So, for history buffs, I am going to quickly review the basis on which the ratings companies quantified credit support requirements for private issue MBS in the first ten or fifteen years of mortgage securitization. And for those who endure the walk down memory lane, I am going to also highlight a great little study by UBS structured products analysts that addresses the question ‘which companies ratings are the most reliable?’
But first, a comment on terminology: I insist on using the term ratings companies. Calling them “agencies” makes it sound as if they perform some kind of administrative or quasi-governmental function, as if they function as watchdogs or ombudsmen for investors’ interests. The performance of rated subprime, Alt-A and ABS CDOs gives the lie to that notion. These enterprises may assert a freedom of speech defense, model error, or outlier event (like a hundred-year storm or thousand-year wave), but the reality is that each sold their ratings in a competitive marketplace, like any other for-profit firm.
Where it all started
A recent consulting project had me scrambling around the files in my attic, where I unearthed a short piece from S&P, “Texas Default Study Confirms Loan-Loss Assumptions,” originally published in November 1990 and reprinted in February 1993. The title refers to the assumptions that once underlied S&P’s ratings of mortgage-backed securities.
S&P originally defined its loan-loss assumptions in the mid-1970s – it rated the first private MBS in 1977. At the time, there wasn’t much contemporary information on which to base extreme stress scenarios. In the decades following WWII, credit losses on existing mortgage portfolios were insignificant – nationwide foreclosures amounted to fewer than one-half of 1 percent of all conventional loans (not FHA/VA). To develop a worst-case, benchmark scenario of mortgage foreclosures and losses, the company had to look back to the Great Depression.
Although data was very limited by contemporary standards, S&P was able to derive base foreclosure-frequency assumptions from a study of the behavior of urban mortgage loans originated by 24 life insurance companies between 1920 and 1946 (published by the NBER). Based on this study, S&P defined a AAA depression as one in which 15 percent of all borrowers in the lowest risk category will default, a AA depression as one in which 10 percent will default. (In other words, to be rated AAA, a bond would require credit support that would withstand a number of iterations of the AAA depression scenario.)
The benchmark loan is a first-lien mortgage on an owner-occupied, single-family, detached house with an original LTV of 80 percent or less. (At the time, it was also fully underwritten to a borrower with good credit. Loans made in the early 20th century by banks and insurance companies tended to have low LTVs, shorter maturities, partial amortization and bullet repayments. Thrifts, holding about 1/5th of residential mortgages, introduced the 30-year, fully amortizing loan.) Foreclosure frequencies would be adjusted higher for loans with additional risk factors, including historical delinquencies and severities, lien type, loan type, geographic concentrations and borrower quality.
The other component of the loss formula is loss severity, as most in the industry know. Again, S&P’s original benchmark for market value losses was the Great Depression. In an extreme stress scenario, market value declines would be a major component of loss. (Other components include unpaid accrued interest, legal and selling costs, property maintenance, and so forth. The severity of these vary from state to state as well as with economic conditions.) Based on Depression experience, S&P had determined the market value of single-family detached properties would decline by 37 percent under the AAA depression scenario, 32 percent under the AA scenario.
Moody’s began rating RMBS in the 1980s, and it also originally used the Great Depression as the basis for its credit enhancement requirements. Moody’s identified positive correlations between mortgage performance and economic events, which in turn were employed in a Monte Carlo model to generate a worst-case loss distribution for a pool. Credit protection was quantified based on that distribution – for instance, Aaa losses equated to approximately three standard deviations from the mean, Aa losses about 2.5 SDs and A losses 2.0 SDs and so forth. Ultimately, credit enhancement levels were defined such that expected losses would result in maximum declines in annual yield consistent with expected basis point yield changes for similarly rated corporate bonds.
The birth of modern ratings models
The ratings companies got their first look at a contemporary benchmark-quality housing downturn during the mid-1980s in Texas, oil-patch and southwestern states. These regional housing depressions coincided with the collapse of domestic oil and gas booms, accompanied by overbuilding fed by easy credit from competing thrifts and banks. Citing Fannie’s study of loans in its portfolio as well as foreclosure and house price data from other sources for Houston (Harris County, TX, one of the hardest hit markets), S&P affirmed its methodology.
Texas/Oil Belt experience is of particular interest in the present crisis. The Fannie study indicated lifetime default rates of 8.5 percent on Texas loans originated in 1981-1983, while other data indicated foreclosures in Houston between 1980 and 1989 amounted to 16 percent of housing stock. Houston home prices declined about 30 percent. Likewise, S&P found loss severities reported by Fannie (and largely attributed to the Oil Belt states) were easily in the ball park of Depression-based assumptions: the GSE charged off 25 percent of aggregate principal balances of foreclosed loans in 1987, 28 percent in 1988 and 31 percent in 1989.
We can all humbly hope that OFHEO does not have the “opportunity” to update its benchmark using the worse experience now being generated in the current foreclosure debacle.
The fresh Texas experience was also reflected in other MBS rating methodologies. Fitch, whose original methodology was similar to S&Ps, devised new benchmarks based on the 1980s Texas depression (using the Fannie as well as FHA data). Moody’s, already using a Monte Carlo model to size credit support, augmented its data sets with the Texas information, as well as its own surveillance data from the early 90s housing downturns in California and the Northeast.
Bear in mind, private MBS issuance was negligible until 1986, when tax code changes enabled issuers to use a senior/sub credit structure. Still, issuance averaged just about $11 billion a year through the end of the decade. The private MBS market exploded with the early 90s rally, jumping from $24 billion in 1990 to $98 billion in 1993 (issuance statistics from Inside Mortgage Finance Publications). This massive increase in rated deals, as well new attention given to reporting by most issuers, gave the ratings companies machine-readable performance data from which to build risk models.
It also provided an excuse to stray from the Great Depression benchmark.
When the Fed brought the mortgage rally to a halt in 1994, originator/issuers shifted from prime to off-prime loans to keep their pipelines full. These were the loans stranded when the thrift industry collapsed in the late 1980s and ignored in the 90s rally. Subprime (called home equity at the time, because borrowers were required to make down payments as large as 50 percent) and Alt-A (originally, good borrowers, but not benchmark, fully documented loans) lending both got their start during this lull.
This watershed change in mortgage lending practices coincided with the transition to newer models at the ratings companies. By 1995, S&P was using the predecessor of its LEVELS model, and Moody’s had refined its model to accommodate the new borrowers and vehicles. At the same time, the GSEs and and big private lenders introduced their automated underwriting systems, beginning the sweeping shift away from fully documented manual underwriting relying on a complete credit history. The age of the machine – and the almighty FICO - had arrived.
Although they may still assert the underlying Depression-style stresses still lurk in the depths of the ratings models, the addition of data has effectively recalibrated those stresses to the performance of thousands of pools issued since 1987, the preponderance of which have enjoyed unprecedented home price appreciation.
The result is that the lessons of the Great Depression ended up buried under an avalanche of more recent good news. Put another way: there is no variable in newer rating methodologies for “originated in a period of free money, easy credit, lite documentation and double digit home price appreciation.”
UBS takes a look at credit risk
Fast forward to the present. A year or so into this disaster, with no end in sight — John Mack’s pep talk to Morgan Stanley shareholders notwithstanding — UBS structured product analysts have performed a fascinating and simple experiment on Moody’s, S&P and Fitch subprime ratings. Taking as their sample the bonds underlying the four rolls of the ABX, they asked which company rates them lowest and, given the bonds UBS expects to default on its model, where do the three ratings companies rate them?
The sample includes 400 bonds, all of which have S&P and Moody’s ratings. Fitch on the other hand, only rated 200. This is a direct reflection of competitive forces in the ratings industry.
Fitch has always been third in terms of market share in the RMBS market, but its share shrank pretty dramatically as the subprime market expanded. In part, this reflects investors’ reflexive reliance on the two larger, older bigger name companies, but most market insiders think it also is a function of Fitch’s tougher credit criteria (which translates into less profitable “execution” for the originator customers of underwriters).
Certainly in UBS’ experiment, Fitch is the most conservative. On average, Fitch currently rates the ABX bonds an average 2.3 rating notches lower than S&P, S&P 2.7 rating notches lower than Moody’s. Of the 400 bonds in the index, UBS analysts predict that 292 will default. Only 134 of those are rated by Fitch, but all are rated AA+ or lower, and 77 are rated CCC or worse. By contrast, of the bonds expected to be written down, Moody’s still rates 35 of them AAA, S&P 24. By both tests, also-ran Fitch is the most conservative.
Which leaves us looking to the future: While the pols, policymakers and public interest groups are busy pushing forward high-profile, spintastic solutions, common sense suggests we start by putting the current mess into historical perspective – a lot of things we used to do in mortgage finance weren’t broke until we fixed them.