The recent news about Goldman, especially the deep dive by the NYT, has focused on a transaction that, until now, not many people knew about and even fewer people understood. These deals, collectively known as ABACUS, were based on an extremely esoteric type of transaction known as synthetic CDO's. Attached to this article is a termsheet that shows the important terms for one of the earlier iterations of the ABACUS trades.
Unlike the transaction that is the subject of the SEC investigation, this transaction wasn't marketed as a CDO with a third party actively managing the transaction. Instead, as one can see when comparing the documents, this transaction is stripped down and much simpler. This is an important detail because, in effect, this transaction was a way for Goldman to go very, very short on residential or commercial mortgage bonds and make no representation of being a customer-centric transaction.
Further, as I'll show, there are a lot of aspects of the transaction that allow Goldman to take significantly less risk and structure a "heads I win, tails you lose" transaction. The whole point of these synthetic trades was to short the real-estate markets as cheaply as possible. However, because these are huge trades, they are costly to unwind and costly to hedge, so Goldman employed other features to minimize the risk they would have a losing transaction on their books for a significant period of time.
There are some key terms to understanding these trades that I'll go over here. They don't all impact the trade equally. But first we'll start with an overarching description of the trade.
Overview: Here is how the ABACUS trade works. Goldman takes a reference portfolio (an example portfolio of bonds, none of which Goldman actually has to own) and uses the performance of that portfolio as the basis for performance of the various securities being offered. The structure of this transaction is just like other securitizations: each set of bonds is senior to all the bonds below it and they get paid off in order of their seniority. For example, the First Loss will take losses before Class L, and Class L will take losses before Class K. The selection of the reference portfolio, and improper disclosures surrounding how it was picked, are the main topic of the SEC complaint against Goldman in the ABACUS 2007-AC1 transaction. Unlike that transaction, this one doesn't have a third-party collateral manager -- all roles here are assumed to be Goldman.
It's important to note that 92% of this transaction is expected to be rated AAA. That's right, 92%! Why so much? Well, the entire reference portfolio is rated single A, which is obviously considered safer than some of the collateral backing other transactions. The ABACUS 2007-AC1 transaction, which the SEC has targeted, for example, was backed by lower-rated BBB bonds. Only 79% of that transaction was rated AAA.
I'm not going to go into much detail about the next aspect of the trade, but will give a brief description. The difference between bonds and CDS contracts, mainly, is the principal payments.
When one buys or sells credit protection -- as in a credit default swap transaction -- the parties exchange a monthly premium and any mark-to-market gains or losses from the underlying bonds (remember, people always compare these contracts to insurance). But, if you owned the actual bond, as the bond pays down you would get principal payments. Since many investors can't invest in derivatives, only bonds, Goldman structures a complex transaction that re-creates the principal payments. Goldman then structures the ABACUS tranches as bonds and uses this agreement to be able to pay those tranches like bonds.
Also, keep in mind that the securities are long CDS contracts on the reference portfolio. A CDS contracts is between two parties -- someone is long and someone is short. It should be unsurprising that Goldman is the party going short here (just see the term "Protection Buyer" as "Goldman Sachs Capital Markets, L.P."). Also, Goldman is able to mark its short position to market -- actual losses do not need to occur for Goldman to make money. Put another way, if Goldman is short these bonds (synthetically, using CDS) and their price drops by 20%, but no losses occur, Goldman makes 20% on its short when it revalues its position using current market levels (which it is required to do). Of course, when losses do occur, Goldman profits then too.
An important point to note is that Goldman, at any time, doesn't have to create classes A through L, the Super Senior, and the First Loss; they can create whatever they want to create and however much of it they like. Remember, and this is important, this whole transaction is synthetic. Goldman never has to sell a penny of this transaction to anyone, or it can sell just the pieces it wants -- as long as they keep the economics of the transaction the same for the buyers of the securities, and as long as its documented, they have complete flexibility. Put another way, the portfolio is merely a reference to track performance, and the rest of the bonds and structure are synthetic. If Goldman decided it only wanted to sell Class H, for example, it could sell billions of just that class. The $1.25 billion is only a reference used so that people could see the capital structure and so that people could run the portfolio for analytical purposes. I've beat this over the head because it's a very, very important point. It's not as if Goldman has to keep the Super Senior or First Loss pieces (listed as "Not Offered"). Those could easily be fictitious, or Goldman could just buy these classes themselves. This allows Goldman to sell only the class of the transaction that they view as the most profitable for them to sell -- the part that is the best tranche for them to be short.
For example, if Goldman thought the underling portfolio would take losses of 6%, they could sell only Class C and Class D, the most senior tranches that would take losses. Why would they only sell these? They are the most senior, and thus cheapest for Goldman to short (the safer a security is the cheaper it is to short). Some language around this can be found in the section entitled, "Additional Issuance of Notes." (Tracing this through the document is difficult, as it makes reference to authorized amounts and Maximum Issuable Amounts, but there is nothing stopping Goldman from creating classes on their own, using this ABACUS trade only for a reference.)
I think it's important to note that, while a lot of these results and terms are counter-intuitive to someone not well-versed in these complex derivative markets, this might not be the case for seasoned market participants.
Important Terms
1. Call Option: This is probably the most underreported feature of the transactions. In the section entitled, "Optional Redemption" on page 6 it describes how Goldman can "call" or cancel the entire transaction and settle with buyers at that time (or any time afterwards), at its sole discretion. Further, earlier, the termsheet defines the first date on which Goldman can exercise this option as three years after the transaction is completed. Its also important to note that the call occurs at the discretion of the "Protection Buyer"--no effort is made, even in the terminology, to separate the short position with the ability to cancel the trade.
In English, this means that in three years (keep in mind these are 30-year bonds and they are expected to be outstanding between 8 and 13 years, according to the last few pages) Goldman can look at how the transaction is performing and just stop if they aren't happy. This is essentially a "heads I win, tails you lose" feature. If the market has rallied hard, and the expected drop in real estate hasn't happened, Goldman can cancel the transaction. If the market has begun to fall, however, Goldman can just keep the transaction outstanding and reap the profits. This is a hugely valuable option. To put this in perspective, most deals have call features, but they are generally much, much longer. Usually, they depend either on the underlying loans or securities being mostly paid off (90%+ having been paid off) or some very, very long time period having passed.
2. Credit Events: This is all on page 3. It describes exactly what events will cause payments to be made. Note that restructuring the debt also counts as a credit event that has to be settled up. Everything here is legal-ese, but follows a fairly intuitive prescription. If the underlying bonds take a loss, then the securities take a loss (Goldman's short position, though, earns money). Also, keep in mind that Goldman's short is being marked to market, so Goldman doesn't need actual losses to see a profit from this transaction.
3. Terms Related to the Principal Payments: As I said, I won't go into this in great detail, but the terms related to this section are found on page 5. Once again, Goldman is the counterparty to all of these other agreements.
Summary: Basically, it should be clear by now that the ABACUS trades were designed to be efficient shorts of various real estate markets. However, beyond that basic function, they were structured in a manner that allowed Goldman to extract much more value than they could in the open markets by just using CDS contracts. As I've stated, the very short call feature, the ability to synthetically create the most efficient short possible, and the structure of the transaction all are much better for Goldman than normal, open-market transactions. This allowed Goldman to maximize profits by reducing their costs and as much as possible and still short the real estate markets.