PrefBlog

Manfred Gärtner and Björn Griesbach have published a paper titled Rating agencies, self-fulfilling prophecy and multiple equilibria? An empirical model of the European sovereign debt crisis 2009-2011. The introduction for the paper was reproduced badly on the link provided; there’s another version on Scribd:

We explore whether experiences during Europe’s sovereign debt crisis support the notion that governments faced scenarios of self-fulfilling prophecy and multiple equilibria. To this end, we provide estimates of the effect of interest rates and other macroeconomic variables on sovereign debt ratings, and estimates of how ratings bear on interest rates. We detect a nonlinear effect of ratings on interest rates which is strong enough to generate multiple equilibria. The good equilibrium is stable, ratings are excellent and interest rates are low. A second unstable equilibrium marks a threshold beyond which the country falls into an insolvency trap from which it may only escape by exogenous intervention. Coefficient estimates suggest that countries should stay well within the A section of the rating scale in order to remain reasonably safe from being driven into eventual default.

The literature review shows some controversy:

Among the first to put rating agencies into the game, in the sense that ratings might have an influence on outcomes if multiple sunspot equilibria exist, were Kaminsky & Schmukler (2002). In a panel regression they show that sovereign debt ratings do not only affect the bond market but also spill over into the stock market. This effect is stronger during crises, which could be explained by the presence of multiple equilibria. As a consequence they claim that rating agencies contribute to the instability in emerging financial markets. Carlson & Hale (2005) argue that if rating agencies are present, multiple equilibria emerge in a market in which otherwise only one equilibrium would exist. The purely theoretical paper is an application of global game theory and features heterogeneous investors. Boot, Milbourn & Schmeits (2006) arrive at the opposite conclusion: ratings serve as a coordination mechanism in situations where multiple equilibria loom. Using a rational-herd argument, they show that if enough agents base their investment decisions on ratings, others rationally follow. Since ratings have economic consequences, they emphasize that the role of rating agencies is probably far greater than that of the self-proclaimed messenger.

“Multiple sunspot equilibria”? I had to look that one up:

‘Sunspots’ is short-hand for ‘the extrinsic random variable’ (or ‘extrinsic randomizing device’) upon which agents coordinate their decisions. In a proper sunspot equilibrium, the allocation of resources depends in a non-trivial way on sunspots. In this case, we say that sunspots matter; otherwise, sunspots do not matter. Sunspot equilibrium was introduced by Cass and Shell; see Shell (1977) and Cass and Shell (1982, 1983). Sunspot models are complete rational-expectations, general-equilibrium models that offer an explanation of excess volatility.

The authors regress a nine-factor model:

• Rating
• GDP Growth
• GDP per capital
• Budget Surplus
• Primary Surplus
• Debt Ratio
• Inflation
• Bond Yield
• Credit Spread

These indicators explain 60 percent of the variance of sovereign bond ratings in our panel. All estimated coefficients possess the expected signs, though not all are significantly different from zero. Ratings are found to improve with higher income growth and income levels, or with better overall and primary budget situations. Ratings deteriorate when the debt ratio, inflation or government bond yields go up.
…
Applying the test proposed in Davies (1987), the null hypothesis of no break was rejected, and the break point was found to lie between a BBB+ and a BBB rating.23 Regression estimates for the resulting two segments are shown as regressions 2a and 2b in Table 3. The differences between the two segments are striking. The slope coefficients differ by a ratio of ten to one. While, on average, a rating downgrade by one notch raises interest rates by 0.3 percentage points when ratings are in the range between AAA and A-, which comprises seven categories, a downgrade by one step raises the interest rate by 3.12 percent once the rating has fallen into the B segment or below.

That makes sense, at least qualitatively – default probabilities are not linear by notch, according to the agencies.

Now they get to the really controversial part:

This means that at sovereign debt ratings outside the A-segment, i.e. of BBB+ or worse, a downgrade generates an increase in the interest rate that justies or more than justies the initial downgrade, and may trigger a spiral of successive and eventually disastrous downgrades. Only countries in the A-segment of the rating scale appear to be safe from this, at least when the shocks to which they are exposed are only small. However, this only applies when marginal rating shocks occur. Larger shocks, and these have not been the exceptions during Europe’s sovereign debt crisis, may even jeopardize countries which were in secure A territory. We may illustrate this by looking at the impulse responses of equation (11) to shocks of various kinds and magnitudes. This provides us with insolvency thresholds that identify the size of a rating downgrade required to destabilize the public finances of countries with a given sovereign debt rating.
…
When rating shocks last, however, as has apparently been the case for the eurozone’s PIGS members, much smaller unsubstantiated rating changes may play havoc with government bond markets and suce to run initially healthy countries into trouble, as shown in Figure 6(b). In this scenario, an arbitrary, yet persistent, downgrade by two notches would trigger a downward spiral in a country with an AA rating. Rising interest rates would call for further downgrades, which would appear to justify the initial downgrade as an apparently good forecast.

And then they get to the real meat:

A more detailed look at the dynamics of the effect of debt rating downgrades on interest rates revealed that at least for countries with sovereign debt ratings outside the A range even erroneous, arbitrary or abusive rating downgrades may easily generate the very conditions that do actually justify the rating. Combined with earlier evidence that many of the rating downgrades of the eurozone’s peripheral countries appeared arbitrary and could not be justified on the basis of rating algorithms that explain the ratings of other countries or ratings before 2009, this result is highly discomforting. It urges governments to take a long overdue close look at financial markets in general, and at sovereign bond markets in particular, and at the motivations, dependencies and conflicts of interest of key players in these markets.

This paper has S&P’s shorts in a knot, and they have indignantly replied with a paper by Moritz Kraemer titled S&P’s Ratings Are Not “Self-Fulfilling Prophecies”:

In questioning the agencies’ integrity, the authors appear to suggest that the agencies follow some hidden agenda that has led them to act “abusively”. As is usually the case with conspiracy theories, little by way of evidence or persuasive rationale is offered to explain who benefits from the agencies’ supposed “strategic” or “disastrous” downgrades. Alas, the reality is not nearly as spectacular: rating agencies take their decisions based on their published criteria and are answerable to regulators if they fail to do so.

The authors also claim that the agencies’ rating actions “cannot be justified” because they do not accord with a mechanistic “ratings algorithm” of the authors’ own devising. Apart from the fact that ratings are subjective opinions as to possible future creditworthiness (and, like other opinions, neither “right” nor “wrong”), the authors fail to justify why their algorithm has more merit than the published comprehensive methodologies of the rating agencies. Nevertheless, so persuaded are the authors of their own algorithm they admonish the agencies for “manipulating the market by deviating” from the authors’ “correct rating algorithm”!

Standard & Poor’s, for one, long ago rejected an algorithmic approach to sovereign ratings as simplistic and unable to account for the subtleties of a sovereign’s political and institutional behavior.

Even more seriously:

At the heart of the paper’s confusion is its treatment of causality and correlation. The paper suggests that investors react to rating changes by asking for higher interest rates when a rating is lowered, but provide no evidence for their claim. In fact, the authors probably cannot provide such evidence as their data set has merely an annual observation frequency. To show causality, the paper should present data that played out during a period bounded by at least two yearly observation points. With such limited data, one cannot determine what came first: rating action or interest movement, or, indeed, whether one caused a change in the other at all!
…
The suggestion that in Europe’s financial crisis, the underlying pattern was one of ratings causality is effectively contradicted by the fact that spreads did not react for several years to our downgrades (starting in 2004) of several eurozone periphery countries.
…
Until early 2009, the CDS-market traded swaps on Portugal as though it were a ‘AAA’ credit (i.e. four notches above our rating at the time). When sentiment changed rapidly, the market “downgraded” Portugal to around ‘B’ in 2010, a full eight notches below the S&P rating at the time. Suggesting that the relatively modest rating changes had caused this massive sell-off appears far-fetched.

And then they get downright nasty:

We note that under the paper’s algorithm Greece should have been downgraded by a mere 0.15 notches between 2009 and 2011. In our view, the algorithm therefore would have entirely missed the Greek default in early 2012, the largest sovereign restructuring in financial history. By contrast, far from having acted in an “arbitrary or abusive” manner, Standard & Poor’s anticipated Greece’s default well before it occurred.