Fool's Gold: How the Bold Dream of a Small Tribe at J.P. Morgan Was Corrupted by Wall Street Greed and Unleashed a Catastrophe (13 page)

Precisely because subprime loans were risky, the home owners who took out such debt typically paid a higher rate of interest than prime borrowers did, and that meant that the “raw material” of subprime loans produced higher-returning CDOs than those built out of “prime” mortgages. For returns-hungry investors, subprime-mortgage-based CDOs were gold dust.

The only real constraint on the business was the need for brokers to find the cash to extend loans. In reality, though, that had become hardly a constraint at all. Brokers and banks alike no longer kept most of the mortgage loans they extended on their books any longer than a few days or even hours. Mortgage lending had become an assembly-line affair in which loans were made and then quickly reassembled into bonds immediately sold to investors. A bank or brokerage’s ability to extend a loan no longer depended on how much capital that institution held; the deciding factor was whether the loans could be sold on as bonds, and the demand
for those was rapacious. In this way, the lending of the mortgages began to be driven by the demand of end investors, in what would prove to be a vicious cycle.

A fundamental danger of the mortgage CDO business was that there was no good information on what might happen to subprime mortgage defaults in a severe house price slump. As Terri Duhon and Krishna Varikooty had discovered when they tested the idea of bundling up mortgage loans to put into a BISTRO product, because the United States hadn’t experienced a nationwide housing crash for seventy years, data on default patterns was extremely scarce. Worse, it was virtually nonexistent for the riskier subprime sector, because subprime mortgages had represented such a small portion of the mortgage market before the turn of the century. While that information gap had worried the J.P. Morgan team enough back in 1999 to lead them to forgo BISTRO deals with mortgage debt, by the middle of the new decade most bankers were willing to ignore the risks and sell these investments on a massive scale.

Banks repackaged mortgage-based bonds in ever-more-creative ways. The best known product was a CDO of asset-backed securities, or CDO of ABS. This was usually (but not always) filled with mortgage-linked bonds. In a sense, then, CDOs of ABS were like CDOs of CDOs. They had an added layer of complexity to add more leverage. Within that field, another popular product was known as a “mezzanine CDO of ABS,” which took pools of subprime mortgage loans and used them as the basis for issuing bonds carrying different degrees of risk. The bankers would then take just the risky bonds, say those rated BBB,
not
A or AAA, and create a new CDO composed entirely from those BBB bonds. That CDO would then issue more notes that were also ranked according to different levels of risk. The scheme looked fiendishly complex on paper, but it essentially involved bankers repeatedly skimming off the riskiest portions of bundles, mixing them with yet more risk, and then skimming them yet again—all in the hope of high returns.

The schemes became more creative still when banks started creating these products not out of actual bundles of mortgage loans but out of derivatives made of mortgage loans. The idea was borrowed from the world of credit default swaps, and as with corporate-loan-based CDOs,
these derivatives versions of CDOs enabled investors to place bets on whether mortgage bonds would default or not. These clever products were referred to as “synthetic CDO of ABS.” They would lead to a frenzy of speculation, all based upon the fundamental premise that the default risk of bundles of mortgages had been virtually erased by the process of bundling and then slicing them into tranches. If banks chose to hold more and more of the risk in these tranches on their own books, selling only the more popular tranches of notes, such as mezzanine, that was no worry because the risk had been so effectively dispersed that the chance the banks would ever take a hit from it seemed so remote as to be unfathomable.

The banks also began to turn to inventive devices for moving those deals off of their books. One of these was a type of quasi–shell company known as a structured investment vehicle (SIV) for purchasing the loans and selling the bonds sliced and diced from them. This species had actually first emerged two decades earlier, when two bankers at Citibank, Stephen Partridge-Hicks and Nicholas Sossidis, hit on the idea as a way of getting around the Basel rules for capital requirements. SIVs were tied to banks, not completely separate as with the shell companies known as SPVs. The banks provided some of their funding, as opposed to all of that being raised by the shell company itself, through selling notes. But SIVs did partially fund themselves independently, and they sat off the balance sheets of banks. They were thus a bit like the garage of a house: a useful place for banks to park assets they did not want inside their home banks. Another structure that fulfilled a similar function was a so-called bank conduit. This was similar to a SIV, but more closely linked to a bank.

The hybrid status allowed the banks to evade the Basel rules that limited the amount of assets they could hold on their balance sheets, thereby freeing them to leverage their capital a good deal more. The key reason the banks were allowed to do so concerned the way in which SIVs raised the portion of their funding that didn’t come from the banks, and this particular exploitation of Basel loopholes would lead to terrible consequences once the credit bubble began to burst. The loophole was this: the Basel Accord stated that banks didn’t need to hold capital resources
for any credit lines that were less than a year in duration. So banks typically extended credit lines to SIVs and conduits that were 364 days or less. The banks did not reckon, though, that the SIVs and conduits would ever need to draw on these credit lines. In normal circumstances, they raised their funding in the short-term commercial paper market. In this market, the SIVs and conduits would sell notes that paid off in only a few months, somewhat like a CD. Those buying the notes were, therefore, extending credit for many fewer days than a year. The cash they raised was used to purchase safe, long-term debt instruments, such as mortgage bonds. They made a tidy profit because their short-term borrowing costs were lower than the returns they made on the long-term bonds they bought. They were thus playing what’s referred to as a “carry trade,” and while the profit margins on this bit of alchemy were small, the SIVs leveraged themselves so much—making substantial purchases of bonds vis-à-vis the amount of capital they had raised—that all in all they made quite reasonable income. The strategy carried a key risk. Leverage not only magnifies gains, it also magnifies losses, and with such a constant need to replenish short-term funding, the SIVs and conduits were vulnerable to finding themselves cash poor. The danger was that if buyers of commercial paper—such as pension fund managers—ever stopped buying such notes, the SIVs and conduits would see their normal funding dry up. And the SIVs were usually required to continuously report the value of their assets at market prices—to mark-to-market. If those values ever dropped precipitously, commercial paper buyers might well decide to stay away. But the SIVs stuck to buying top-quality assets, only those carrying the triple-A tag from credit rating agencies, so the chances of that turn of events seemed vanishingly slim or so they assumed.

One of the truly staggering things about this boom in newfangled credit investment products was that very few nonbankers had any idea that institutions such as SIVs and CDOs even existed. Even regulators seemed only vaguely aware of what the banks were really doing. Yet SIVs were proliferating like mushrooms after a rainstorm. The financiers had created a vast “shadow banking” system that was running out of control.

As the pace of innovation heated up, credit products were spinning off into a cyberworld that eventually even the financiers struggled to understand. The link between the final product and its underlying assets was becoming so complex that it appeared increasingly tenuous. Bankers were becoming like the inhabitants of the cave in Plato’s tale, who—at best—could see only shadows, not tangible reality. Most financiers lacked the cognitive skills to truly understand the connections in this new world. These complex products could not be analyzed with just a pen and a piece of paper, or even a handheld computer or two. The debt was being sliced and diced so many times that the risk could be calculated only with complex computer models. But most investors had no idea how the banks were crafting their models and didn’t have the mathematical expertise to evaluate them anyway. Each player had its own twist on modeling, after all, and as Terri Duhon observed, “Many different investment banks will provide significantly different prices on the same CDO tranche because they are using different models of correlation.”

Investors generally relied on the ratings agencies to guide them through this strange new land, which seemed a rational, easy solution to contending with the complexity. The ratings scale was so simple: if something was triple-A, it was supposed to almost never default; if it was triple-B or triple-C, it had far more risk. In a world where so much else was baffling, those clear-cut designations were wonderfully comforting. Better still, they were backed up by massive research, which was a key element in the rating agencies’ sales pitch.

Like priests in the medieval church, ratings agency representatives spoke the equivalent of financial Latin, which few in their investor congregation actually understood. Nevertheless, the congregation was comforted by the fact that the priests appeared able to confer guidance and blessings. Such blessings, after all, made the whole system work: the AAA anointment enabled SIVs to raise funds, banks to extend loans, and investors to purchase complex instruments that paid great returns, all without anyone worrying too much.

Some bankers warned about the seduction. “People who are focused on ratings alone are prime fodder for the investment banks to stuff [sell] things too,” argued Charles Pardue, a key player on the team that created
BISTRO. “I don’t think we should kid ourselves that everything being sold is fair value. I have been to dealer events where bankers are selling this stuff, and the simplicity of the explanation about how it works scares me…there are people investing in stuff they don’t understand, who really seem to believe the models, and when models change, it will be a very scary thing.”

The ratings agencies, unsurprisingly, were adamant that such concerns were unfounded. Moody’s, Standard & Poor’s, and Fitch had each invested heavily to develop cutting-edge systems for modeling the risks of the full range of new products. To allay fears that their calculations might be faulty, they also had tried to show investors exactly how these systems worked. “We are very transparent in everything we do,” Paul Mazataud, a senior official in the structured finance team of Moody’s, explained in an interview. Moody’s even voluntarily posted details of its own model, called CDOROM, on the internet in 2004. “Our model has become a bit like a template in the market,” Mazataud observed. “Most CDOs are rated with this model, and it is used by management in most synthetic transactions,” he continued, bursting with pride.

Yet such assurances failed to allay the unease of Pardue and others. Precisely because the agencies had diligently posted the details about how their models worked on the net, bankers found it easy to comb through the models looking for loopholes to exploit. And by 2005, they were doing quite a bit of that. Whenever a banker had an idea for a new innovation, it would be run through the agency models to see what rating the product was likely to earn. If it looked too low or high, the design would be tweaked. The aim was to get as high a rating as possible, with the highest level of risk—so that the product could produce all-important higher investor returns. In banking circles, the game was known as “ratings arbitrage.”

Officials at the ratings agencies knew perfectly well that this game was going on. But they felt in a poor position to fight back. Banks had far more resources than the agencies, so they could build better models and hire the smartest structured finance experts. The banks also held the whip hand in a commercial sense. While in the corporate bond world, the agencies rated the bonds of thousands of companies and were not
dependent on any one company for fees, these credit products were being produced by a much smaller circle of banks. Those banks constantly threatened to boycott the agencies if they failed to produce the wished-for ratings, jeopardizing the sizable fees the agencies earned from the banks for their services. From time to time, the ratings agencies took a stand, to show they couldn’t always be pushed around, but they were careful not to offend the banks too deeply. When an agency gave a rating to a CDO, it sometimes commanded a fee of $100,000 per shot, or even several times that level. Moreover, the business was growing fast—so fast, in fact, that by 2005 Moody’s was drawing almost half of its revenues from the structured finance sector; two decades before, that proportion had been modest.

On top of that conflict of interest, the ratings that the agencies were issuing were subject to another pernicious problem. In trying to judge the risk of these products, the agencies faced the same vexing issue that had dogged the old J.P. Morgan BISTRO team when it considered going into mortgage-based BISTRO deals: How could default patterns be modeled? There was so little good data to work with. Was it safe to assume that defaults would play out in the future as they had in the past? Even if so, the historical data was so limited. The trickiest issue of all was working out the level of “correlation”—figuring out how likely it was that one default would trigger others. Different modelers had alternate ways of dealing with the problem, partly because they often selected different pools of data to work from. “The purest information to use is data on [historic] defaults, but the sample is just too small,” Gareth Levington, a London-based managing director at Moody’s, explained. “So we look at correlations on ratings movements. But there are other ways you can do it with equity prices, say.”

Almost all of these slightly different techniques did, however, use the same fundamental mathematical approach—or “statistical engine,” as Moody’s officials called it—which tried to plot the probability of future defaults based on historical data, using a bell curve type of chart that assumed that losses would occur in a relatively steady manner.