John Kiff, Michael Kisser, and Liliana Schumacher
Credit ratings are often employed in fixed income portfolio composition and collateral acceptability guidelines, in bond covenants and other financial contracts, and various financial rules and regulations. Given the different needs of its various end-users, credit rating agencies (CRAs) have to strike a challenging balance between delivering stable and yet accurate credit ratings.
On a conceptual level, CRAs can assign ratings on either a point-in-time (PIT) or a through-the-cycle (TTC) basis. Loosely speaking, the PIT approach can be thought of as using current information when computing the default risk metrics that are mapped into ratings. Credit ratings assigned under the PIT approach should provide the most accurate estimate of future default probabilities and expected losses. On the other hand, the TTC approach is supposed to balance the need for accurate default estimates and the desire to achieve rating stability.
Altman and Rijken (2006) investigate the conflicts of interests arising from the CRAs’ often competing objectives of the two approaches. Using credit-scoring models, they show that CRAs focus on a permanent credit risk component when assigning ratings. Besides, they argue that CRAs are slow in adjusting their ratings and that the slow reaction is the most important source of rating stability. Topp and Perl (2010) investigate actual corporate ratings assigned by Standard & Poor’s (S&P) and show that even though the CRAs claim to only focus on a permanent risk component, actual ratings reveal cyclical patterns.
Evidence from the CRAs themselves is consistent with the above findings. In a special comment to Moody’s rating users, Cantor and Mann (2006) analyze the trade-off between ratings accuracy and stability and argue that CRAs desire to deliver both accurate and stable ratings. Also, S&P’s claim that “when assigning and monitoring ratings, we consider whether we believe an issuer or security has a high likelihood of experiencing unusually large adverse changes in credit quality under conditions of moderate stress. To promote rating comparability, we use hypothetical stress scenarios as benchmarks for calibrating our criteria across different sectors and over time.”1
This article investigates the stability and accuracy of credit ratings within a stochastic framework (Kiff, Kisser, and Schumacher, 2013). Specifically, it first employs contingent claims analysis to simulate asset values, which are subject to both transitory and cyclical shocks. In the contingent claims analysis framework, a sovereign defaults when the value of its assets falls through a distress threshold that is related to its liabilities (Gray, Merton, and Bodie, 2007).
The simple model used here is based on Loeffler (2004) and assumes that asset values are driven by (1) the sovereign’s fundamentals and (2) cyclical factor fluctuations. Conceptually, the PIT rating process involves estimating the difference between future values of the assets and liabilities (“distance to default”), and mapping this difference into a default-probability-related credit rating. A TTC rating process estimates the distance to default based on fundamental values but imposes a stress scenario on the cyclical component.
In a second stage, the CRAs typically apply a smoothing rule to rating changes to avoid overshooting or subsequent reversals. In other words, it is a two-step process in which ex-ante ratings are based on fundamentals and a stress scenario and ex-post rating changes are smoothed, and not adjusted immediately. More specifically, the factor that represents the sovereign’s “fundamentals” is assumed to follow a random walk, whereas the cyclical component is assumed to follow an autoregressive process.
Figure 1 shows the distribution of actual S&P sovereign ratings. It can be seen that while most sovereigns receive an investment-grade rating, i.e., a minimum rating of BBB, the largest single fraction of sovereigns are rated B, that is below investment grade.
Figure 1:Empirical Rating Grade Distribution for Sovereigns as Rated by Standard & Poor’s
This figure displays the distribution of sovereign ratings as of March 2012. Specifically, ratings correspond to foreign currency ratings by Standard & Poor’s. Note: All figures in this article are from Kiff, Kisser, and Schumacher (2013).
Figure 2 shows the rating distribution as implied by the TTC approach. That is, we first stress test each asset, compute the corresponding distance-to-default and map this continuous measure into discrete ratings using Moody’s 5-year idealized default probabilities. Figure 2 shows that the distribution implied by the TTC approach is similar to the empirical rating distribution displayed in Figure 1 and thus provides assurance regarding the choice of the parameter values.
Figure 2:Model Implied Rating Distribution for TTC Approach
This figure displays the distribution of sovereign ratings as of March 2012. Specifically, ratings correspond to foreign currency ratings by Standard & Poor’s.
The main interest of this analysis lies in how ratings evolve over time and how well the two approaches predict future defaults. It therefore assumes that future asset values do not evolve according to their expected values but instead come in well below. While the PIT approach would imply immediate downgrades for this case, a CRA following the TTC approach would typically wait to see if the deviation is only of a cyclical nature.
For example, in the case below, downgrades are assumed to occur only if (1) the rating change is expected to be persistent and (2) the implied change is larger than one notch. This is one of several smoothing rules discussed in Cantor and Mann (2006), which also accounts for the empirically documented fact that CRAs are slow in adjusting their ratings (Loeffler, 2005).
Figure 3 visualizes rating downgrades under the PIT and smoothed TTC methodologies and compares them to the case in which a CRA switches from a TTC to a PIT rating method once the initial stress scenario is breached (“unsmoothed TTC”). One can see that ratings decline faster under the PIT approach whereas a downgrade is less likely if the CRA followed a TTC approach. The intuitive reason is that TTC ratings build in a pessimistic forecast so the rating is already lower and does not have to fall as much as the more “optimistic” PIT ratings would imply.
Figure 3:Example of Rating Downgrades
However, as time passes the PIT rating would eventually drop below the smoothed TTC rating (Period 3), which is precisely the point when the smoothed TTC approach becomes prone to potential cliff effects. By not reacting to new information in Periods 3 and 4, the TTC ratings would drop from BB to CCC in Period 5, thereby generating a rating downgrade of four notches. From a stability perspective it would therefore be optimal if a CRA followed the TTC approach ex ante but would immediately adjust the rating once the initial forecast has been breached.
Finally, the analysis looks at how well both approaches predict future defaults by computing the cumulative accuracy profile (CAP) for defaults taking place at the end of Periods 1 and 2. “Ideal” CAP curves look almost like vertical lines because all the defaulters should be among the lowest rated issuers. The closer the CAP curve to the ideal curve, the better the discriminatory power of that CRA’s ratings. Figure 4 shows that initially the TTC approach is only slightly less accurate at forecasting future defaults.
Figure 4:Cumulative Accuracy Profile for Defaults at the End of Period 1
However, as time passes the PIT approach becomes clearly more accurate (see Figure 5) as it immediately incorporates new information into its ratings whereas the TTC approach only reacts with a lag owing to its smoothing policy.
In summary, the experiment has shown that, from an ex-ante viewpoint, the TTC approach produces more stable and only slightly less accurate ratings when current net asset values are higher than in the stress scenario. However, once ratings drop below those implied by the stress scenario, the smoothed TTC approach is less accurate at predicting defaults and it runs the risk of generating rating cliff effects that may lead to dangerous second-round liquidity effects.
Figure 5:Cumulative Accuracy Profile for Defaults at the End of Period 2
Current discussions on the usefulness of the TTC approach should therefore focus on the reaction to new information when net asset values drop below those implied by the initial stress scenario. The implementation of a “through-the-crisis” methodology, which has been mentioned by the CRAs themselves, seems to require a more severe stress test ex ante. However, it currently does not address the slow adjustment typically taking place once the cushion built in by a TTC methodology is eroded, and nor does it address the potential for cliff effects created by smoothing policies.
AdelsonMarkGailI. HessolFrancisParisi and ColleenWoodell2010 “Methodology: Credit Stability Criteria.” Standard and Poor’s RatingsDirect.
AltmanEdward I. and HerbertA. Rijken2006 “A Point-In-Time Perspective on Through-The-Cycle Ratings.” Financial Analyst Journal Vol. 62 No 1. (Jan.-Feb. 2006) pp. 54–70.
CantorRichard and ChrisMann2006 “Analyzing the Tradeoff Between Ratings Accuracy and Stability.” Special Comment Moody’s Investors ServiceSeptember2006.
GrayDale F.Robert C. Merton and BodieZvi2007 “Contingent Claims Approach to Measuring and Managing Sovereign Credit Risk.” Journal of Investment Management Vol. 5 No. 4 pp. 5–28.
KiffJohnMichaelKisser and LilianaSchumacher2013 “Rating Through-The-Cycle: What Does the Concept Imply for Rating Stability and Accuracy?” IMF Working Paper 13/64 (Washington: International Monetary Fund).
LoefflerGunter2004 “An Anatomy of Rating Through the Cycle.” Journal of Banking and Finance Vol. 28 No. 3 pp. 695–720.
LoefflerGunter2005 “Avoiding the Rating Bounce: Why Rating Agencies Are Slow to React to New Information.” Journal of Economic Behavior and Organization Vol. 56 No. 3 pp. 365–381
Adelson, Hessol, Parisi and Woodell (2010) in Standard & Poor’s, Global Credit Portal. Ratings Direct. Page 2