Ratings are up, spreads are down. Chart 1 plots cumulative upgrades for Moody’s and S&P for Emerging Markets (EM’s) and the EMBI global illustrating the dramatic increase in emerging market credit quality the reduction in emerging market spreads. In this blog I want to address 3 issues:
Chart 1: Cumulative Upgrades and The EMBI Global
i) what can we learn explaining ratings movements?
ii) is the change in spreads commensurate with that in ratings ?
iii) do ratings matter ? This note is based on a recent paper written with Juan Francisco Martinez Sepulveda, also in the Research Department, at the IDB. Explaining ratings, what can we learn? There have been several recent papers on ratings and in fact it is not difficult to explain ratings across countries, and although perhaps slightly more difficult, also over time. Panel regressions with only a few “country fundamentals” do a pretty good job – in fact probably too good, with R-Squared’s suspiciously high! On closer inspection some models include the current account deficit with the “wrong sign” i.e.: a wider deficit is associated with a better rating, smacking of the reverse causality; that countries with better ratings may finance a deficit more easily. Others include Kaufmann’s Government Effectiveness variable. But is this survey of surveys really independent of something so basic as the sovereign rating? Surely respondents to the various surveys included in the construction of the index may have been influenced by the “opinion” of a major rating agency? As an aside Government Effectiveness is highly correlated not only with S&P and Moody’s ratings but also with the other indices in the same family (such as Voice and Accountability, Control of Corruption, Political Stability, Rule of Law) and others such as Business Competitiveness, Growth Competitive Index Score or Ranking etc. Various combinations of these indices can be added to the regression to get the R-Squared up if the researcher so chooses. Still, even if we stick to harder economic data the models explain a reasonable percentage of the variation in ratings over time. There is some evidence that additional variables to the standard ones also matter. Debt composition is one example. We re-compute the Hausmann and Panizza Original Sin 3 variable and also debt composition variables recently available thanks to the IDB’s Living with Debt publication and the IMF’s new database on debt composition. The data does not cover many countries so treat this as a call for better homogenous data on currency composition of (even public sector) debt! Interestingly though, even in a model with just a few harder economic variables and fixed country effects, global financial variables such as the US interest rate, the VIX or the US high yield spread appear to play no role in explaining ratings changes. So far so good but let me quickly throw a large spanner in the works. A simple Principal Component (PC) analysis of the changes in ratings of emerging countries shows that 3 PC’s explain about 90% of the variation in ratings. A similar analysis of 7 economic fundamentals per country (for about 30 countries, so about 210 variables), shows that 3 global factors explain almost 70% of the variation! If we add them together so we include 210 country fundamentals plus 30 ratings, about 240 variables in total, then 3 PC’s explain almost 80% of the relevant variation. We are still working to try to interpret these “global factors” but the message is pretty clear; while “economic fundamentals” are important, they appear to be very largely endogenous to developments in the global economy. Explaining Spreads Now let me turn to the second question, are the recent reductions in spreads that we have seen commensurate with the improvements in ratings? However one chooses to analyze this question the answer is a resounding, no. Spreads are significantly lower than would be expected even considering today’s high ratings given the past relationship between ratings and spreads. This is true on an individual country basis, on a pooled basis, using linear panel techniques, using log scales, using various non linear techniques (as the change from a BB to a BB+ may not have the same implications as a change from an A- to an A) etc. The reduction in spreads can be explained however by including financial variables such as the US interest rate, the VIX and/or the US high yield – we resist putting in the global EMBI across all countries as this may be endogenous to a change in the spread of one large emerging economy (even if we left out that economy in the definition of the EMBI in the relevant row of the panel), but we think using the US high yield is just about ok. Interestingly, a principal component analysis of the US interest rate, the VIX and the US High Yield reveals that two Principal Components explain 90% of the variation of these 3 variables with one PC essentially explaining the US interest rate and a second explaining both the VIX and the High Yield. It is tempting to suggest that this latter PC represents risk aversion and the former, lets call it the global supply of funds. It is the risk-aversion-PC (that explains the VIX and the High Yield) that is important in explaining the divergence between emerging market spreads and ratings that has emerged over the last few years. The global supply of funds PC (that explains mostly US rates) is not significant. Simulations indicate that without the reduction in the VIX or the High Yield spreads, EMBI spreads would be some 153 (for S&P ratings) and 176 (for Moody’s ratings) basis points higher on average across the EM’s in the EMBI. Do Ratings Matter? Finally, let’s consider the question as to whether ratings actually matter? Obviously in some trivial sense they do; a regression of spreads on ratings reveals ratings to be highly significant. But the question that should be posed is, do ratings matter over and above (i.e.: controlling for) country fundamentals? The agencies may follow the market in that they may delay decisions to see if changes are temporary or more permanent but here we are interested in the opposite causality, whether a change in ratings may (also) affect the spread controlling for any changes in fundamentals. There are several ways that one may attempt to do this. Barry Eichengreen and Ashok Mody in an older paper (1998) regress ratings on economic variables and then take the residual and introduce that in an equation for spreads including the economic fundamentals and this “unexplained” component of the rating. These results are easily replicated and the residual is highly significant. One interpretation is that this residual is the rating agency “opinion” over and above what one can explain of the rating with harder economic data. However a criticism is that if this is so, how can it conform to the standard properties of an error to ensure the first regression is valid? It is simply an act of faith to state that this error is the agency’s opinion appropriately measured. A second technique is to run a system of equations, one for the rating and another for the spread. Again this reveals that ratings are significant determinants for spreads controlling for fundamentals but the methodology makes some strong assumptions to gain identification. Interestingly, the economic fundamental that is required here to explain spreads (in addition to the rating) is growth. A different sort of analysis would be an event study. But in this context such a venture would appear problematic. Rating agencies do their best to make rating changes predictable. They announce positive or negative outlooks, they announce credit watch periods, they even say what would have to happen for an upgrade to be made (Peru must increase tax revenues for example). This hardly conforms to the classic “exogenous event” used in corporate finance event study literature (merger announcement, stock split etc). Indeed it is not obvious what the “market model” should be for the spread as again it would of course be affected by the fundamentals – that the rating agency may have deemed must change for the upgrade to occur. In short, it doesn’t look a promising technique. Another route is to exploit differences in opinions. It turns out that rating agencies, Moody’s and S&P specifically, disagree about 50% of the time. Using standard mappings between the ratings of these two agencies they are the same about 50% of the time and differ the rest. So, one can ask the question, do spreads respond to a rating change by one agency when the rating of the other agency remains the same. The answer is that the spread does change. Again this is not perfect but it does seem to be another result that suggests that the while the agencies may follow the market, rating changes may also have some impact. Putting it all together So, what does it all mean? On the one hand ratings appear to be determined by country economic variables, and financial ones such as the VIX, High Yield and US rates are not significant. Rating agencies may take heart in this result that suggests that their indicated credit quality responds to fundamentals rather than changes in risk aversion or liquidity. The agencies may also take some comfort in our results suggesting that their ratings do indeed appear to matter. However, rating changes and the economic fundamentals that explain rating changes are largely determined by a few “global factors” – presumably real in nature. This suggests that much of the improvement in EM credit quality is endogenous to developments in the world economy. Whether this is temporary or permanent then depends on whether one thinks this remarkable period of world growth and stability is here to stay. Second, spreads have come down more than ratings have improved. An optimistic interpretation is that this derives directly from the fact that we can now splice up EM risk more effectively and there are new investors taking different chunks. This smacks of a permanent change in the technology of risk transfer that allows the world to better exploit “comparative advantages” in risk taking – to use a phrase coined by Don Lessard some 30 years ago now as he lamented the lack of risk sharing in debt contracts for EM’s. The less optimistic interpretation is that there is a “global credit Tsunami” washing over all risks, driven by the competitive lending decisions of investors desperate for returns and creating a huge bandwagon. In a Tsunami, the true extent of the damage is only revealed when the tide recedes. The high tide washing over all risk makes everything look great and it has brought lots of beautiful multicolored fish to marvel at. We can only speculate what it would look like if that tide receded. It might be argued that our results are exactly what you would expect and would perhaps even desire. On the one hand, ratings appear to respond more to real developments whereas spreads reflect both ratings and financial variables. Rating agencies should be cautious and should delay in giving upgrades until improvements are “consolidated”. This implies we should really expect them to “follow the market” – at least to follow those market moves that turn out to be persistent. Note that Moody’s appears to be rather more cautious than S&P assuming the cumulative upgrades plotted in Chart 1 is an appropriate measure. An open question I will leave you with is whether, if EM credit quality is largely endogenous to a set of small world factors, whether this is being signaled appropriately in individual rating decisions.