Responding to market participants' demand for more precise and granular measures of risk, Fitch Ratings conducted a study of the level of program-wide credit enhancement (PWCE) with a sample of large, diverse multi-seller ABCP programs in the U.S. Using VECTOR CP as the analytical tool, the study suggests that many partially-supported multiseller ABCP programs in the U.S. maintain excess levels of PWCE. This information should be of comfort to investors as it confirms the low level of risk in established U.S. multiseller programs. In addition, multiseller ABCP program sponsors may find the application of VECTOR CP useful, as it can compliment their internal credit scoring and support their risk and regulatory management objectives. A more detailed discussion with respect to the modeling assumptions and results will be presented in a report to be published by end of the month.
Multiseller ABCP programs generally calculate their PWCE levels by applying a fixed percentage against the outstanding amount excluding any "highly rated assets". These percentages typically range between 5% and 10%. However, the collateral risk that multiseller ABCP programs take ties into the underlying asset portfolio, and this risk greatly depends on the distribution of asset type, asset size, and asset rating, while fixed percentages are not reflective of these factors. Given this disconnect, the study's objective was to measure this portfolio risk using VECTOR CP and compare the model's outputs to the actual levels of PWCE using representative programs.
While Fitch rates a significant number of multi-seller ABCP programs, this study's sample included partially supported multiseller programs sponsored by Bank of America, JPMorgan Chase, Royal Bank of Canada and Sumitomo Mitsui Banking Corp. The inputs to VECTOR CP were based on the latest portfolio information made available for each program. Except for explicitly rated assets, bank internal scores were converted to Fitch equivalent ratings and used as inputs. Fitch views banks' internal scores to be a reliable indicator of risk, especially with respect to large and established banks typically associated with mature and seasoned multi-seller ABCP programs.
The table on this page shows the results of base case model runs. All programs had PWCE well in excess of the amount calculated by the model. The program with the smallest margin still had PWCE that exceed the VECTOR CP level by 3.7%, which translates to a multiple of 1.6. The program with the largest margin had 16.4% of extra enhancement that is more than nine times the amount calculated by VECTOR CP. The majority of programs had excess enhancement of 5% to 8% and multiples between two and four. Ranger resulted in the highest VECTOR CP output and the lowest multiple mainly because of one large explicitly rated security (wrapped by a monoline bond insurer) that was supported by an "eligible liquidity facility" (per the new capital rules) that provides 0% support should the security (or the monoline bond insurer) be downgraded to non-investment grade. The model's inputs are adjusted to account for the weaker support provided by such eligible liquidity facility and consequently may result in a higher output level.
For each subject program, model runs were repeated with varying levels of transaction rating adjustments from one to four sub-rating-categories lower. These adjustments were applied only to ratings derived from bank internal scores, and ratings of explicitly rated transactions remained constant. By adjusting the converted rating downward, it would not only add a degree of conservatism, it would allow a break-even analysis and get a sense of how much buffer is afforded by the available PWCE. When VECTOR CP is formally used to evaluate multiseller ABCP programs, the level of adjustments may be scaled based on the bank's level of experience in managing ABCP programs, the soundness of its credit underwriting process, and its administrative and operating capabilities.
The results with downward rating adjustments are shown in Chart 1. On average, each downward adjustment increased the credit enhancement amount by 22%, although the variance was large. In fact, Ranger exhibited no sensitivity to the adjustments and its VECTOR CP level remained at 6.1%. This may appear to be a strange phenomenon, but it is possible especially when there are large explicitly rated transactions that were not subject to the rating adjustment. As mentioned earlier, there was one explicitly rated transaction that acted as the main driver in determining the model's outcome for Ranger. This was the case when converted ratings on unrated transactions were not adjusted, but the security remained as the main driver even when converted ratings on unrated transactions were adjusted downwards through four rating sub-categories, resulting in a constant 6.1% enhancement. Because of this, although Ranger had the narrowest margin under the base case assumptions, it had a larger buffer than most other programs when the four-sub-category adjustment was applied to the converted ratings.
Overall, all of the programs had sufficient enhancement even when internal ratings were conservatively adjusted down by three categories. Surprisingly, seven of the programs still had sufficient enhancement even when applying an extremely conservative four level adjustment. This outcome demonstrates the robustness of protections afforded by these programs.
Additional model runs were performed with varying levels of correlation. Correlation levels were increased to levels that are 0.1 and 0.2 points higher than the levels originally estimated by VECTOR CP under the base case model runs. Similar to lowering the converted ratings of assets, increasing correlation for multiseller ABCP programs with PWCE has the effect of "stressing" the model runs and it provides a more conservative treatment against the possibility that the correlation assumptions may not be accurate. By increasing the correlation by 0.2 point, most programs had correlation above 0.4, which is a correlation level rarely estimated as high even for an intra-industry asset pair.
The results of increasing correlation to portfolio with asset ratings adjusted downward by four sub-categories are shown in the graph above. The model's outputs increased on average by 26% for every 0.1 point raise in correlation. In addition to the two programs that failed with just the downward rating adjustment by four sub-categories, one more failed with 0.1 point correlation increase, and two others failed with 0.2 point increase to correlation. Close to half of the programs still passed under this severe stress case, which further underscores the robustness of credit enhancement provided to these programs.
(c) 2006 Asset Securitization Report and SourceMedia, Inc. All Rights Reserved.