Insider's Game

Hopefully, the report released tonight by the Senate Permanent Subcommittee on Investigations, entitled “Wall Street and the Financial Crisis: Anatomy of a Financial Collapse,” will address the biggest scandal overlooked by the FCIC, how Goldman used the subprime market index, known as the ABX, as a pump and dump scheme.

A review of Goldman’s mortgage trading activity–about 14,000 trades executed between May 2007 and November 2008–offers additional evidence of Goldman’s market manipulations. It also refutes dissembling offered up by Goldman executives in their testimony last year.

As you may remember, Goldman CFO David Viniar claimed that his firm didn’t really break out its trading activity between cash and derivative instruments. There was just one mortgage trading desk that bought and sold cash and synthetic instruments, which were presumably interchangeable. Except they weren’t. There was virtually no overlap between Goldman’s cash trades and its synthetic trades.

During the five-month stretch between May and September 2007, Goldman’s trading volume for residential mortgage related instruments exceeded $77 billion. But only $105 million, about 1/10 of one percent of the total, represented trades of the same instruments, identified by CUSIP number, in both the cash and CDS market. In the cash market and the synthetic market were entirely different with regard to the risk levels involved. The lion’s share of Goldman’s cash trades were among triple-A and double-A tranches of subprime and Alt-A bonds. In the synthetic market, the trading volume was overwhelmingly in triple-B tranches, with very limited trading in single-A tranches.

This makes perfect sense, for several reasons. First, there were a lot more triple-A tranches available to be sold. They represented about 80% of a subprime deal’s capitalization, and about 94% of an Alt-A deal’s capitalization. By contrast, triple-B tranches (BBB+, BBB and BBB) constituted a tiny sliver, about 3% of any subprime deal. More importantly, very few of the triple-B tranches were available for sale. As the FCIC pointed out, about 90% of those subordinate tranches were stuffed into mezzanine CDOs. CDS trading of residential mortgage bonds had no cash basis, or any real market with price discovery, to be used as a reality check.

These credit default swaps were never intended to be interchangeable with cash investments. They were designed by a handful of bankers, including Rajiv Kamilla of Goldman and Greg Lippmann of Deutsche Bank, to be profoundly different from the mortgage bonds that they referenced. The bankers wanted a standardized contract to provide for a payout long before the underlying bond declared a default.

Residential mortgage bonds have no amortization schedule; they prove no fixed date for repayment of principal prior to the final maturity date, 30 or 40 years after closing. Monthly cashflows are first applied for payment of interest to all tranches, and any remaining cash is used to pay down principal of the most senior tranche. Each deal has its idiosyncrasies with regard to the mechanics of cash distributions, but the general rule is that a subordinated tranche can be “current” on its debt obligations, i.e. interest payments, for years and years after it became obvious to everyone that the subordinated tranches would never recover a nickel of principal.

Kamilla, Lippman et al. created a template for buyers of credit protection who did not want to wait for a formal default. These bankers developed the pay-as-you-go ISDA contract, which facilitated a payout to credit protection buyers based on various credit events such as an implied writedown. Without the pay-as-you-go format, released by ISDA in June 2005, almost none of Goldman’s synthetic trading, which included synthetic CDOs and ABX trades, would have been possible.

Why was so much of Goldman’s trading activity concentrated on such a small triple-B segment, which represented 3% of a deal’s capitalization? Because by May 2007, an awful lot of people had seen and passed around Greg Lippmann’s flipbook presentation, titled, “Shorting Home Equity Mezzanine Tranches.”

Lippmann’s thesis was that the credit ratings on subprime mortgage bonds were fatally flawed, and the odds strongly favored a big windfall if you shorted the most subordinate tranches, i.e. the triple-B tranches, of a subprime securitization. He made his pitch to hedge funds and handing out T-shirts with the phrase, “Short my house,” in January 2006, around the same time that he was promoting the launch of the ABX.

Moody’s, S&P and Fitch purported to rate deals, “through the cycle.” That is, the ratings were supposed to measure the likelihood that an entity would survive a cyclical downturn and still meet all its debt obligations. But by the early 2000s, the rating agencies’ approach toward residential mortgage bonds was to do the opposite; they rated deals through the bubble. More specifically, they assumed that credit losses would be no higher than those experienced during the real estate bubble. For subprime deals issued during 1998, 1999 and 2000, vintages that were followed by unprecedented growth rates in virtually all housing markets, the cumulative loss rate in the first five years was about 4.7%, and losses were still climbing. (Remember, these were average losses; a lot of deals incurred loss rates that were considerably higher.) For most of the deals issued during 2005 and 2006, Moody’s calculated estimated losses below 5.5%.

One of the reason why the ABX was so popular was that it was so easy to figure out. Each vintage composite was comprised of 20 large subprime mortgage deals equally weighted at 5%. So if four deals tanked, the recovery of principal on the composite would never exceed 80%. By October 2006, it was obvious that six of the 20 constituents of the ABX 2006-1 had effectively tanked.

By October 2006, when the triple-B tranches of the ABX continued to trade at close to par, it was it was obvious that many of the constituents of the indices would be wiped out. There are a lot of moving parts in a mortgage securitization, but one yardstick that tells you that the triple-B tranche will get wiped out is if, within the first 15 months, 5% of the loan balance is in foreclosure. That doesn’t automatically translate into a 5% loss on the original loan balance, but it does mean that loan performance has deteriorated so quickly that there was no way, under the rating agencies’ original assumptions, that the triple-B tranches could recover their principal. In October 2006, Goldman began marketing the now-notorious CDO, Hudson Mezzanine I, which was primarily comprised of the triple-B and triple-B-minus constituents of the ABX.

With the rating agencies’ blessing, these deals were structured in a way that allowed almost no margin for error. In almost every case, the level of overcollateralization used to protect the triple-B tranches was far less than Moody’s estimated loss on the mortgage pool. A deal with a 5.5% estimated loss would have maybe 3% overcollateralization, otherwise known as the unrated equity tranche. The difference was supposed to be covered by Moody’s guesstimate of total “excess spread.” It’s axiomatic that interest income on the good loans is the only thing that offset the credit losses on the bad loans. For the triple-B tranches, the excess spread to cover credit losses was only made available after: (a) current interest payments were made on all tranches in the deal, and (b) all the principal had been repaid on the 94% of all tranche debt that ranked senior to the triple-B tranches. If the good loans prepay faster than expected, or if the bad loans become delinquent sooner than expected, the value of the excess spread can quickly shrink to the point where, if you update the cashflow projections using Moody’s original assumptions, it becomes obvious that the triple-B tranches will never recover their principal.