On April 10 , 2019, S&P Global Ratings published a request for comment (RFC) on proposed revisions to its criteria for analyzing collateralized loan obligations (CLOs) and corporate collateralized loan obligations (CDOs; see "Request For Comment: Global Methodology And Assumptions For CLOs And Corporate CDOs," published April 10, 2019). Following feedback from the market, we finalized and published our criteria, titled "Global Methodology And Assumptions For CLOs And Corporate CDOs," on June 21, 2019.

We'd like to thank market participants who provided feedback. We made no substantive change to the RFC based on market comments. The first part of this article outlines our replies to the comments and the second part describes the changes and clarifications we made to our guidance for applying the criteria.

S&P Global Ratings' Response To Market Feedback On The Proposed Criteria

Comment

The performance data over the last decade provides no additional insight into the likely performance of CLO pools in an extreme level of stress. In fact, the performance of speculative-grade corporates in this period was better than the worst observed, hence probably closer to a modest ('BB') stress scenario.

Response

We agree that performance in the past decade has not been representative of an extreme level of stress and our 'BBB' target scenario default rates (SDRs) remain calibrated to the worst observed default rates since 1981. However, we do believe there is value in the information available from the last recession and have conducted backtesting to test that the adjusted 'BB' SDRs reflect this recent performance data. (see "Credit FAQ: Understanding S&P Global Ratings’ Request For Comment On Proposed Changes To Its CLO And Corporate CDO Criteria," April 10, 2019)

We believe there is variability in projected default rates in an extreme stress scenario, which supports the range of outcomes for 'AAA' SDRs calculated with the calibration of these updated criteria.

Below, we elaborate further on the data used, what changed in the calibration, and why we believe the outcome is consistent with our rating definitions.

As in our previous corporate CDO criteria, our target SDRs for a moderate ('BBB') level of stress are informed by the post-1981 maximum observed default rates for each rating and tenor. Each maximum cohort default rate may come from a different time period. For example, the worst seven-year 'B' cohort started in June 1998, the worst five-year 'B' cohort started in October 1998, and the three-year 'B' cohort began in September 2000.

We continue to use the same targets for a 'AAA' scenario as in the previous criteria. However, these targets were previously associated with highly diversified pools, much more so than the pools of CLOs we have rated in the past decade. This means that for the previous criteria, our targets effectively constituted minimum SDRs for the projected portfolio default rates, as SDRs for less diversified CLO pools would have generally been higher. In these revised criteria, we redefined the targets to reflect the portfolio default rate for average pools, which we refer to as archetypal pools. These have 105 homogeneous assets distributed across 23 industries. This means that some of the actual CLO portfolios may have higher (lower) default rates than the archetypal pool if they are less (more) diversified than the archetypal pool.

To achieve the target scenario default rates using the new archetypal pool, we modestly lowered the asset default rates. We further adjusted the rating quantiles used in the model (CDO Evaluator) to achieve the targets and preserve monotonicity across rating levels and rating differentiation. The overall effect of the calibration for CLO pools, given their composition and diversification, is a slight decrease in SDRs.

We assessed the impact of this change on SDRs at all rating levels, and we believe the revised criteria remain consistent with our rating definitions. Particularly for the 'AAA' SDRs, we analyzed the dependency of corporate defaults on key macroeconomic variables. The intent of this analysis was to build a relationship between corporate default rates and macroeconomic variables and make inferences about the projected levels of default in an extreme stress scenario. Our analysis showed that projected default rates for 'BB' and 'B' rated assets in an extreme level of stress support the 'AAA' SDRs computed using the revised criteria (see paragraphs 87 to 89 in the criteria article).

The macroeconomic analysis established a relationship between the performance of corporate 'BB' and 'B' rated obligors and key macroeconomic variables over the past several decades since 1981. The performance data includes information about the creditworthiness of corporate obligors, at the obligor level. The macroeconomic factors comprised financial market information, business cycle characteristics, and credit market information. We used four covariates for the regressions: slope of U.S. Treasury yield curve (the difference between the 10-year and the one-year yield: financial markets), return volatility on the S&P 500 (financial markets), real GDP growth (business cycle) and the yield spread between 'Baa' and 'Aaa' rated corporate bonds (credit markets). There are inherent limits in our regression analysis as the post-1981 period did not incorporate an extreme level of stress. However, we find this analysis to be very useful for gaining insight into how defaults are linked to the macroeconomy. We acknowledge data limitations for the rated universe from periods like the Great Depression and the need for robust data (such as from the post-1981 period) to perform accurate statistical analysis.

Assuming the above relationship, we then calculated up to 10-year projections for defaults using macroeconomic variables from the Great Depression. We used four periods of 10 years that were relevant: 1928-38, 1929-39, 1930-40, and 1931-41. Using the actual macroeconomic variables from those periods, we preserved the historical data and the relationship between the macroeconomic variables. Using four periods of 10 years as opposed to a single one, we obtained four default projections for up to 10 years that show the sensitivity to the specific dataset we used. We concluded that the projections for tenors of five years and more supported our targets and exhibited a range of potential outcomes as we had expected, depending on the specific scenario used (see criteria paragraph 87). Moreover, our projections of portfolio default rates within the range of, for example, 45%-60% for a five-year horizon and 'B' rated assets, imply that a single projection for a five-year horizon cannot capture all economic responses and trends during a period of extreme stress such as the Great Depression, which was also followed by a slow recovery.

For four years and less, our targets remained informed by historical studies of bond defaults during the Great Depression and around the time of World War I, such as the Hickman study (see criteria paragraphs 82 to 84). We used the aggregated default data from such studies and did not rely solely on the econometric model.

The variability of the projected default rates in the extreme stress scenario, described above, supports the range of outcomes for our 'AAA' SDRs calculated with the calibration for the archetypal pools. The band of potential outcomes for default rates in an extreme stress scenario also supports the decrease in 'AAA' SDRs for existing transactions.

Further, as our calibration exercise decreased SDRs at all stress levels beyond 'AAA', we performed a backtesting exercise to assess how model-projected SDRs compare with the empirical data for actual defaults in the respective pools. The most relevant for this exercise were model-projected SDRs for 'BBB' and 'BB' stress scenarios. Our sample included pools initiated between 2005 and 2008. Given their varying starting dates, they were affected differently by the 2007-2009 recession. Some of the earlier pools benefited from the macroeconomic environment in the years prior to the peak of the crisis, and some from that in the post-recessionary years and the subsequent recovery. To complete our backtesting analysis, we compared the projected portfolio default rates with empirical data for defaults in a moderate ('BBB') and a modest ('BB') stress. In our view, the two scenarios define boundaries for the historical performance of the pools, given their different impact during the recessionary period, and show that our calibrated model remains consistent with our ratings definitions.

Comment

We understand that the diversification of the archetypal portfolio should reflect recent CLO pools. However, we do not understand or believe it is appropriate for 'AAA' SDRs to decrease, including for archetypal pools of five-year 'B' rated assets which are typical of CLOs, as suggested in table 7 of the RfC.

Response

We believe the criteria calibration is enhanced by the use of archetypal portfolios that are more similar to actual pools that are securitized and produce ratings consistent with our ratings definitions, including at a 'AAA' rating level.

In our previous criteria, the 'AAA' target SDRs were associated with highly diversified pools of assets. As a result, most typical CLO pools, which are less diversified, achieved higher 'AAA' SDRs than the target SDRs associated with the highly diversified pool. With the criteria now associating the same 'AAA' target SDRs with a less diversified pool more representative of the pools we typically see, actual CLO pools will now have 'AAA' SDRs that are either higher or lower than this 'AAA' target, depending on whether they are less or more diversified than the archetype.

We do believe the ratings are consistent with our ratings definitions when applying the revised criteria. Our macroeconomic analysis suggests a variability of potential projected default rates in a 'AAA' scenario, which supports our view that the updated SDRs are consistent with our definition of a 'AAA' rating. For a fuller explanation, see the answer to the previous question.

Comment

The proposed criteria article does not explain why the worst observed default rates that inform the target SDRs for a 'BBB' stress scenario (see table 1) are lower than they were in the 2009 version of these criteria.

Response

The worst observed default rates presented in table 1 in the revised methodology are based on an analysis of global CreditPro transition and default data instead of using Creditpro data only from the U.S., Canada, Western Europe, Australia, and New Zealand in the previous version of the criteria. However, this adjustment results in changes to the 'BBB' targets, which are not material and thus do not result in any changes to our calibration. Prior to this request for comment, we had also made limited changes to our target SDRs for the 'BBB' stress scenario between 2009 and the RFC. We provide more detail about our calibration of 'BBB' target SDRs in the first answer above.

Comment

It would be more appropriate to continue using percentiles in the breakeven analysis so that outliers are still excluded.

Response

Our criteria still allow outlier runs to be excluded. However, because of the revisions we made to the cash flow analysis, we believe outliers should be less frequent and therefore the use of a percentile break-even default rate (BDR) is no longer needed.

We focused our cash flow analysis on a smaller number of the most relevant scenarios that we believe are the key drivers of the likelihood that a CDO note will be repaid in full. By doing so, we expect it less probable that these scenarios generate outlier BDRs that are not relevant to our analysis. In addition, with a reduced number of scenarios, we believe the use of a percentile of BDR distribution becomes inappropriate from a quantitative perspective. For both these reasons, in general we maintain the use of the minimum BDR approach in the final criteria. In the unlikely scenario, however, that a particular run does not appear to be relevant, we may decide not to consider it in our analysis depending on the specific facts and circumstances.

Comment

The criteria change is based on performance data from a decade that was very benign for corporate risk, which may not be representative of performance in a more stressed environment, so it is not clear what supports the recalibration. It is also ill-timed as we are in the late stage of a credit cycle; it will encourage further flexibility in transaction structures at a time when corporate leverage and loan covenants have deteriorated, instead of fostering the standardization that is desirable in this market.

Response

We believe we have robust analytical grounds for the recalibration, and we expect the revised criteria to lead to appropriate ratings regardless of when the next economic downturn occurs. Our criteria are calibrated such that each rating level is associated with a specific stress scenario; these rating definitions do not change as a function of where we are in the credit cycle, and as such our criteria is not calibrated to a specific point in the cycle.

We agree that performance in the past decade has not been representative of an extreme level of stress. However, our 'BBB' target SDRs remain calibrated to the worst observed default rates since 1981. We do believe there is value in the additional information available from the last recession and have adjusted our 'BB' SDRs to reflect that.

Our criteria calibration for higher degrees of stress is consistent with our ratings definitions, with SDRs, and other rating assumptions becoming more stringent for higher rating levels, as they are expected to withstand increasingly harsh levels of economic stress. For example, we expect our 'BBB' rated notes to survive a moderate level of stress (see "Understanding S&P Global Ratings' Rating Definitions," published June 3, 2009). For more details, see the first part of this document.

With respect to conditions in the leveraged finance market, our criteria also provide for the use of recovery ratings, and these are currently available for the vast majority of loans we see in CLO portfolios. Recovery ratings reflect our forward-looking view of recovery prospects for a specific loan of a specific corporate and hence are intended to take into account characteristics such as higher leverage, smaller debt cushions, covenant-lite status, and other factors we believe will influence the level of recoveries in the case of a default. For example, in recent years, the proportion of loans with lower recovery ratings has increased substantially (see "When The Cycle Turns, Assessing How Weak Loan Terms Threaten Recoveries," Feb. 19, 2019; "Lenders Blinded By Cov-Lite? Highlighting Data On Loan Covenants And Ultimate Recovery Rates," April 12, 2018; and, "Lean Senior Debt Cushion Threatens Recovery Prospects For U.S. Leveraged Loans," Nov. 30, 2017). In our rating approach for CLOs, weaker recovery ratings lead to lower recovery assumptions and hence higher enhancement levels, all else equal, for a given stress scenario.

Additionally, the criteria contemplate the use of tools such as CDO Monitor, which enable ongoing testing of the credit quality of portfolios prior to reinvestments, which we believe limits the ability for collateral managers to lower the credit quality of the CLO in a way that would be inconsistent with its rating.

Finally, we note that market participants determine which transaction structures and characteristics they consider appropriate. We believe these criteria constitute an adequate framework to enable S&P Global Ratings to analyze these transactions and for us to provide an opinion about the creditworthiness of the instruments issued. The criteria do not constitute an endorsement of one or another structural feature.

Comment

Asset default rates should not decrease as we are on the verge of an economic downturn when defaults rate are likely to go up. Assuming the cycle turns, the ratings on CLOs based on the proposed methodology will be affected negatively. At least then the quantiles should be made more stringent to reflect higher correlation risk when default rates increase in a possible downturn, and given S&P Global Ratings is not proposing changes to its correlation assumptions. Finally, S&P Global Ratings should reduce recovery rate assumptions to reflect the increasing proportion of covenant-lite loans for which the average recovery has been constantly decreasing over the past years.

Response

Asset default rates, pairwise asset correlations, and rating quantiles are key parameters calibrated to achieve portfolio scenario default rates (which increase with higher stress levels) that we believe are commensurate with our ratings definitions (see above). Our ratings definitions do not change with the credit cycle, and as such our criteria are not calibrated to a specific point in the cycle or for a possible downturn. We have adopted a target-based approach to minimize model dependency and inform our output based on credit fundamentals and our credit view, rather than model choices. Our calibration to the 'AAA' and 'BBB' targets is focused on key assumptions as a whole, rather than individually. These assumptions are informed by historical data and calibrated to achieve our desired targets. We also consider the stability of ratings in calibrating this methodology, and believe these proposed criteria meet our credit stability criteria (see "Methodology: Credit Stability Criteria," May 3, 2010). The revised asset default rate assumptions we use in our analysis are informed by historical default data and, for 'BB' and 'B' rated assets, for example, they remain higher than the average default rates experienced since 1981. Correlation assumptions address the interdependency of defaults of separate credits within an asset pool. For the purpose of calibrating portfolio scenario default rates, we make certain assumptions about correlation, including the assumption that correlation is likely to remain constant over time, as well as being uniform across many industries within our classification system. While these assumptions are, by their nature, qualitative, we believe they are reasonable for reducing the complexity of the modeling process and enhancing its transparency. We base calibration of the quantiles on our target portfolio default rates for 'AAA' and 'BBB' scenarios and aim to ensure a proper rating differentiability and monotonicity across the entire rating scale.

Regarding recovery rates, we believe our assumptions are robust. They primarily rely on the use of recovery ratings, which we assign on most loans found within broadly syndicated CLOs (see above). We also reviewed the calibration of recovery rate assumptions for assets that don't have a recovery rating against the most recent historical observations, considered the level of stress experienced at the time and concluded that those assumptions remain valid.

Comment

It is more appropriate to use an average maturity for the pool of assets that is extended to take into account the length of the reinvestment period, rather than the pool's actual weighted average maturity as proposed in the RFC.

Response

The use of the assets' actual maturity profile in our analysis is related to our "stable quality" rating approach, described in the criteria. We apply this approach where the collateral manager commits to generally maintain or improve the consistency of the proposed portfolio's credit quality with the notes' original rating as a condition of reinvesting, for example, using S&P Global Ratings' CDO Monitor. In this case, we reflect this ongoing commitment by focusing our credit analysis primarily on the characteristics of the actual portfolio. For example, when the pool is modeled at closing using the actual maturity, our ratings reflect the credit risk of that particular pool. If a later reinvestment resulted in an extension in time of the credit risk, that reinvestment would only be possible provided that the extended exposure remained consistent with the notes' rating or did not deteriorate the portfolio's credit risk further. We believe, therefore, that the change to using actual maturities in the stable quality approach does not result in increased risk compared to the previous approach.

As an alternative to this approach, we may also rate CDO transactions based on their covenants and the length of the reinvestment period in particular, under the "stressed portfolio" approach.

S&P Global Ratings' Changes And Clarifications To The Proposed Guidance

Changes

Change: We removed the guidance regarding recovery assumptions for unitranche loans

Rationale

We received mixed feedback from market participants about what they believe constitutes a unitranche loan compared to the description we proposed in Appendix E of our RFC. As a result, we removed the definition. Instead, we will apply our recovery assumptions based on loan-specific characteristics described in the transaction documents.

Change: We updated the guidance relating to the cash flow analysis of combination notes

Rationale

It came to our attention that combination notes with equity components may be subject to very strong upward rating movements in a very short period of time after issuance, in scenarios where we assume no credit to excess spread in our analysis and if a limited amount of excess spread turns out to be available in the first year after issuance. We believe we can address this by taking into account our outlook for the sector in making that assessment, and contribute to a better stability of our ratings.

Change: We gave more guidance about defaulted assets for coverage tests

We supplemented our guidance description of what assets we believe may be carved out of the definition of defaulted assets for the purpose of coverage tests, consistently with our rating criteria. This can now also include obligors that have defaulted but that have emerged from bankruptcy, which we believe may be treated as 'CCC-' rated assets, pending the assigning of a new rating.

Rationale

This situation has arisen in the past, and we believe it is a situation similar to debtor-in-possession and current pay loans.

Clarifications

Clarification

We clarified the language in the guidance relating to our expectations regarding the ability of a collateral manager to extend asset maturities.

Clarification

We removed the reference to discounted securities as an example of application of guidance relating to the calculation of market value. This was previously inconsistent with the guidance presented in table 12, which remains unchanged.

Clarification

We also simplified the examples presented in table 13 relating to stresses applied in our analysis to 'CCC' rated and current-pay assets.

Related Criteria

• Global Methodology And Assumptions For CLOs And Corporate CDOs, June 21, 2019