Credit Ratings vs. Credit Default Swaps

As an alternative to relying overly on ratings produced by credit rating agencies, several ratings reform proposals offer the usage of bond or credit default swap (CDS) prices or spreads as a more plausible option. Some of these proposals are positively suggestive of the fact that market prices are both more accurate and more predictive than credit ratings.

I’m not convinced.

Firstly, with ratings being so deeply embedded throughout our financial structure, the ratings of the assets themselves become an integral component of the market-implied risk assessment. For example, even when analyzing securitized products Vink and Fabozzi (2009) show credit ratings to be a major factor accounting for the movement of primary market spreads. Thus, for any proposal to be convincing it would have to test the accuracy and reliability of CDS spreads on unrated bonds or companies. Alternatively, a study would need to compare the performance of traded securities whose ratings are not publicly known (also known as shadow ratings) to the performance of those shadow ratings.

Secondly, bond yields (or spreads-to-swaps) and credit default swap premiums are largely incomparable to credit ratings for many reasons. These differences will have to be tackled in a separate piece, but at the very least there’s that non-insignificant concept of liquidity. Both CDS premiums and bond yields include the various risks – not just credit risks – that come with investing in, or buying protection on, a security. Credit ratings speak solely to long-term credit risks.

One may argue that the ratings were far less accurate than CDS spreads during the crisis, and that this (i.e., during a market dislocation) is the only time we depend on accurate default projections and we should therefore abolish rating agencies in general. While I don’t wish to complain of these proposals, I fear that they complain unfairly of the rating agencies.

Yes the CDS spreads may better reflect default probability during a crisis. By definition they’re more adaptive to changing market conditions, versus the ratings which are long-term predictors. But would you want ratings to change in as volatile a fashion as CDS spreads? Would you want ratings to depend on headline news, or on audited (or lightly audited) financial data? Also, one shouldn’t forget that CDS spreads on CDOs and RMBS tranches were just as poor reflections of market-perceived asset quality before the crisis. The crisis could only occur, in part, because the banks were able to buy protection so cheaply from the monolines, by way of being long the CDS -- the infamous negative basis trades.

But even if these proposals made sense and even if their hypotheses were correct, they would be missing at least one crucial point: we need ratings. Meaningful ratings are essential – certainly now. Let me explain why, albeit by way of a long-winded explanation.

For financial reform to be successful it needs ultimately to deal with the flaws in our banks’ risk management procedures – and to deal with them in an environment in which the very serious practice of risk mitigation is left by senior management to risk managers, just as the serious business of growing revenues while attending to shareholder pressure is left by risk managers to upper management.

That these two functions are more adversarial than independent in nature is a concept not to be lost on us. Overly cautious risk management might hinder the implementation of growth opportunities, or the extent thereof. At times, indeed, they may be thought by the skeptic to be mutually exclusive.

Indeed the overpowering pressures that come with business initiatives can influence even the most judicious risk manager’s ability to perform her function in an objective manner, even though her function ought to be both separate from and independent of the business strategies. (See for example “Lehman’s Worst Offense: Risk Management.”)

With both traders and management being compensated for revenue generation, and with prudent risk managers acting only as a hindrance to the initiation and exploitation of growth opportunities, there remains little incentive for senior managers to maintain a healthy risk management environment. Instead of cultivating an environment in which risk managers are educated in monitoring the real risks (which requires expensive resources including personnel, data and systems) they are seen rather as a burden and a cost center, and are therefore starved of the resources necessary to question traders, trades, and trading strategies.

In sum, we remain in the infancy of creating a functioning risk control practice in place at our major banks. We are yet to promote adequate business-peer challenge processes and our price verification processes remain immature. Credit ratings, if created and applied properly, can provide a healthy starting point for internal skepticism; they can provide the independent credit risk assessment that supplements an analysis performed by the front-office or by the back-office.

Conclusion

CDS spreads are untested as a predictor of long-term default probability on unrated securities. Perhaps the reliability of CDS spreads depends on the underlying referenced entity being rated. There’s no doubt that CDS spreads are useful indicators – but I seriously doubt that they’re anywhere near as useful as ratings in predicting long-term default probabilities or losses.

I remain convinced there's an important place in our market for one or more independent agencies to provide their objective opinions in the form of a rating. For ratings reform to be successful, however, requires that the necessary measures be put in place to ensure that rating analysts are unfettered by market share concerns, and are incentivized only by ratings quality and accuracy. If we can achieve these objectives, ratings will return to providing a meaningful utility.

The SEC v. Goldman Sachs

The civil suit filed by the Securities and Exchange Commission against Goldman Sachs revolves around a combination of three patterns we have seen recently:

(1) the ability of external parties to adversely influence the portfolios that support securitized vehicles (see for example the Magnetar trade or the TPG trade);

(2) the harmful effects of the negative basis trade (according to the New York Times article, seven of Goldman’s Abacus deals were wrapped by AIG); and consequently

(3) the capacity for ratings to be “gamed.”

Ultimately, the complexity and opacity of the structured finance product make it susceptible to abuse by poorly incentivized parties. Securitization is abounding with conflicts of interest while informational asymmetries, resulting in various forms of moral hazard, proliferate.

But none of these problems would exist if ratings were always perfect: the negative basis trade would never have existed and the Abacus deal’s reportedly poorly-selected portfolio would never have received the ratings it was able to achieve.

And so we must deal with the very serious business of underwriters “gaming” the ratings system, which jeopardizes both the accuracy and the integrity of ratings and the ratings process.

But first we need to understand the observer effect.

The Observer Effect

In the social sciences any framework or methodology is subject to what is called the observer effect. The observer effect deals broadly with the process by which subjects alter their behavioral patterns on becoming aware of a test’s objectives.

Consequently, behavior induced by the methodology can be different from that on which the historical data was based. The subjects’ response varies based on the incentives of the subjects and the potential complexity of the underlying product.

The problems for ratings become most apparent when the incentives of the subjects (e.g. mortgage bankers) vary significantly from those of the creators of the methodology (i.e. rating agencies). The assumption, or hope, that models and careful study of the historical data can overcome the misalignment of incentives has been shown to be a fallacy: poorly incentivized market participants were able to create products and scenarios never contemplated by historical analysis.

How? Suppose we have a duopoly of rating agencies X and Y. Rating agency X decides that according to its internal calculations, it will more heavily scrutinize a particular quantitative factor – the borrower’s FICO score – when considering the quality of a mortgage for the purposes of its ratings methodology. Rating agency Y, however, discloses that its methodology predominantly hinges on the loan-to-value (LTV) ratio of each mortgage.

As an originator of mortgage loans (or structurer of RMBS securities) if you have a package of loans possessing high FICO scores, capitalistic tendencies will encourage you to approach rating agency X for your rating.

This is known as ratings arbitrage, or “gaming the system.” A more realistic scenario may be that once rating agency X publicly discloses that it considers FICO to be the key driver of mortgage performance, mortgage originators or structurers, will seek to put together a bunch of comparatively cheap mortgages representing high FICO score borrowers but with very poor other qualities and bundle those together in an RMBS to be rated by rating agency X. Originators might specifically target high-FICO individuals, knowing they’ll be able to off-load these mortgages by way of an RMBS to be rated by rating agency X, almost irrespective of the other qualities pertaining to the borrower (e.g. salary) or the mortgage itself (e.g. documentation level). While FICO score may have originally been a key driver of mortgage performance, all else equal, these high-FICO pools now significantly underperform.

Thus, due to what we call customization — or active adverse selection – rating agency X’s RMBS analysis turns out to have been inaccurate or compromised. (The “problem” of customization is not limited to the selection of the portfolio, but may include adverse selection of the securitized vehicle itself, or counterparties to the structure. Like the product itself, the problem is multidimensional.)

If the rating agencies continue to publicly disclose their procedures they will need to be increasingly adaptive and accurate in our computationally-intensive market, lest their rating models be otherwise gamed. They will have to be nimble and swift, like investment banks: they will need to be everything they currently are not.

The regulatory drive towards creating multiple rating agencies presumes that increased competition encourages higher standards. Niels Bohr noted that “the opposite of a great truth is also true.” In this situation, perhaps the flip-side is a greater truth: that as far as creating rating agencies goes, less is more.

Ratings competition was a key driver in the decline in standards, with rating agencies competing on ratings for market share: higher ratings translate into increased market share, which is crucial for publicly-traded companies.

The more models and options available, the easier it is to arbitrage the system, finding the least conservative rating agency to rate each particular pool of assets.