the $62 billion portfolio, Merrill’s deals had the worst weighted average rating. Their deals come out at Caa2. The CDOs of Goldman Sachs were the second worse, with a weighted average rating of B1 and ranging between C and A1 (for a 2002 transaction).
In addition, for the deals on which Goldman Sachs was a counterparty, which represented approximately $14 billion, the weighted average rating was quite a bit worse than the deals for which they were just the banker (meaning they had created the CDO but a different bank owned it and insured it with AIG). On these deals, the weighted average rating was Caa2. This means that the deals which Goldman chose to retain, either on their own behalf or for the benefit of clients, performed much worse than the deals they sold to others. Of course, it was for these deals which Goldman received insurance from AIG. And when AIG was bailed out, Goldman kept the collateral which AIG had already posted (approximately $8 billion) and got paid out on the remaining amount from the Federal Reserve via the Maiden Lane III transaction.
Since Goldman was the most aggressive bank when it came to seeking collateral from AIG, it stands to reason that the deals on which they could require such posting would be particularly bad (and of course, Goldman may have also been more realistic about the likely value of the AIG insured CDOs, compared to the other banks).
This data analysis does not prove that Goldman actively created bad CDOs so that they could profit when the deals failed. However, since it is based on more current ratings and relates to a specific strategy that Goldman apparently had with AIG, it provides a more detailed viewed into how Goldman managed their CDOs. Because the CDOs underwritten by Goldman in AIG’s portfolio were among the worst in this portfolio, and because the CDOs on which Goldman was the counterparty performed even worse than the banked deals, I feel confident in saying that Goldman has not been exonerated.
One way to determine whether Goldman is innocent would be to examine the performance of the Abacus synthetic CDOs, on which Goldman presumably would have held a short position from the outset. One would assume that since Goldman devised these deals with the intention of profiting from its short position, those CDOs would have performed even worse than the other Goldman deals. Unfortunately, Goldman has declined to make this data available.