A New Approach to Sovereign Ratings

The debt ceiling debacle has rejuvenated the discussion about the adequacy and effectiveness of sovereign credit ratings, as they're currently being provided.

We remain agnostic on the value added of a US downgrade today – does anybody really benefit from a downgrade at this stage? The objective of a downgrade watch notification, or a downgrade itself, is to encourage governmental concern, which precipitates action. We already have international concern. A downgrade at this stage seems only to cause further turmoil.

We are additionally skeptical of the adequacy of sovereign ratings being provided today: we wonder at the raters' ability to determine what they refer to as “willingness to pay.” Are they any better than anybody else at estimating this highly subjective measure? And if so, why is it they cannot provide to users the separate outcomes of each of these two components of the credit rating - the ability to pay and the willingness?** (Note also distinguished researcher Arturo Cifuentes’ timely suggestion of a third component: permission to pay.)

Sadly it seems a crisis has to occur before we resolve flawed systems. Since we believe sovereign ratings ought to be more consistently applied and more transparent, if worthwhile at all, we are happy to host guest author Marc Joffe's posts. We hope his views continue to encourage meaningful debate around the importance and implementation of sovereign ratings, and we welcome your comments.

A New Approach to Sovereign Ratings

By Marc Joffe*

A couple of my friends in the Moody’s diaspora have argued that rating agencies should not assign sovereign ratings due to difficulties in managing conflicts. I disagree for a couple of reasons. First, to the best of my knowledge, sovereign ratings have performed fairly well over the past few decades. While rating changes could have been faster, I have not seen evidence of systematic bias – except perhaps in the maintenance of the home country AAA rating.

Second, sovereign ratings provide a type of government accountability not unlike that offered by an independent press. For this reason, I am actually sympathetic to the idea that ratings warrant First Amendment protection. The best sovereign rating analysis is politically aware without being politically opinionated. As S&P and Moody’s have repeatedly stated during the current debt ceiling debate, they have no view about which fiscal measures should be employed to adjust the nation’s fiscal trajectory, they just believe that inaction or minimal action is no longer consistent with the highest rating.

Although the prevailing sovereign rating methodology may not actually display bias, it is certainly being subjected to accusations of bias, especially in Europe. Also, the present qualitative approach necessitates delays in rating changes unless the rating group is heavily staffed. In last week’s Congressional testimony, S&P President Deven Sharma said that his agency’s sovereign group consists of roughly 100 analysts rating 126 countries – less than 1 full time equivalent per issuer.

Thus, the biggest opportunity for improvement in sovereign rating performance is the application of transparent, quantitative modeling technology fueled by frequently updated data for exogenous variables. Model driven ratings can be calculated daily, weekly or monthly and then reviewed by analysts prior to release. And models don’t worry about being criticized for a lack of patriotism or for affecting interest rates.

Quantitative credit assessment techniques have been successfully applied to consumers, public firms and private firms. Bloomberg and Morningstar have already implemented ratings based on Merton-style public firm models first popularized by KMV Corporation, while RapidRatings uses financial statement data to automatically rate both public and private firms.[1]

Sovereign risk modeling is newer and less developed. The most interesting work I have seen in this area has come from Nouriel Roubini and colleagues, from Dale Gray and colleagues, and from Kamakura Corporation (full disclosure: Kamakura is a client and also has a public firm model). These efforts are generally focused on emerging market issuers, since that is where just about all recent default observations are available. Data for advanced economy sovereign defaults (or, more precisely, defaults by nations that went on to become advanced economies) is older, hard to gather and may require restatement to be meaningful in modern terms. Reinhart and Rogoff have assembled this data for their book and related IMF papers, lowering the data collection barrier.

I propose a different modeling approach for advanced economies that focuses on their primary risk factor: the impact of population aging on social insurance spending. The approach leverages the wealth of budget, economic and demographic data and forecasts available for these countries. While the remainder of the discussion focuses on the US, it should be equally applicable to major European economies.

In the US, the Congressional Budget Office and other government agencies (including OMB and GAO), periodically issue long term budget forecasts. Typical lengths of these forecasts are 10 and 75 years. For Treasury investors, the longest relevant time horizon is 30 years, which also happens to be the length of the longest term macroeconomic forecasts provided by IHS Global Insight and Moody’s Analytics.

The CBO forecasts include projected Debt-to-GDP ratios. It is also possible to derive Interest Expense to Revenue ratios from the CBO outlooks. This latter ratio may be more predictive of default than Debt-to-GDP because it embeds information about the government’s ability to tax economic activity and the level of its Treasury interest rates. (This ratio effectively ignores principal repayment schedules – an exclusion that could be justified by the assumption of continued liquidity and absence of rollover risk for advanced economy sovereign debt.)

Consequently, the CBO data provides expectations for key ratios at the maturity date of long term Treasuries. These expectations are based on policy and macroeconomic forecasts. If different policy and macroeconomic parameters are used, different projected ratios can be generated. If we provide a range of possible scenarios to a Monte Carlo simulation engine, we could generate a distribution of ratio outcomes.

Next, we can use historical data to identify ratios that are associated with default. A nice property of interest expense to government revenue is that, with rare and extremely idiosyncratic exceptions, it is always in the range of zero (absolute certainty of non-default) to 100% (absolute certainty of default). This relationship prevails across all countries and through time. For purposes of this discussion, let’s assume that an interest expense to revenue ratio of 30% is determined to be a reasonable default point. This would mean that a government would be unwilling to pay interest beyond this threshold, because the political pain associated with further crowding out other kinds of spending exceeds that stemming from a default. Admittedly limited post-Reconstruction experience in the US suggests that the critical value of this ratio, i.e. the default point, is 30% (see my earlier blog post entitled Correction: The US Has Defaulted Before and it Can Default Again).

With a distribution of ratio realizations and critical point both in hand, rating the issuer is then simply a matter of calculating the proportion of the distribution beyond the default point. If this proportion is very low (perhaps 0.25% for a thirty year Treasury), the issuer is AAA. If the proportion of “default-indicative” realizations is higher, then a lower rating is appropriate.

What would such a model conclude about the US? Since the model only exists in the form of the rough outline above, I can’t be certain – but I have a pretty good idea. In April, I calculated a projected interest expense to revenue ratio based on an adjusted version of the June 2010 CBO Long Term outlook. The adjustments mostly reflected some inputs from CBO’s more recent March 2011 10-year forecast. This calculation yielded a ratio of 37.82% in FY 2041. (Although the debt ceiling deal lowers this figure somewhat, CBO may also have to make an offsetting adjustment due to the disappointing GDP numbers published last week). Assuming that the 30% ratio is indeed the default point, the implied default probability is quite substantial.

This rough calculation is based on a number of assumptions, most important of which is that the CBO forecast provides a reasonable expected value. While CBO’s economic forecasts are well within the mainstream, CBO more controversially assumes no substantive policy change. If major tax increases or entitlement reforms are implemented, the expected ratio could be far lower. But the recent debt ceiling debate has shown just how difficult major policy change is in an environment of divided government and party polarization (the concept of Congressional polarization can be quantified as Political Scientist Keith Poole regularly does at VoteView).

Contrary to what we hear from politicians and the media, I do not see much reason to expect this situation to be resolved by the 2012 election. The likelihood that divided government will continue can be estimated by consulting political markets, such as InTrade, which reflects the expectation that Obama will be President and that the Republicans will control the House and Senate in 2013. Also, the further we get into the cycle of baby boomer retirements, which started in 2008, the more difficult entitlement changes will become given the economic inflexibility and heightened voting participation of seniors.

One assumption in the CBO projections that could be questioned is the reversion of interest rates to modern historical averages over the next few years. If, instead, the US has entered a period of sustained low interest rates a la Japan, the terminal interest expense to revenue ratio would be far lower.

Regardless of which interest rate, policy or growth assumptions are used, the simulation model outlined above provides a formal way of evaluating the implications of a range of political and economic scenarios for sovereign creditors. Further, if rating agencies make the simulation model and their assumptions publicly available, investors could substitute their own exogenous variables and form their own conclusions. The benefit is that evaluations of US and other advanced sovereign credits become more rigorous and more transparent. Quantitative approaches do not guarantee objectivity because the choice of assumptions can itself be biased, but any bias and its effect on the rating becomes more apparent and much easier to address.

----------------------------------------------------
[1] Application of models to structured products has proven more controversial, but I would argue that the problems of 2007-2008 are not an indictment of quantitative assessment in general. Instead, the crisis is an indictment of the specific modeling procedures and assumptions employed. Just because models based on Gaussian Copulas failed to adequately weight tail risk does not mean that models relying on more empirically appropriate distributions will not work. And just because RMBS models failed to consider negative Housing Price Appreciation in 2006 is not evidence that models that included a proper set of HPA scenarios would not have been effective.

* Marc Joffe (joffemd@yahoo.com) is a consultant in the credit assessment field. He previously worked as a Senior Director at Moody’s Analytics. This article reflects his personal opinion of sovereign rating practice. Although previously employed by Moody’s Analytics, the author no longer works at Moody’s and, when he did work there, his area of professional responsibility was software development and data collection. He had no professional experience as a ratings analyst, and no knowledge of Moody’s ratings practices beyond what is in the public record.

** It is worthwhile to note that rating agency BMI, though not an SEC-licenced NRSRO, divulges each of the two components separately in its rating analysis.