Goldman’s controversial “ABACUS 2007-AC1″ synthetic CDO turns out to be a very complicated deal. This is not your grandfather’s vanilla mezzanine RMBS synthetic CDO. It is, in some sense, a supersynthetic CDO.
There’ve been some excellent posts dissecting the deal, including…
- Alea: Abacus for Dummies
- Bionic Turtle: Goldman’s Abacus illustrated
- Sandrew on Finance: Reverse Engineering the ABACUS Timeline
In what way was ABACUS a “supersynthetic CDO”? Despite notionally having seven classes of investors, just two classes of notes were actually sold. When I read this at Alea, it blew my mind. The only notes that were sold were AAA debt, from senior (but not “super senior”) tranches. I didn’t understand how this could work. CDOs turn low-class debt into AAA gold by segregating losses. Senior notes are made safe by shifting losses to junior tranches, and remain safe until the junior tranches are wiped out. I had seen synthetic CDOs with unfunded senior classes, in which case the issuer retains some risk if the CDO fails catastrophically. But if there are no junior tranches, who takes the first loss? Who stands in the line of fire to protect AAA noteholders?
I spent some time squinting over the prospectus to understand. But there is no clearly stated explanation. On the contrary, there is a lot of language like this:
On (i) each Payment Date and (ii) any other Business Day on which Currency Adjusted Notional Principal Adjustment Amounts are paid by the Issuer to the Noteholders, the Class SS Notes will be senior in right of payment to the Class A-1 Notes, the Class A-1 Notes will be senior in right of payment to the Class A-2 Notes, the Class A-2 Notes will be senior in right of payment to the Class B Notes, the Class B Notes will be senior in right of payment to the Class C Notes, the Class C Notes will be senior in right of payment to the Class D Notes and the Class D Notes will be senior in right of payment to the Class FL Notes.
That sounds like the standard CDO waterfall. But in reality there was nowhere for the water to fall, because no B, C, D, or FL notes were sold. If losses were allocated to any investor, they would be allocated to AAA tranches. So what was going on?
The ABACUS prospectus doesn’t say. But there is a hint. Rather than buying credit default swaps on the aggregate reference portfolio, then dividing the cash flows among the tranches based on seniority, the CDS payments are calculated separately for each “series” of notes (where the series are subdivided by class). In other words, each class of notes writes its own distinct insurance policy.
As best as I can tell, there are two distinct levels of abstraction in the ABACUS deal. First there is the reference portfolio, a hypothetical portfolio of debt. Then there is a notional CDO, a hypothetical entity that we imagine to have purchased (or synthesized) the reference portfolio. We pretend that this notional CDO is “fully funded”, with a $1100M SS tranche (“super senior”), a $200M Class A-1 tranche, a $280M Class A-2 tranche, a $60M Class B tranche, a $100M Class C tranche, a $60M Class D tranche, and a $200M FL tranche (“first loss”). In reality, no one has purchased any of the reference portfolio, and the notional CDO, which would have required
$1.8B $2B of investor interest to build, was never constructed. Instead, the notional CDO forms the basis for a thought experiment: Given any performance scenario for debt in the reference portfolio, we can compute the loss that would have been experienced by holders of the various tranches. So, we could write a kind of swap (somewhat different from an ordinary credit default swap), whereunder a “protection buyer” pays a predetermined, fixed spread and a protection seller pays the losses that a hypothetical holder of a tranche in the notional CDO would have experienced.
Effectively, the seven tranches of the notional CDO serve to define seven new kinds of bets that one could take on the reference portfolio. Since these bets are designed to mimic the experience of investors in a real CDO, S&P and Moody’s were able to associate ratings with these bets. However the bets themselves — highly customized variants credit default swaps — are not securities.
For a regulated entity that wished to hold AAA debt, securities had to be constructed based on these bets. The actual “ABACUS 2007-AC1″ legal entity offered synthetic securities designed to mimic the experience of tranches in the notional CDO. It did so in the usual way, just as a commodity ETF or “vanilla” synthetic CDO would: The entity accepts money from investors, and uses those funds to purchase ordinary, low-risk debt. (In this case, the low risk debt was not so ordinary; it was itself a synthetic security. But we’ll set that aside.) The entity then takes a side bet. Investors’ earnings are interest on the low risk debt adjusted by the gains or losses they experience on the side bet. The net effect of all this is that buyers of “notes” from the entity experience outcomes that are almost exactly as if they had invested in a tranche of real CDO. However, notes synthesized this way do not need to be backed by a funded CDO structure (cash or synthetic). In fact, the scheme completely eliminates all constraint on the quantity of funding invested in a given tranche, and severs any relationship between the quantity of funding and the characteristics of the securities. Goldman could have sold $1 worth of Class A-1 notes or a $1T dollars of Class A-1 notes, as long as it was able to make itself comfortable with taking the other side of the Class A-1 side bet. It happened to sell $0 worth of Class C notes, but it could have sold any quantity, without altering the characteristics of the notes or the structure of the notional CDO.
Once you “get it”, the scheme is not very difficult to understand, and it is clever. But it is not clearly described in either the ABACUS pitchbook or prospectus. I don’t know why the three-level structure is not clearly diagrammed (reference portfolio -> notional CDO -> funded entity that replicates the experience of arbitrary tranches in the notional CDO). “Notional CDO”, by the way, is my term. It is nowhere in the prospectus or pitchbook. The distinction between the notional CDO and the actual funded entity are blurred in the documentation. Perhaps the structure of this sort of deal would be obvious to insiders, or perhaps there are clearer descriptions elsewhere, in documents that have yet to be made public. Both Alea and David Harper have pointed out that this structure is similar to a “bespoke” or “single-tranche” CDO. Effectively ABACUS describes a hypothetical cash CDO with seven tranches, then chops it up into seven “single-tranche” CDOs, only three of which were ever actually invested. But in none of the documents is it represented as a bespoke CDO.
A remaining issue that has not received much scrutiny is how the deal was priced. IKB earned LIBOR + 0.85% on its Class A-1 tranche, prior to any credit events. Both ACA and IKB earned LIBOR + 1.10% on Class A-2 notes. In a cash or more vanilla synthetic CDO, the above-LIBOR cash flow to CDO investors is determined by the credit spread on the underlying debt, potentially plus a “basis” if demand for insurance has pushed the market price of protection above the underlying’s credit spread. Effectively, cash flows into the structure are market determined. (The allocation of spread between the tranches is an internal matter among the CDO’s investors.) With ABACUS or a bespoke CDO, there is no market in the tranche-specific credit default swaps and no security with an observable credit spread that can serve as a basis for pricing. So the price of protection must be negotiated between the protection seller (the ABACUS SPV and its investors in this case) and the protection buyer (usually the deal’s sponsor) without a very clear benchmark. Disclosure of the fact that there was an adversarial counterparty on the other side of the deal would likely have affected the character and perhaps the outcome of those negotiations. Since investors may have believed the ABACUS deal was offered and underwritten at Goldman’s initiative, it’s unclear whether there were active negotiations at all, or whether ABACUS investors simply accepted spreads computed by Goldman on the theory that as customers of a reputable bank they would be given reasonable prices. (“Fair” prices would have to be modeled, and modeling a fair price of a bespoke CDO tranche might be within the competence of an investment bank but beyond the competence of even “sophisticated” institutional investors.) Sponsors of bespoke CDOs often hedge their exposure in public markets, so ABACUS investors need not have suspected that there would be an identifiable counterparty, who was also a customer of Goldman’s, negotiating against them on price. Alternatively, Goldman undoubtedly had more efficient means of hedging its exposure that it otherwise would have, since it could just lay off the risk on Paulson. So Goldman might have been able to offer unusually good pricing to ABACUS investors. We cannot say a priori whether ABACUS investors ended up receiving better or worse pricing than they would have had Goldman underwritten this deal on its own initiative and hedged its exposure. But investors did not have the opportunity to negotiate price in full awareness of an adversarial counterparty, so the fairness of the spreads investors received merits further examination.
To summarize, ABACUS defined seven “side bets” based on the performance of the reference portfolio. Under each bet, one party would insure the losses of a hypothetical tranche of a notional CDO in exchange for fixed payments from the other party. The ABACUS legal entity synthesized securities based on two of those side bets, and sold those synthetic securities to IKB ($150M) and ACA ($42M). But as Alea points out, the largest “investment” — by ACA via ABN-AMRO — was not actually a purchase of notes from the ABACUS SPV, but an unfunded side bet. ACA/ABN took a $909M “long” positions in one of the seven side bets, with Paulson (via Goldman Sachs) on the other side. This was an unfunded CDS-like arrangement that occurred some time after the ABACUS legal entity was formed and funded.
I think in judging Goldman Sachs’ behavior, the fact that the ACA/ABN “investment” was a side bet arranged after the deal closed is important. The SEC’s main allegation, that Goldman was less than candid about Paulson’s role during the selection of the reference portfolio, would have affected all parties, IKB, ABN-AMRO, and ACA, both as noteholders and bond insurers (side bettors). But the question that I find most interesting is whether or not Goldman mistreated investors by virtue of a conflict between its roles as market maker and underwriter. That conflict directly affected only IKB and ACA as purchasers of newly underwritten notes. The ACA/ABN “wrap” of the super senior tranche occureed after the ABACUS LLC had been underwritten, so Goldman was only a counterparty to ABN/ACA at that point in time.
Update: Correction: IKB invested the A-1 tranche, not ACA as originally stated. Many thanks to commenter gennitydo for pointing out the error.
Update 3-May-2010: Yves Smith publishes a note from an anonymous correspondent claiming that ABACUS was just a failed underwriting of a vanilla CDO, not several “singe tranche CDOs” as described above. I think her correspondent is mistaken, and stand by the post as written.
If ABACUS had been constructed as a vanilla synthetic CDO, but the junior tranches had been left unfunded, Goldman would have been on the hook for that risk (as well as the risk of the super senior tranche and the unfunded portion of the Class A-1 and A-2 tranches). Goldman would have lost at least $708B on the deal if that had been the case, probably much more, depending on how worthless the super senior tranche turned out to be. Goldman could have synthesized the full reference portfolio and then dynamically hedged its exposure to the whole unfunded portion of the structure, but that would have been an elaborate and inefficient means of reaching an economically identical result. The prospectus notes that the structure would sell CDS by series of note, where series are within-tranche groupings, which it would not have done if it were synthesizing the full reference portfolio. ABACUS was built from single-tranche CDO’s, with Class A-2 notes covering a 21% – 35% slice of a notional CDO built from the reference portfolio and Class A-1 notes covering a 35% – 45% slice, while the unfunded but eventually insured super senior tranche was 45% – 100%. No one funded or ever bore the risk of the 0% – 21% bit.
- 26-April-2010, 6:45 a.m. EDT: Corrected misstatement of which parties invested which tranches, with thanks to commenter gennitydo.
- 3-May-2010, 3:00 a.m. EDT: Added update re a purported debunking of this description published at Naked Capitalism.
- 3-May-2010, 3:45 a.m. EDT: While reviewing the piece after its alleged debunking, I notice that I am arithmetically inept. It would have taken $2B, not $1.8B to fully fund the structure. Corrected in the text with the old value scratched.