Column Poor standards, moody ratings

In response to the civil lawsuit filed by the US Department of Justice in February 2013, Standard & Poor?s affirms that its ratings were ?objective, independent and uninfluenced by conflicts of interest?. This column presents empirical evidence opposing this claim. The data suggests a systematic rating bias in favour of the agencies? largest issuer clients.

While the US Department of Justice did not give any statistical evidence in its deposition, our new research (Efing and Hau 2013) suggests that rating favours were indeed systematic and pervasive in the industry.

In a sample of more than 6,500 structured debt ratings produced by Standard & Poor?s, Moody?s and Fitch, we show that ratings are biased in favour of issuer clients that provide the agencies with more rating business. This result points to a powerful conflict of interest, which goes beyond the occasional disagreement among employees.

The beneficiaries of this rating bias are generally the large financial institutions that issue most structured debt; they in turn provide the rating agencies with most of their fee income. Better ratings on different components (so-called tranches) of the debt-issue amount to a lower average yield at issuance?a cost reduction pocketed by the issuer bank.

The evidence also suggests that the two other rating agencies, Moody?s and Fitch were no less prone to rating favours towards their largest clients than was Standard & Poor?s. For all three rating agencies, we can reject the null hypothesis that their respective deal-level rating favours are independent from the overall rating business each agency obtained from a particular issuer.

In its motion to dismiss the lawsuit, Standard & Poor?s points out that one would need to know the ?true? credit risk of a security in order to show that certain ratings were ?objectively false? (McGraw-Hill Companies, Inc and Standard & Poor?s Financial Services LLC 2013). Such objectivity may not be available about any individual credit rating, but a rating bias can nevertheless be revealed at high levels of statistical certainty within a large sample of ratings. Importantly, the true credit risk of a security can be proxied by observable control variables. Our regression analysis controls for collateral quality, levels of credit enhancement and issuer fixed effects. Yet inclusion of these credit-risk determinants does not destroy the statistically and economically highly significant correlation between rating favours by a particular rating agency and its securitisation business with the issuer.

Could reverse causality be at work?

Correlation evidence typically has alternative interpretations. Rating errors can also be made in good faith and banks are likely to privilege rating agencies which err in their favour and provide them with more business. Could this explain the observed correlation? In the sample, large issuers account for dozens of structured deals and only if a rating agency makes repeated rating mistakes in the same direction about the same issuer should we expect to see the correlation. In other words: The rating error of each agency must have a substantial common and directional component with respect to a specific issuer bank in order to invoke reverse causality as a plausible explanation. However, such error clustering at the level of the agency-issuer relationship is difficult to reconcile with rating errors made in good faith as these should be independent across deals and not have a common bias in the same direction.

We report a strong and highly significant Pearson (Spearman) correlation of 0.175 (0.187) between the rating errors of deals sold by issuers that provide the rating agencies with substantial securitisation business. By contrast, Low Value agency-issuer relationships do not generate a systematic rating-error correlation. For small issuers the null hypothesis of a random non-directional rating error produced in good faith cannot be rejected; it can be rejected for large issuer clients.

While a common error component (specific to deals of High Value clients) is unlikely to occur in good faith for similar deals, it is even less plausible across deals which differ in the underlying asset classes, because ratings are produced on the basis of different historical default data and methodologies. It is therefore instructive to examine separately the error correlation calculated for heterogeneous deals pairs within the same agency-issuer relationship. For High-Value agency-issuer relationships, we find a large and significant Pearson (Spearman) correlation of 0.379 (0.339) between deals that belong to different asset classes. Again such error clustering does not extend to heterogeneous deals of Small Value relationships. The strong error correlation across different collateral pools and asset classes (only for large issuers) is difficult to attribute to reverse causality or an omitted variable specific to a particular asset class.

What could be done?

The lawsuit against Standard & Poor?s highlights the conflicts of interest inherent in the rating business, but can do little to resolve them. If new and complex regulation and supervision of rating agencies provides a remedy is unclear and remains to be seen. But three alternative policy measures could make the existing conflicts much less pernicious:

l Similar to US bank regulation under the Dodd-Frank act, Basel III should abandon (or at least decrease) its reliance on rating agencies for the determination of bank capital requirements.

l As forcefully argued by Admati, DeMarzo, Hellwig and Pfleiderer (2011), much larger levels of bank equity as required under Basel III could reduce excessive risk-taking incentives and ensure that future failures in bank-asset allocation do not trigger another banking crisis.

l More bank transparency in the form of a full disclosure of all bank asset holdings at the security level would create more informative market prices for bank equity and debt, with positive feedback effects on the quality of bank governance and bank supervision.

Our reliance on bank ratings could thus be greatly reduced.

This is an abridged version of the column. The full article can be found at http://www.VoxEU.org