Rather than roll over and play dead in the wake of massive downgrades, market pundits are biting back. Efforts to put a leash on correlation, evidentially the beast with the longest teeth, made headway last week with the release of a new rating tool from Fitch Ratings and a new research paper from Nomura Securities aimed at alerting the market to overlooked dangers.
Accounting for why a wide spectrum of sectors - from airplanes to meatpacking - will all move together during particular credit crunches does not necessarily follow intuition. For CDOs, one would have thought pooling together McDonald's and Boeing debt, along with paper from other non-related industries, would have achieved a comfortable level of diversification. Yet one of the most jarring results of the soured CDO environment has been that diversification, as a defense against credit risk, stumbled in the end zone.
Portfolios said to be diverse, and thus protected from credit events, actually experienced more downgrades and defaults than ever anticipated, according to all three rating agencies.
"These CDOs did not necessarily perform better than less diversified portfolios," noted Richard Hrvatin, managing director, Fitch Ratings.
According to Fitch, so far this year, the agency issued 119 CDO downgrades versus a total of 223 in 2002. "You can see the pace isn't really slowing yet," noted John Schiavetta, Fitch managing director. Up through the month of May, eight tranches were upgraded versus 19 in all of 2002.
Diversity, experts believe, is still useful and not entirely kicked to the curb. The updated theory relies on better diversifying by understanding how and why dissimilar credits behave in similar fashion - or correlation risk.
Fitch launches Vector
Moody's Investor Service and Standard & Poor's, to a certain extent, address correlation in their rating methodology. Fitch has gone one step further and recently launched Vector, an analytical
tool that looks at inter-industry correlation.
"The default correlation is the glue that hangs together the constituents of a portfolio," explained H. Gifford Fong, president of Gifford Fong Associates, which was tapped in a consulting role on Fitch's new Vector product. The firm specializes in analysis and model validation in fixed income. "Portfolio credit risk has relevance to fixed income in general," he said.
Currently in the rollout stage, Vector targets correlation between industries. The analytical tool does not replace traditional ratings, according to Hrvatin, and will be a mainstay in the market by mid-summer.
"That underlying CDO collateral has correlation, other than zero, is not a new concept - its measurement was not formally introduced in the quantification of risk," said Val Goldstein, a structured finance consultant. "Just as the transition defaults matrix is not a constant, the correlation matrix is not a constant over time," he said.
The Vector model is based on the probability of default from an idealized default curve. According to the rating agency, it first defined 25 specific corporate industries. Then, by looking at correlation between issuers and industries, the agency found Vector useful in gauging portfolio diversification risk correlation. It believes investors will agree.
"The real world is not a point estimate. You need more dynamic underpinning (in modeling)," argued Fong. "A correlation number is not a static number. This suggests that if you have a single-step approach, but are trying to use correlation, you are trying to model something that's a moving target," he said. "Capturing the behavior of correlation with a single step is, perhaps, impossible."
In its early simulation runs, Vector revealed that across the 25 Fitch-defined industries in the U.S., the highest industry on average was broadcast and media, at 30%; conversely the lowest industry on average was energy, at 13%. Correlation ranges were from 8% to 37%.
In addition to the U.S. industries, Vector currently encompasses 25 industries in Europe, with Japan to be included soon. Five classes of ABS securities have also been modeled for correlation, although Hrvatin stated that those correlation figures are in final development.
Nomura considers model risk
Modeling with limited data is in itself a risk and compared to other financial instruments like bonds, the timeline of CDOs spans less than two decades; so does its data. As the securitization market hummed along happily during most of that time, some of those overlooked or under compensated for risks created wrinkles in models. Now some experts believe these inconsistencies in the models, in part, contributed to the record number of downgrades in deals seen over the last 18 months.
A recently issued detailed research piece from Nomura Securities suggests that historical reliance on models that ignored correlation may have contributed to some of the downgrade activity.
"Our assumptions affect our models very strongly. We want to believe in the tools we create," said Mark Adelson, head of asst research for Nomura. "Sometimes we run the risk that we are putting important things aside for the convenience of using a model."
Model risk, Adelson contends, partially explains the poor credit performance of certain areas of the structured finance market over the last two years, namely CDOs as well as aircraft ABS, franchise loan ABS and mutual fund 12b-1 fee ABS. In looking over market data from 2001 and 2002, Adelson took note of similarities he found in the poor credit performance in all four of these areas, despite the fact the instruments are very different from one another and despite the fact they are diverse packages of credits. To Adelson, the signs pointed to the inherent dangers with modeling, as well how far-reaching correlation had become.
The market compensates for unknowns by building certain levels of expected risk into models. Modeling efforts have gone so far as to create a different style of risk simulations for these four securities, due to their shorter track record, than had traditionally been done for other securities - the multi-step Monte Carlo vs. actuarial methods, for example. In this sense, CDOs, and securitizations overall, have become a market very indebted to the land of academia.
In a research report titled "What a Coincidence?" released last week, Adelson sought out answers from the experts in academia. Many influential names, from J.M. Keynes, published in 1936, to R.K. Skora, published in 1998, have sought to quantify risk and, as with Skora, the effect of correlation.
In turning to academia, Adelson aimed to find solutions. He wound up, however, with more questions. The headache posed by time-varying correlations, correlations that change with the passage of time, was one such quandary.
"I can't say I've created a lot of new ideas," said Adelson. "But I've pulled things together that haven't been before and I've brought that into structured finance, an area where these items haven't been publicized so much," he said. Skora, said Adelson, caught the correlation issue years ago, as a really important point that represented a threat to performance in CDOs. "That was a research item I had never seen before," he said.
"Investors should favor deals where credit risk is analyzed by actuarial methods," said the report. "These deals generally benefit from having less model risk than deals where risk is analyzed by Monte Carlo simulations, or similar techniques that rely on too many assumptions."
While Adelson champions analytic methods, he believes it is too harsh to say that the rating methodologies of rating agencies failed. What is fair to say it that the securitization industry, collectively, did not fully appreciate the potential effects of the methodologies' limitations, he said.
"Had such players been able to anticipate - either quantitatively or qualitatively - the havoc that correlation effects can wreak, they probably would have demanded higher yields or stayed out of those sectors altogether," said Adelson.
While his research has been ongoing since January, Adelson toiled through the last few weeks to get Nomura's findings out to the market in time for Vector's road show, but his report was not meant as an endorsement of the new model. He published his findings last week to make use of the interest and heightened awareness in modeling and correlation.
As an analytical tool, Vector incorporates correlation; to what degree of success, as with anything new, remains to be seen. Nomura's study probes prevailing conventional wisdom in modeling and brings to light exhaustive work on correlation from academia. Investors should be aware that the two are distinctly different.
"Both checkers and chess are played on a similar board with the same number of pieces but you would be hard-pressed to say that checkers models chess," noted Goldstein.