International Review of Financial Analysis 21 (2012) 45–55

Contents lists available at SciVerse ScienceDirect

International Review of Financial Analysis

Rating agencies' credit signals: An analysis of sovereign watch and outlook Rasha Alsakka a,⁎, Owain ap Gwilym b a b

Bangor Business School, Bangor University, Bangor, LL57 2DG, UK Bangor Business School, Bangor University, UK

a r t i c l e

i n f o

Article history: Received 25 June 2010 Received in revised form 13 October 2011 Accepted 30 October 2011 Available online 4 November 2011 JEL classification: G1 G2

a b s t r a c t We analyse sovereign watch and outlook signals from Moody's, S&P and Fitch. Prior literature shows strong market reactions to these signals, which arguably contain more new information than rating changes. We show that the agencies' actions imply different policies: S&P has more emphasis on short-term accuracy, while Moody's actions are consistent with greater stability. We find evidence of momentum in negative (not positive) outlook signals, but no watch momentum. We also examine the lead–lag relationships, finding that S&P (Fitch) demonstrates the least (most) links with other agencies' actions. Moody's tends to be the first mover for positive outlook and watch signals. © 2011 Elsevier Inc. All rights reserved.

Keywords: Credit rating agencies Sovereign outlook Sovereign watch Lead–lag relationship Momentum

1. Introduction The sub-prime mortgage crisis in the United States placed credit ratings agencies (CRAs) under the spotlight, and brought increased attention to their performance. There is an ongoing debate on issues of revenue versus reputation (e.g. Mathis, McAndrews, & Rochet, 2009). However, the U.S. Securities and Exchange Commission (SEC, 2011) finds ‘no material regulatory deficiency’ based on its recent examinations of the ten registered Nationally Recognized Statistical Rating Organizations (NRSROs), despite ongoing concerns about whether CRA policies are entirely adequate to avoid conflicts of interest. 1 Meanwhile, criticism of CRAs during the European sovereign debt crisis was more focused on the extent and timing of downgrades. In response to the perceived role of CRAs in the sub-prime crisis, the International Organization of Securities Commissions (IOSCO) revised the Code of Conduct Fundamentals for CRAs in 2008 to address issues of independence, conflict of interest, transparency and competition. A formal European Union (EU) regulation on CRAs entered into force in December 2009, and CRAs are now subject to legally binding rules based on the IOSCO Code. Within the EU, the responsibility for the registration and regulation of CRAs was handed to the European

⁎ Corresponding author. Tel.: + 44 1248 383571. E-mail addresses: [email protected] (R. Alsakka), [email protected] (O. ap Gwilym). 1 The ten registered NRSROs are A.M. Best, DBRS, Egan-Jones, Fitch, JCR, Kroll, Moody's, Morningstar, R&I and S&P (at end-2010). 1057-5219/$ – see front matter © 2011 Elsevier Inc. All rights reserved. doi:10.1016/j.irfa.2011.10.002

Securities and Markets Authority (ESMA) in July 2011. Ratings issued outside the EU can be used for regulatory purposes by regulated entities in the EU by means of either endorsement or certification with ESMA. The Basel Committee also reviewed the role of external ratings in its capital adequacy regulations, mainly to incorporate the IOSCO Code in the eligibility criteria. 2 CRAs aggregate information about the credit quality of borrowers, reducing information asymmetry faced by lenders, and hence allowing borrowers to access financial markets and attract investment funds. 3 Rating changes are the means by which CRAs signal permanent changes in an issuer's credit quality. However, CRAs' rating outlook and watch are supplemental tools to communicate potential changes in credit quality. Outlook and watch signals were developed to provide indicators of the likely direction and timing of future rating changes (Hamilton & Cantor, 2004). A complete CRA credit opinion on an issuer consists of a credit rating and a rating outlook/watch status. One criticism of CRAs is their apparently slow reactions in changing ratings. However, because of CRAs' “through the cycle” methodology and the sound reasons for stability in ratings (Altman & Rijken, 2006; Löffler, 2004, 2005), watch and outlook signals are 2 For more details about the Basel II capital framework, see Sundmacher and Ellis (2011), for example. Further, in 2009, the U.S. SEC amended its regulations for CRAs to require enhanced disclosure of performance statistics, rating methodologies and annual reporting, and additional restrictions on activities that could produce conflicts of interest. 3 Credit ratings are now heavily hardwired into investment processes, financial contracts and regulatory frameworks.

46

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

very likely to be the source where CRAs reveal more private information. Bannier and Hirsch (2010) analyse the economic function of the watchlist, and find that CRAs employ watch signals to improve the delivery of information. Several prior studies demonstrate that outlook and watch signals have a significant market impact. Hand, Holthausen, and Leftwich (1992) show that watch signals by Moody's and S&P (pooled together) for corporate issuers are associated with stronger abnormal bond and stock returns than are actual rating changes. Hull, Predescu, and White (2004) show that negative watch signals by Moody's contain information for the credit default swap (CDS) market, while rating downgrades do not. The average increase in the CDS spread at the time of a watch event is almost 10 basis points. Norden and Weber (2004) find that negative watch actions by Moody's and S&P for corporate issuers affect stock returns and CDS spreads while rating downgrades by S&P do not. Kaminsky and Schmukler (2002) illustrate that sovereign outlook and watch signals by the larger three CRAs have a stronger impact than rating changes for emerging stock and bond markets. Pukthuanthong-Le, Elayan, and Rose (2007) show that sovereign outlook and watch events by S&P influence bond and equity markets, while the effect of ratings changes is either insignificant or weaker. Hooper, Hume, and Kim (2008) find that the impact of sovereign outlook/watch changes by the larger three CRAs is twice as strong as the impact of rating changes. Sy (2004) finds that S&P and Moody's sovereign credit signals, including negative watch and outlook events, help predict the likelihood of distressed debt events within the next year. Hill and Faff (2010) highlight that sovereign outlook and watch events are more timely and more informative than rating changes. The International Monetary Fund (2010a) emphasises that CRAs affect stock and bond markets by revealing new information and a ‘certification’ role, though this is most evident in their use of outlook and watch signals rather than actual rating changes. Kim and Wu (2011) provide evidence that improvements in sovereign credit quality encourage international bank flows from developed to emerging economies, but note that outlook and watch events are associated with much stronger economic effects than are rating changes. Gande and Parsley (2005), Ferreira and Gama (2007) and Ismailescu and Kazemi (2010) find that the impact of sovereign outlook and watch signals is also transmitted to stock, bond and CDS markets in other countries.4 Given the economic importance of outlook and watch signals, we investigate the behaviour of sovereign outlook and watch status assigned by Moody's, S&P and Fitch. Specifically, we aim to answer four main questions: (i) Do previous sovereign outlook/watch events carry any predictive power for the direction of future sovereign outlook/watch changes?; (ii) Do the CRAs' polices differ in relation to outlook/watch?; (iii) Do sovereign outlook/watch changes by one CRA appear to be affected by prior actions by another CRA?; (iv) Does any one CRA demonstrate a lead in providing signals to the market through outlook/watch actions for sovereigns? Prior actual rating changes are demonstrated to carry predictive power for the direction of future rating migrations by the same CRA (rating momentum). Downgrade (but not upgrade) momentum in corporate ratings is supported by Bangia, Diebold, and Schuermann (2002) and Lando and Skødeberg (2002). Fuertes and Kalotychou (2007) and Alsakka and ap Gwilym (2009) provide evidence of downgrade momentum in sovereign ratings. However, the literature is silent on the existence of momentum in outlook and watch signals. Therefore, we examine whether outlook or watch status is affected by previous outlook or watch actions by the same CRA. We find evidence of momentum in negative (not positive) outlook actions, while watch

4

See Section 2.1 for further details about the importance of outlook and watch signals.

signals do not carry predictive power for the direction of future watch changes. Within the literature on the market impact of rating signals, there is evidence of unequal reactions to different CRAs' actions. Cantor and Packer (1996) find that Moody's sovereign rating changes have a greater effect on bond spreads than do S&P actions. Brooks, Faff, Hillier, and Hillier (2004) provide evidence that Moody's sovereign upgrades are associated with positive abnormal returns, but S&P and Fitch upgrades are not. Norden and Weber (2004) show that downgrades by Moody's only significantly impact CDS spreads, and negative watch actions by Moody's and S&P are associated with significant negative abnormal stock returns, while no abnormal performance is associated with Fitch actions. Hill and Faff (2010) highlight that S&P is more active and provides more new information than Moody's and Fitch during crisis periods. Outside crisis periods, Moody's tends to lead for ratings of advanced economies, and S&P leads for ratings of non-advanced economies. Each CRA has a clear interest in maintaining a strong reputation in financial markets by providing high quality credit signals (Güttler & Wahrenburg, 2007). Our lead– lag analysis aims to identify whether any given CRA demonstrates a lead in supplying credit signals to the market. Our evidence shows that different policies are applied across CRAs, whereby Moody's has more emphasis on stability, while S&P puts more weight on short-term accuracy. Our findings on lead–lag analysis are summarized as follows. S&P is the most independent CRA, while Fitch is the most dependent. Fitch watch and outlook actions have an insignificant impact on future outlook/watch adjustments by Moody's, but not vice versa. Moody's and Fitch tend to follow S&P negative outlook/watch actions to a greater extent than S&P follows the others. Moody's tends to be the first mover in positive outlook and positive watch signals. The remainder of the paper is organized as follows. Section 2 discusses key themes associated with the empirical analysis. Section 3 describes the data, while Section 4 presents the ordered probit models. Section 5 analyses the empirical results and Section 6 concludes the paper. 2. Key themes associated with the empirical analysis 2.1. The importance of outlook and watch signals A rating outlook is an opinion regarding the likely direction that a credit rating may take over the next one- to two-year period. The rating outlook categories are: positive, stable, negative and developing. Credit watch status is a much stronger statement about the future direction of a credit rating within a relatively short horizon (ex-ante target of 3 months). The watch categories are: watch for upgrade, watch for downgrade, and watch with direction uncertain. Watch assignments are formal rating reviews that are likely to result in some rating action (including confirmation of the existing rating). The CRAs' perspective is that an issuer which is on watch has a higher probability of experiencing a rating change than one with a rating outlook assigned. Rating outlooks and watch are designed to signal when risks are imbalanced but a rating change is not certain. Many rating changes are preceded by a non-stable outlook or a credit watch placement, but a positive or negative rating outlook/watch does not imply that a rating change is inevitable. Additionally, ratings with stable outlooks or which are not on watch are frequently changed before the outlook/watch status is revised (see Hamilton & Cantor, 2004; Klaar & Riley, 2005; Vazza, Leung, Alsati, & Katz, 2005). 5 Previous studies emphasise the economic importance of outlook and watch signals, since these signals offer important information 5 Outlook developing and watch with direction uncertain are a very small minority of the cases of outlook/watch status. As they do not signal a future rating direction, we exclude these cases in the empirical analysis.

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

content for equity, bond and CDS markets (see Section 1). In comparison, actual rating changes are, to some extent, anticipated since investors are aware of the prior rating outlook/watch status. Outlook and watch signals also help mitigate the tension between stability and accuracy, the two targets of a credit rating system (Hamilton & Cantor, 2004). CRAs only adjust ratings when they believe that a given issuer has experienced a stable and permanent change in basic creditworthiness. Many market participants, such as bond issuers, investment management firms (particularly pension and mutual funds) and financial regulators, prefer this approach since they often take actions based on rating adjustments. Thus, they may incur unrecoverable costs if these actions need to be reversed if the rating change is subsequently reversed (Cantor, ap Gwilym, & Thomas, 2007; Cantor & Mann, 2007). Outlooks provide an indication that a change in the creditworthiness of an issuer has been observed, but its permanence has not been established. When the CRA believes that a permanent change in an issuer's creditworthiness has occurred, the issuer may be placed on watch for a rating change. When any remaining uncertainty is resolved, the rating is either changed or confirmed (Hamilton & Cantor, 2004; Vazza et al., 2005). Outlook and watch status are good predictors of future corporate and sovereign rating migrations (e.g. Alsakka & ap Gwilym, 2009, Hamilton & Cantor, 2004; Vazza et al., 2005). Credit outlook/watch also helps to identify issuers that are more likely to default or have their rating withdrawn (Metz & Donmez, 2008). Incorporating outlook/watch status into a portfolio's analytical methodologies is likely to result in more accurate assessment of risk, leading to more efficient allocation of capital (Vazza et al., 2005). 2.2. The importance of sovereign ratings Many reasons motivate our focus on sovereign issuers. SEC (2011) reports that the proportions of total ratings which are government securities are 80.95%, 81.13% and 72.06% for Moody's, S&P and Fitch, respectively, at year-end 2010. Sovereign credit actions have attached huge attention in recent years. The International Monetary Fund (2010b) states that sovereign default was the most pressing risk facing the global economy, with indebted European countries, including Greece, Iceland, Ireland and Portugal, causing widespread concerns. Breitenfellner and Wagner (2010) indicate that government support packages to rescue the financial system amounted to approximately €5.55 trillion by October 2008, as calculated by the Bank of England. This led to increasing public debt and thereby higher risk of sovereign default. The European Central Bank (ECB, 2011) highlights that tensions in euro area sovereign debt markets are a key risk factor for ongoing financial stability, and warned that any form of sovereign debt default could trigger a major crisis. The European sovereign debt crisis highlighted many examples of immediate reactions to CRA news. For example, on 1 June 2011, Moody's downgraded Greece to Caa1 from B1 (with negative outlook). In response, Greek 10-year government bond yields increased by 12 basis points, and the prices of Irish, Spanish and Portuguese bonds declined. The Stoxx Europe Index fell 0.7%, and the Euro was also affected. On 13 June 2011, S&P downgraded Greece from B to CCC (with negative outlook), and warned that any attempt to restructure the country's debt would be considered a default. In reaction to this, the five-year CDS on Greek government debt rose 58 bps to 1600 bps. Greek, Portuguese and Irish 10-year bond yields closed at euro-era highs of 16.79%, 10.66% and 11.34% respectively. 6 The ECB was concerned that a Greek default would lead to contagion, causing Greek banks to default and major damage to large banks in France and Germany. 6

Greece's sovereign rating was downgraded to Ca (CC), with developing (negative) outlook, by Moody's (S&P) on 25 (27) July 2011.

47

Sovereign ratings represent a ceiling for the ratings assigned to provincial governments, corporates and financial institutions. 7 Sovereign ratings have a strong influence on borrowing costs and they are the most important stimulus for enhancing the capability of governments and private sectors to access global capital markets, attracting international capital and investment (Kim & Wu, 2008). Duggar et al. (2009) find that sovereign risk is a key factor in corporate defaults both during and outside sovereign crises, and also show how sovereign defaults can spill over into the corporate sector driven by institutional and political factors. 2.3. Lead–lag analysis of ratings CRAs have varying experience in different countries, they differ in the methodologies used in judging the creditworthiness of a sovereign borrower, and release only limited information about their methodologies. Differences across CRAs could affect both the time frame and the manner in which they react to any new information by adjusting the outlook/watch status. CRAs would rationally treat an outlook/watch adjustment by another CRA as a trigger leading them to review their own ratings. It could be viewed as costeffective to follow up a competitor's signal. Issuers seek credit quality improvements to be reflected in their ratings and/or outlook/watch status as quickly as possible in order to enable them to reduce borrowing costs and/or enhance capital inflows. Similarly, investors appreciate timely information about any change in credit risk affecting their invested funds. The earlier a change is signalled through an adjustment in the credit opinion, the better it is for the CRA's credibility in the market. Rating leadership can be considered as a sign of the predictive ability of a given CRA (Güttler & Wahrenburg, 2007). Prior literature on lead–lag analysis across CRAs is very limited. Additionally, all previous studies are focused on lead–lag behaviour for actual rating changes only, and mainly for corporate ratings. However, significant discrepancies between sovereign and corporate ratings performance have been demonstrated. CRAs apply different approaches and consider different inputs to evaluate the creditworthiness of sovereign and corporate issuers (see Alsakka & ap Gwilym, 2009; Cantor & Packer, 1996). Johnson (2004) shows that S&P follows Egan-Jones (a small CRA active since 1995) in downgrading corporate issuers. Güttler and Wahrenburg (2007) analyse the lead–lag relationship for credit ratings of near-to-default corporate issuers with ratings from Moody's and S&P during 1997–2004. They find that given a rating change by Moody's (S&P), the subsequent rating adjustment by S&P (Moody's) is significantly of greater magnitude in the short-term (1–180 days). Further, Güttler (2011) analyses the lead–lag relationship between Moody's and S&P corporate ratings during 1994–2005, and reveals that previous upgrades (downgrades) by one of these CRAs are associated with higher rating intensities for most one-notch upgrades (downgrades) by the other CRA. Güttler's (2011) evidence suggests that positive watch additions by one CRA increase the upgrade intensities of the other CRA even more sharply than negative watch additions increase the downgrade intensities. Alsakka and ap Gwilym (2010) is the only study to investigate lead–lag relationships in sovereign ratings. They use five CRAs: Moody's, S&P, Fitch, Japan Credit Rating Agency and Japan Rating & Investment Information, and find that S&P demonstrates the least dependence on other CRAs, and Moody's tends to be the first mover in upgrades. They point out that rating actions by Japanese agencies tend to lag those of the larger CRAs, although there is some evidence that they lead Moody's actions. Alsakka and ap Gwilym (2010) only consider actual rating changes, whereas this paper is the first to 7 Moody's, S&P and Fitch have recently eliminated their sovereign ceiling rule. Though the ceiling effect is no longer absolute, there remains a “sovereign ceiling lite” (Alsakka & ap Gwilym, 2010).

48

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

focus on ‘credit signal leadership’ by focusing on watch and outlook announcements.

Table 1 Descriptive statistics of the full data sample. Moody's

S&P

Fitch

108 91 15

120 0 17

107 25 14

1 2 3

106 55 4

17 59 0

39 45 2

4 5 6

59 165 80

59 76 180

47 86 101

7 8 9

18 98 59

68 248 208

25 126 103

10 11 12

56 264 512 588 0.5% 2.4% 2.0% 8.0% 18.0% 24.2% 17.7% 27.2% 22.3%

26 129 255 341 5.6% 5.9% 4.7% 9.1% 16.7% 20.2% 17.3% 20.5% 21.2%

13 14 15 16 17 18 19 20 21 22 23 24 25

3. Data sample The dataset consists of daily observations of long-term foreigncurrency (LT FC) ratings, outlook and watch of all sovereigns rated by three international CRAs (Moody's, S&P, and Fitch) during the period from 10 August 1994 to 31 December 2009. All event dates are obtained from relevant publications from each CRA. These three CRAs dominate the credit rating industry, with Moody's (S&P) accounting for 36.90% (42.27%) of the market, while Fitch's share is 17.93%, based on the number of ratings reported by the ten registered NRSROs at year-end 2010 (SEC, 2011). In terms of staffing, Moody's, S&P and Fitch employ 30.18%, 33.71% and 26.29% of the total number of credit analysts and supervisors reported by NRSROs at year-end 2010. Also, the shares of the total government securities ratings reported by NRSROs at year-end 2010 are 38.49%, 44.19% and 16.65% for Moody's, S&P and Fitch, respectively (SEC, 2011). In 1989, S&P was the first CRA to start applying outlook/watch to its sovereign ratings (Chambers & Ontko, 2007; Vazza et al., 2005). Moody's started using watch signals in 1991, while outlooks came into extensive use in 1995 (Hamilton & Cantor, 2004). Fitch has used rating watch since it began assigning ratings to sovereign issuers on 10 August 1994 (Klaar & Riley, 2005). Fitch began to assign outlook to sovereign LT FC ratings on 21 September 2000. In order to construct a consistent data period for the three CRAs, 10 August 1994 is the earliest feasible start date. 8 Table 1 presents summary statistics on the rating outlook and watch data. The dataset comprises: 165 watch and 174 outlook actions by Moody's for 108 sovereigns, 76 watch and 512 outlook actions by S&P for 120 sovereigns, and 86 watch and 255 outlook actions by Fitch for 107 sovereigns (see Rows 1, 8 and 15). For S&P, there is no watch for possible upgrade and, thus, no action of confirming rating after being placed on watch for upgrade (Rows 2 and 6). As a matter of policy, S&P tends not to place sovereigns on watch for possible upgrade (Chambers & Ontko, 2007). 9 The number of positive watch and outlook signals (106 and 98) exceeds the number of negative actions (59 and 76) in the case of Moody's, while it is vice versa in the case of S&P and Fitch (see Rows 4, 7, 11 and 14). In rows 8 and 15, the total number of watch actions by Moody's is similar to the total number of outlook adjustments (165 vs. 174). In contrast, the number of outlook changes by Fitch is 3 times greater than watch adjustments (255 vs. 86), bearing in mind that Fitch started using outlook (watch) in September 2000 (August 1994). The number of outlook actions by S&P is approximately 7 times greater than watch adjustments (512 vs. 76). This can be partly explained by S&P's tendency to reverse its outlook actions far more frequently than Moody's. S&P illustrates the highest percentage of reversals in outlook actions; 22.3% of the total outlook adjustments, while this is 14.4% for Moody's and 21.2% for Fitch (see Row 25). This is also in line with the findings of Alsakka and ap Gwilym (2010) that S&P sovereign ratings show the highest rating volatility, while Moody's shows the greatest sovereign rating stability. This suggests that S&P's policy tends to aim for greater short-term accuracy, while Moody's policy puts more weight on stability. 10 This highlights

8 There are only 10 (10) watch changes by Moody's (S&P) in the period 1991 (1989) to 9 August 1994. Only 35 outlook changes are observed (for S&P) between 1989 and 9 August 1994. 9 However, on 22 July 2010, S&P placed Ukraine (rated at ‘B’) on positive watch. Ukraine was upgraded to ‘B+’ one week later. This is the only time that S&P has placed a sovereign on positive watch. 10 Many market participants seem to support Moody's policy of avoiding rating reversals (Fons, Cantor, & Mahoney, 2002). However, Moody's has not specified its policy in more detail (Löffler, 2005).

No. of countries Watch for possible upgrade Confirm rating after being placed on Watch for downgrade PW (Positive Watch Signals) Watch for possible downgrade Confirm rating after being placed on Watch for upgrade NW (Negative Watch Signals) Total Watch Signals (PW + NW) To positive Outlook from stable/negative Outlook To stable Outlook from negative Outlook PO (Positive Outlook Signals) To negative Outlook from stable/positive Outlook To stable Outlook from positive Outlook NO (Negative Outlook Signals) Total Outlook Signals (PO + NO) Total Outlook/Watch Signals (rows 8 + 15) Investment grade PW % of Total Speculative grade PW % of Total Investment grade NW % of Total Speculative grade NW % of Total Investment grade PO % of Total Speculative grade PO % of Total Investment grade NO % of Total Speculative grade NO % of Total Outlook reversals % of total outlook actions

17 76 174 339 17.4% 13.9% 7.7% 9.7% 13.9% 15.0% 10.6% 11.8% 14.4%

Row number

This Table presents summary statistics for the dataset, which comprises three international rating agencies. The sample consists of daily long-term foreign-currency outlook and watch signals for all sovereigns rated by each agency during the period from 10 August 1994 to 31 December 2009.

different practices applied by CRAs in adjusting the outlook and watch status of sovereign issuers, which is ultimately one of the paper's contributions. The percentages of speculative-grade issuers which experienced watch or outlook changes generally exceed those of investmentgrade issuers, with the exception of Moody's positive watch changes (see Rows 17 to 24). This is not unexpected as sovereigns rated at the lower (higher) range of the rating scale, are more (less) likely to experience rating changes, and thus outlook and watch adjustments (Alsakka & ap Gwilym, 2010). Fig. 1 illustrates the net actions (i.e. positive minus negative) of outlooks/watch announced by each CRA during the sample period. S&P and Fitch appear to be more alike, which is consistent with Alsakka and ap Gwilym's (2010) finding that S&P and Fitch have the lowest frequency of sovereign rating disagreements between CRAs. Additionally, Moody's tends to offer more positive signals than S&P and Fitch, which is also in line with Alsakka and ap Gwilym's (2010) finding that Moody's shows a slight tendency to assign the higher rating. 11 This is also supported by higher positive outlook/ watch actions than negative ones by Moody's, but not by S&P and Fitch as discussed earlier (see Rows 4, 7, 11 and 14 of Table 1). The net outlook and net watch are negative or around zero in 1998 and 2001 reflecting the Asian and Russian crises, and the crises in Latin America, respectively. The net outlook and net watch actions rise in 2000 and 2004–2006 to mirror economic growth, especially in emerging countries. In contrast, there is a strong negative trend through the 2007–2009 period (with negative net values in 2008 and 2009) reflecting the financial crisis.

11 This is particularly clear in watch for 2006 and 2007. However, bear in mind that S&P does not typically assign positive watch to sovereigns.

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

49

Outlook 10 0 -10 -20

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

1999

1998

1997

-40

1996

-30 1995

Net Outlook

20

Year Moody's

S&P

Fitch

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

1999

1998

1997

1996

1995

Net Watch

Watchlist 20 15 10 5 0 -5 -10 -15

Year Moody's

S&P

Fitch

Fig. 1. Annual net outlook and watch actions for sovereigns during 1995–2009.“Net outlook/watch” equals positive minus negative signals.

Table 2 Distribution of the outlook and watch signals for sovereigns rated by each pair of agencies. Positive outlook signals (PO)

Positive watch signals (PW)

Total signals

Panel I: Moody's and S&P (97 countries) Moody's signals 56 74 S&P signals 53 236

92 223

102 14

324 526

Panel II: Moody's and Fitch (84 countries) Moody's signals 45 56 Fitch signals 44 115

76 111

91 39

268 309

Panel III: S&P and Fitch (97 countries) S&P signals 42 148 Fitch signals 45 122

146 114

13 36

349 317

Negative watch signals (NW)

Negative outlook signals (NO)

This Table provides an overview of positive and negative outlook/watch signals for all sovereigns rated by each pair of agencies during the period from 10 August 1994 to 31 December 2009.

To analyse whether there is interdependence across CRAs, rating outlook and watch actions for sovereign issuers that are rated by at least two CRAs during the sample period are the focus of attention. 12 Table 2 reveals the distribution of the outlook and watch changes (the dependent variable for the lead–lag analysis) by each pair of CRAs. 13

The inferences from Table 2 are consistent with those discussed above based on Table 1. 4. Methodology We apply the ordered probit modelling approach, which considers the discrete, ordinal nature of credit ratings, outlook and watch status and the changes in these variables. The ordered probit model has been widely employed in a variety of contexts in credit rating research (e.g. Güttler & Wahrenburg, 2007; Manzoni, 2004). We first investigate whether there is momentum in outlook/watch actions by each CRA. The existence of momentum would imply that a sovereign which experienced a positive (negative) outlook/ watch action is more prone to subsequently experience further positive (negative) outlook/watch changes (within a given time horizon). The hypotheses that there is no positive nor negative momentum in outlook and watch signals are tested by estimating the following ordered probit regression for each CRA (Moody's, S&P or Fitch) separately:


yit ¼ β1 opM i þ β2 onM i þ β3 wpMi þ β4 wnMi þ εit ; εit e N ð0; 1Þ:

ð1Þ

yit* is an unobserved latent variable linked to the observed ordinal response categories yit by the measurement model: 2

12

For the lead-lag relationship between Fitch and Moody's/S&P, outlook data is included starting on 21 September 2000 when Fitch started assigning outlooks for sovereigns. 13 Table 1 provides a complete picture of the dataset of events. Table 1 considers all sovereigns rated by each CRA, and these sub-samples are used for examining the existence of momentum (based on Eq. 1 later). Table 2 presents specific details for sovereigns rated by each pair of CRAs, i.e. controlling for the differing sets of sovereigns rated by each CRA. These sub-samples are used for examining the lead-lag relationship (based on Eqs. (3) and (4) later).

3 −2 if y
it ≤μ 1
6 −1 if μ by ≤μ 7 it 1 2 7: yit ¼ 6 4 1 if μ by
≤μ 5 it 2 3
2 if μ 3 byit

ð2Þ

The μ represents thresholds to be estimated (along with the β coefficients) using maximum likelihood estimation, subject to the constraint that μ1 b μ2 b μ3.

50

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

yit represents outlook/watch status changes by CRA A at time t. It is an ordinal variable taking the value of −2, −1, 1, or 2, as follows: ‘−2’

if sovereign i experiences negative watch change by CRA A at time t. This includes: placing sovereign i on watch for possible downgrade, and the action of confirming the rating of sovereign i after being on watch for possible upgrade. if sovereign i experiences negative outlook change by CRA A at time t. This contains: changes to negative outlook from stable/positive outlook; assigning negative outlook simultaneously with a rating change, and changes to stable outlook from positive outlook. if sovereign i experiences positive outlook change by CRA A at time t. This contains: changes to positive outlook from stable/negative outlook, assigning positive outlook simultaneously with a rating change, and changes to stable outlook from negative outlook. if sovereign i experiences positive watch change by CRA A at time t. This includes: placing sovereign i on watch for possible upgrade, and the action of confirming the rating of sovereign i after being on watch for possible downgrade.

‘−1’

‘1’

‘2’

opMi is a dummy variable taking the value of 1 if the most recent action by CRA A for sovereign i within the last 24 months is a positive outlook signal, and 0 otherwise. onMi is a dummy variable taking the value of 1 if the most recent action by CRA A for sovereign i within the last 24 months is a negative outlook signal, and 0 otherwise. wpMi is a dummy variable taking the value of 1 if the most recent action by CRA A for sovereign i within the last 12 months is a positive watch signal, and 0 otherwise. wnMi is a dummy variable taking the value of 1 if the most recent action by CRA A for sovereign i within the last 12 months is a negative watch signal, and 0 otherwise. The “memory” of outlook (watch) momentum is limited to 24 (12) months for: (i) comparability with prior literature on rating momentum, such as Lando and Skødeberg (2002), Hamilton and Cantor (2004), Fuertes and Kalotychou (2007); and (ii) given the fact that outlook (watch) duration is on average 12–18 (3–6) months. 14 Potential lead–lag relationships regarding outlook/watch actions are assessed for each pair of CRAs, using a Granger-like method with ordered probit regression (Alsakka & ap Gwilym, 2010; Güttler & Wahrenburg, 2007). We accomplish a relative comparison of the probability of an outlook/watch change by CRA A conditional on a previous outlook/watch action and/or an actual rating change by CRA B. The restriction to a relative comparison arises from the fact that rating signal adjustments are not random events (see Güttler, 2011). We estimate the following models with CRA A as potential follower and CRA B as potential leader in Eq. (3), and vice versa in Eq. (4):


A

yit ¼

3 X

1

B

βh opi;h þ

h¼1

þ

2 X

3 X




B

3 X

5

B

βk upi;k þ

1

A

h¼1

þ

k¼1

2 X

2 X

3

B

βs wpi;s þ

s¼1 6

2 X

4

B

βs wni;s

s¼1

B

βk dni;k þ εi

ð3Þ

k¼1

γh opi;h þ

2 X

B

h¼1

k¼1

yit ¼

2

βh oni;h þ

3 X

2

A

γh oni;h þ

þ

2 X

6 A γk dni;k

3

A

γs wpi;s þ

s¼1

h¼1 5 A γ k upi;k

2 X

þ νi :

2 X

4

A

γ s wni;s

s¼1

yit* is an unobserved latent variable linked to the observed ordinal response categoriesyit. yit is yitM for Moody's, yitSP for S&P, or yitF for Fitch, referring to an outlook/watch status change by CRA A in Eq. (3) or CRA B in Eq. (4) for sovereign i on day t. yit is an ordinal variable taking the value of −2, −1, 1, and 2 (as in Eq. (1)). opi, h (oni, h) equals one if sovereign i experienced a positive (negative) outlook change by the potential leader CRA within three time windows h prior to the outlook/watch action by the potential follower CRA at day t, zero otherwise. The three predefined time windows are h = 1 for 1–15 days, h = 2 for 16–180 days, and h = 3 for 181–540 days. wpi, s (wni, s) equals one if sovereign i experienced a positive (negative) watch change by the potential leader CRA within two time windows s prior to the outlook/watch action by the potential follower CRA at day t, zero otherwise. The two predefined time windows are s = 1 for 1–15 days and s = 2 for 16–180 days. upi, k(dni, k) takes the value of one if sovereign i is upgraded (downgraded) by the potential leader CRA within two time windows k prior to the outlook/watch action by the potential follower CRA at day t, zero otherwise. The two predefined windows are k = 1 for 1–15 days and k = 2 for 16–180 days. 15 To estimate the economic significance of each variable, we follow Livingston, Naranjo, and Zhou (2008) in calculating the marginal effects (MEs). The marginal effects report the impacts on the probability of outlook/watch status changes (dependent variable) when the independent dummy variables take the value of 1. 5. Empirical results 5.1. Momentum in outlook and watch signals Table 3 presents the results of Eq. (1). Outlook and watch signals for all sovereigns rated by each CRA are used. The dataset reported in Table 1 is used for these estimations. The coefficients for the wpM and wnM variables are not statistically significant at the 5% level for all estimations, implying no evidence of momentum in positive nor negative watch signals for Moody's, S&P or Fitch. The coefficients for the onM variable are statistically significant at the 5% level or less for each CRA, revealing evidence of momentum in negative outlook actions. A sovereign which experienced a previous negative outlook adjustment by Moody's, S&P or Fitch is more likely to subsequently experience a further negative outlook change by the same CRA by 6.4%, 5.2% and 5.7%, respectively (see ME of ‘−1’ for the onM variable). In comparison, Fuertes and Kalotychou (2007) find significant downgrade momentum for Moody's sovereign ratings, while Alsakka and ap Gwilym (2009) report significant downgrade momentum for S&P and Fitch sovereign ratings, but insignificant upgrade momentum. The MEs of ‘−2’ for the onM coefficients suggest that a sovereign with a previous recent negative outlook action has increased probability of being placed on negative watch by Moody's, S&P, and Fitch by 25.5%, 4.5%, and 9.1% respectively. The only statistically significant coefficient for the opM variable occurs for Moody's (see Panel I). However, this coefficient does not imply the existence of momentum in Moody's positive outlook actions since the ME of ‘+1’ is negative. A sovereign which experienced a previous positive outlook signal by Moody's is less likely to subsequently experience a further positive outlook change. The ME of ‘+2’ for the opM variable also shows that a sovereign which experienced a recent positive outlook action by

ð4Þ

k¼1

14 Alternative specifications for outlook (watch) momentum with 18 (6 and 24) months “lookback” periods produced qualitatively similar results.

15 The pre-defined windows (h, s, and k) for actions by potential leader CRA prior to an action by potential follower CRA are chosen in line with Güttler and Wahrenburg (2007) and Alsakka and ap Gwilym (2010). Yet we consider the characteristics of watch and outlook signals in this regard, i.e. the medium-term view of outlooks and the ex-ante target of three months for watch status.

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

51

Table 3 Momentum in outlook and watch signals. Coef

t-Val

ME % Avr |Chg|

−2

−1

1

2

Panel I: Moody's opM onM wpM wnM Pseudo R²

0.72** − 0.82** 0.20 0.12 4.9%

4.42 13.5 − 4.91 15.9 0.08 0.3 0.55 2.2 No. of observations 339

− 13.5 25.5 − 0.4 − 2.7

− 11.3 6.4 − 0.2 − 1.8

− 2.1 − 9.1 0.1 0.3

26.9 − 22.7 0.1 4.2

Panel II: S&P opM onM wpM wnM Pseudo R²

0.18 − 0.25* na − 0.22 1.1%

1.89 3.7 − 2.38 4.8 na na − 0.89 4.2 No. of observations 588

− 3.0 4.5 na 4.1

− 4.4 5.2 na 4.3

6.0 − 8.2 na − 7.2

1.3 − 1.5 Na − 1.2

Panel III: Fitch opM onM wpM wnM Pseudo R²

0.01 − 0.38** 0.03 0.004 0.9%

0.05 0.12 − 2.92 7.4 0.06 0.6 0.01 0.07 No. of observations 341

− 0.13 9.1 − 0.6 − 0.08

− 0.11 5.7 − 0.5 − 0.06

0.13 − 8.6 0.6 0.07

0.12 − 6.3 0.6 0.07

This Table reports the results of ordered probit estimations of Eq. (1) using data from: Moody's in Panel I, S&P in Panel II, and Fitch in Panel III, for 10 August 1994 to 31 December 2009. The dependent variable yit(yitM,yitSP and yitF, respectively) refers to an outlook/watch status change by agency A for sovereign i on day t. Four different classes of outlook/watch status changes are employed: − 2, − 1, 1, and 2; i.e. negative watch, negative outlook, positive outlook, and positive watch signals. The independent variables are as follows. opMi (onMi): a dummy variable taking the value of 1 if the most recent action by agency A within the last 24 months is a positive (negative) outlook signal, and 0 otherwise; and wpMi (wnMi): a dummy variable taking the value of 1 if the most recent action by agency A within the last 12 months is a positive (negative) watch signal, and 0 otherwise. We apply Huber–White robust standard errors. We also estimate and report the impact of each variable on the probability of an outlook/watch status change (marginal effect (ME)). **Significant at 1% level; *significant at 5% level. The estimates of the three threshold parameters are significant at the 1% level in all estimations, and are not shown here. ‘na’: no data available/observed.

Moody's has significantly elevated probabilities of being placed on watch for possible upgrade by 26.9%, but is less likely to experience negative outlook/watch actions. In brief, Table 3 provides evidence of momentum in negative outlook actions, but no evidence of momentum in positive outlook nor watch signals. However, the effect of prior negative outlook actions on the subsequent action is not especially strong, particularly for S&P. This can be partly explained by S&P aiming for accuracy more than stability, and the presence of more reversals in outlook actions in the S&P sample (see Table 1, Row 25). 5.2. Lead–lag relationships for outlook and watch Tables 4 to 6 present results for the outlook/watch lead–lag relations across the three CRAs, i.e. the results of estimating Eqs. (3) and (4). For these estimations, we use sub-samples of sovereigns that are rated by each pair of CRAs (see data summary in Table 2). 5.2.1. Moody's and S&P Table 4 considers Moody's and S&P. In Panel I, the coefficients for the on by S&P variable within the h1 and h2 time windows are statistically significant at the 1% level. The MEs in these cases imply that a sovereign that experienced a negative outlook change by S&P has increased probabilities to be placed on negative watch by Moody's within 15 (16–180) days by 26.8% (38.8%), and to experience a negative outlook action by Moody's by 9.1% (8.8%). The coefficients for the wn by S&P variable within the s1 and s2 time windows are statistically significant at the 1% level. The MEs imply that a sovereign assigned a negative watch signal by S&P has increased probabilities to be placed on negative watch by Moody's within 15 (16–180) days by 36.0% (17.2%), and to experience a negative outlook adjustment by Moody's by 6.1% (8.3%). In contrast, positive outlook actions by S&P have an insignificant impact on future outlook/watch actions by Moody's, given the insignificant coefficients for the op by S&P variable. The coefficients for the dn by S&P variable within the k1 and k2 time windows are statistically significant at the 5% level. Therefore,

sovereigns downgraded by S&P have elevated (decreased) probabilities of negative (positive) outlook/watch adjustments by Moody's within 180 days. Additionally, the coefficient for the up by S&P variable within the k2 time window is statistically significant at the 1% level. The MEs suggest that issuers upgraded by S&P have an elevated probability of being placed on watch for possible upgrade by Moody's by 16.8% within 16–180 days. This implies that Moody's may follow (by issuing positive watch, which is considered as a strong signal) positive news released by S&P, only if the latter were upgrades (not positive outlook adjustments). Panel II shows that most coefficients are statistically significant at the 5% level or less, with the exceptions being those for op by Moody's and on by Moody's within the h3 time window and for wn by Moody's within the s2 time window. The ME analysis illustrates that a sovereign that experienced a positive (negative) outlook or watch action by Moody's has increased probabilities of experiencing positive (negative) outlook/watch changes by S&P, while it has decreased probabilities of experiencing negative (positive) outlook/watch changes by S&P, in the subsequent 180 days. Similarly, issuers upgraded (downgraded) by Moody's have elevated probabilities of experiencing positive (negative) outlook/watch changes, and decreased probabilities of experiencing negative (positive) adjustments, by S&P for all time windows. It is clear that Moody's is more likely than S&P to lead in positive actions. In contrast, S&P has a greater tendency to lead in negative actions. Moody's negative outlook/watch adjustments tend to follow S&P actions to a greater extent than vice versa, as suggested by the MEs, particularly during the 1–15 days window (which has an implication for market reactions). S&P negative outlook actions increase the probabilities of Moody's negative watch (outlook) adjustments within 15 days to a greater extent than vice versa (26.8% (9.1%) versus 11.3% (8.2%)). Also, S&P negative watch actions increase the probability of Moody's negative watch adjustments within 15 days to a greater extent than vice versa (36.0% versus 18.1%). Overall, S&P is less dependent because the Pseudo R² value is 13.3% when Moody's is a follower compared to 6.4% when S&P is a follower.

52

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

Table 4 Leads and lags between Moody's and S&P. Coef

t-Val

ME % Avr |Chg|

−2

−1

1

2

Panel I: Moody's as outlook/watch follower to S&P actions, Eq. (3) op by S&P- h1: 1–15 days before 0.87 op by S&P- h2: 16–180 days before 0.33 op by S&P- h3: 181–540 days before 0.51 on by S&P- h1: 1–15 days before − 0.94** on by S&P- h2: 16–180 days before − 1.30** on by S&P- h3: 181–540 days before − 0.28 wp by S&P- s1: 1–15 days before na wp by S&P- s2: 16–180 days before na wn by S&P- s1: 1–15 days before − 1.15** wn by S&P- s2: 16–180 days before − 0.65* up by S&P- k1: 1–15 days before 0.26 up by S&P- k2: 16–180 days before 0.47** dn by S&P- k1: 1–15 days before − 0.68** dn by S&P- k2: 16–180 days before − 0.64* Pseudo R² 13.3%

1.68 16.3 1.80 6.1 1.82 9.1 − 4.17 18.0 − 6.30 23.8 − 0.80 5.5 na na na na − 3.71 21.1 − 2.37 12.7 0.74 4.7 2.88 8.4 − 2.76 13.4 − 2.39 12.6 No. of observations 324

− 9.7 − 5.4 − 7.3 26.8 38.8 6.2 na na 36.0 17.2 − 4.2 − 7.2 18.0 16.4

− 17.0 − 6.8 − 10.3 9.1 8.8 4.8 na na 6.1 8.3 − 5.2 − 9.6 8.7 8.8

− 6.0 0.5 − 0.7 − 14.5 − 20.2 − 2.8 na na − 19.6 − 9.2 0.5 − 0.04 − 9.5 − 8.5

32.6 11.6 18.3 − 21.1 − 27.4 − 8.2 na na − 22.6 − 16.3 9.0 16.8 − 17.2 − 16.7

Panel II: S&P as outlook/watch follower to Moody's actions, Eq. (4) op by Moody's- h1: 1–15 days 1.08** op by Moody's- h2: 16–180 days 0.46* op by Moody's- h3: 181–540 days 0.32 on by Moody's- h1: 1–15 days − 0.53* on by Moody's- h2: 16–180 days − 0.63* on by Moody's- h3: 181–540 days 0.33 wp by Moody's- s1: 1–15 days 1.57* wp by Moody's- s2: 16–180 days 0.78** wn by Moody's- s1: 1–15 days − 0.76** wn by Moody's- s2: 16–180 days − 0.30 up by Moody's- k1: 1–15 days 0.39* up by Moody's- k2: 16–180 days 0.54** dn by Moody's- k1: 1–15 days − 0.92** dn by Moody's — k2: 16–180 days 0.78** Pseudo R² 6.4%

12.54 19.1 2.07 9.1 1.25 6.3 − 2.47 9.7 − 2.38 11.2 0.85 6.5 2.45 24.1 4.05 14.9 − 2.77 13.1 − 1.30 5.7 2.11 7.7 3.07 10.6 − 3.49 15.3 − 4.27 13.6 No. of observations 526

− 7.6 − 5.1 − 3.9 11.3 13.8 − 4.0 − 8.2 − 7.1 18.1 5.3 − 4.5 − 5.7 23.2 18.1

− 30.6 −13.0 − 8.7 8.2 8.6 − 9.1 − 40.0 − 22.7 8.2 6.0 − 10.9 − 15.4 7.3 9.1

24.3 14.8 10.6 − 18.0 − 20.8 11.0 19.8 22.2 − 24.6 − 10.3 12.7 17.0 − 28.7 − 25.3

13.8 3.3 2.0 − 1.4 − 1.6 2.10 28.4 7.5 − 1.7 − 1.0 2.6 4.1 − 1.9 − 1.8

This Table reports the results of ordered probit estimations of Eq. (3) and Eq. (4) using data from Moody's and S&P for 10 August 1994 to 31 December 2009. The dependent variables are: yitM in Panel I (Eq. (3)), referring to an outlook/watch status change by Moody's (follower agency) for sovereign i on day t, and yitSP in Panel II (Eq. (4)), referring to an outlook/watch status change by S&P (follower agency) for sovereign i on day t. Four different classes of outlook/watch status changes are employed: − 2, − 1, 1, and 2; i.e. negative watch, negative outlook, positive outlook, and positive watch signals. The independent variables are as follows. opi, h (oni, h): a dummy variable taking the value of 1 if there is a positive (negative) outlook change by the potential leader agency, in three predefined windows of time h, with h = 1 for 1–15 days, h = 2 for 16–180 days, and h = 3 for 181–540 days, prior to the outlook/watch action for sovereign i at time (day) t by the potential follower agency, zero otherwise; wpi, s (wni, s): a dummy variable taking the value of 1 if there is a positive (negative) watch change by the potential leader agency, in two predefined windows of time s, with s = 1 for 1–15 days and s = 2 for 16–180 days, prior to the outlook/watch action for sovereign i at time (day) t by the potential follower agency, zero otherwise; and upi, k(dni, k): a dummy variable taking the value of 1 if a sovereign i is upgraded (downgraded) by the potential leader agency, in two predefined windows of time k, with k = 1 for 1–15 days and k = 2 for 16–180 days, prior to the outlook/watch action for sovereign i at time (day) t by the potential follower agency, zero otherwise. We apply Huber–White robust standard errors. We also estimate and report the impact of each variable on the probability of an outlook/watch status change (marginal effect (ME)). **Significant at 1% level; *significant at 5% level. The estimates of the three threshold parameters are significant at the 1% level in all estimations, and are not shown here. ‘na’: no data available/observed.

5.2.2. Moody's and Fitch Table 5 considers Moody's and Fitch. Panel I of Table 5 shows that the coefficient for the on by Fitch variable within the h2 time window is statistically significant at the 1% level. The MEs in this case indicate that a sovereign which experienced a recent negative outlook action by Fitch has increased probabilities to experience a negative watch (negative outlook) change by Moody's by 16.0% (6.7%) within 16–180 days. In contrast, watch actions by Fitch have an insignificant influence on future outlook/watch adjustments by Moody's. The coefficients for the up by Fitch variable within the k2 time window and the dn by Fitch variable within the k1 and k2 time windows are statistically significant at the 1% level. The MEs in these cases suggest that sovereigns upgraded by Fitch have a significantly elevated probability (by 33.8%) to be placed on watch for possible upgrade by Moody's within 16–180 days. This implies that Moody's may follow (by issuing positive watch, which is considered as a strong signal) positive news released by Fitch, as with S&P, only if the news involves upgrades (not positive outlook or watch adjustments). Further, a downgrade by Fitch subsequently increases (decreases) the probabilities of Moody's negative (positive) outlook and watch movements within 180 days. Panel II of Table 5 shows that the coefficients for the op by Moody's and on by Moody's variables within the h1 and h2 time windows, the

wp by Moody's and wn by Moody's variables within the s1 and s2 time windows, the up by Moody's variable within the k2 time window and the dn by Moody's variable within the k1 time window are statistically significant at the 5% level or less. The MEs indicate that a sovereign which experienced a recent positive outlook change by Moody's is more (less) likely to experience positive (negative) outlook/watch adjustments by Fitch for time windows up to 180 days. Similarly, negative outlook actions by Moody's significantly increase (reduce) the probabilities of negative (positive) outlook/watch actions by Fitch for time windows up to 180 days. Moody's positive (negative) watch actions increase the probabilities of positive (negative) watch adjustments by Fitch by 28.0% (36.0%) within 15 days and by 13.0% (23.3%) within 16–180 days, while significantly reducing the probabilities of negative (positive) outlook/watch movements by Fitch. In addition, an actual downgrade (upgrade) by Moody's significantly and strongly increases the probabilities of Fitch negative (positive) watch actions by 71.5% (18.5%) within 15 days (16–180 days), while significantly decreasing the probabilities of positive (negative) outlook/watch adjustments by Fitch. It is clear that Fitch outlook/watch adjustments tend to follow Moody's actions to a much greater extent than vice versa, as suggested by the MEs. This is also supported by the outcome that the

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

53

Table 5 Leads and lags between Moody's and Fitch. Coef

t-Val

ME % Avr |Chg|

Panel I: Moody's as outlook/watch follower to Fitch op by Fitch — h1: 1–15 days before op by Fitch — h2: 16–180 days before op by Fitch — h3: 181–540 days before on by Fitch — h1: 1–15 days before on by Fitch — h2: 16–180 days before on by Fitch — h3: 181–540 days wp by Fitch — s1: 1–15 days before wp by Fitch — s2: 16–180 days before wn by Fitch — s1: 1–15 days before wn by Fitch — s2: 16–180 days before up by Fitch — k1: 1–15 days before up by Fitch — k2: 16–180 days before dn by Fitch — k1: 1–15 days before dn by Fitch — h2: 16–180 days before Pseudo R²

actions, Eq. (3) − 0.04 − 0.48 0.8 − 0.02 − 0.09 0.4 0.06 0.23 1.03 − 0.40 − 1.12 7.8 − 0.58** − 3.04 11.4 − 0.50 − 1.47 9.8 na na na Dropped (only two observations) na na na − 0.66 − 1.63 12.9 0.61 0.94 11.8 0.88** 3.22 16.9 − 1.01** − 2.99 19.2 − 0.74** − 2.99 14.4 7.3% No. of observations 268

Panel II: Fitch as outlook/watch follower to Moody's op by Moody's — h1: 1–15 days op by Moody's — h2: 16–180 days op by Moody's — h3: 181–540 days on by Moody's — h1: 1–15 days on by Moody's — h2: 16–180 days on by Moody's — h3: 181–540 days wp by Moody's — s1: 1–15 days wp by Moody's — s2: 16–180 days wn by Moody's — s1: 1–15 days wn by Moody's — s2: 16–180 days up by Moody's — k1: 1–15 days up by Moody's — k2: 16–180 days dn by Moody's — k1: 1–15 days dn by Moody's — k2: 16–180 days Pseudo R²

actions, Eq. (4) 0.61** 0.73** 0.32 − 1.19** − 0.96** − 0.27 1.04** 0.59** −1.21** − 0.87* − 0.21 0.78** − 2.22** − 0.44 12.9%

6.85 11.8 2.91 14.0 1.93 6.5 − 3.30 18.4 − 5.41 16.4 − 0.72 5.3 2.65 18.4 2.60 11.5 − 2.70 18.6 − 2.01 14.9 − 0.73 4.0 2.76 14.8 − 4.98 35.8 − 1.14 8.4 No. of observations 309

−2

−1

1

2

1.0 0.4 − 1.2 10.5 16.0 13.6 na

0.7 0.3 − 0.9 5.13 6.7 5.9 na

− 0.1 − 0.04 0.1 − 2.8 − 5.0 − 4.2 na

− 1.5 − 0.7 2.0 − 12.8 − 17.8 − 15.3 na

na 19.2 − 9.3 − 12.6 32.1 21.6

na 6.7 − 10.1 − 14.1 6.2 7.3

na − 6.7 − 4.2 − 7.1 − 12.9 − 7.6

na − 19.2 23.5 33.8 − 25.4 − 21.3

− 6.9 − 8.2 − 4.6 35.5 25.7 5.5 − 8.9 − 7.1 36.0 23.3 4.0 − 8.4 71.5 9.6

− 16.6 − 19.8 − 8.4 1.4 7.0 5.0 − 27.8 − 16.0 1.1 6.5 4.0 − 21.2 − 23.3 7.1

9.8 11.1 6.8 − 28.5 − 24.3 − 7.0 8.7 10.1 − 28.8 − 22.3 − 5.3 11.1 − 38.4 − 11.6

13.7 16.9 6.2 − 8.3 − 8.4 − 3.5 28.0 13.0 − 8.3 − 7.5 − 2.8 18.5 − 9.8 − 5.2

This Table reports the results of ordered probit estimations of Eq. (3) and Eq. (4) using data from Moody's and Fitch for 10 August 1994 to 31 December 2009. The dependent variables are: yitM in Panel I (Eq. (3)), referring to an outlook/watch status change by Moody's (follower agency) for sovereign i on day t, and yitF in Panel II (Eq. (4)), referring to an outlook/watch status change by Fitch (follower agency) for sovereign i on day t. Four different classes of outlook/watch status changes are employed: − 2, − 1, 1, and 2 (see Table 4). The independent variables are: opi, h, oni, h, wpi, s, wni, s, upi, k, and dni, k (see Table 4 for the definitions). We apply Huber–White robust standard errors. We also estimate and report the impact of each variable on the probability of an outlook/watch status change (marginal effect (ME)). **Significant at 1% level; *significant at 5% level. The estimates of the three threshold parameters are significant at the 1% level in all estimations, and are not shown here. ‘na’: no data available/observed.

Pseudo R² value is 7.3% when Moody's is a follower compared to 12.9% when Fitch is a follower. Watch signals by Fitch have an insignificant influence on future outlook/watch adjustments by Moody's. This is consistent with Norden and Weber's (2004) findings that negative watch actions by Moody's and S&P affect both stock and CDS markets, while no abnormal performance is detected in response to credit actions by Fitch. Further, the average probabilities for outlook/watch changes following negative news by the other CRA are generally higher than these probabilities following positive news, which is possibly influenced by the stronger negative reputational effects of being slow in the case of negative actions. It is evident from Tables 4 and 5 that Moody's tends to lead positive outlook and watch actions. This is in line with Alsakka and ap Gwilym's (2010) evidence on lead–lag relationships for sovereign rating changes, and with Brooks et al.'s (2004) result that only Moody's sovereign upgrades are associated with positive abnormal stock returns. 5.2.3. S&P and Fitch Table 6 considers S&P and Fitch. Panel I shows that all coefficients for independent variables are significant at the 5% level or less, with the exceptions of those for the up by S&P and the dn by S&P variables within the k1 time window. This implies that Fitch outlook and watch actions are affected by prior S&P rating, outlook and watch changes for every time window, with the exception of S&P rating changes within 15 days prior to a Fitch action. Panel II demonstrates that the coefficients for the op by Fitch variable within the h1, h2 and h3 time windows are significant at the

1% level. The MEs imply that issuers which experienced recent positive outlook actions by Fitch have significantly elevated (decreased) probabilities of experiencing positive (negative) outlook/watch changes by S&P for all time windows. In contrast, Fitch negative outlook actions have an insignificant influence on future outlook/watch adjustments by S&P. The coefficient for the wn by Fitch variable within the s1 time window is significant at the 1% level, and the MEs suggest that negative watch actions by Fitch lead to a greater probability of negative watch adjustments by S&P within 15 days by 62.5%. The coefficients for the up by Fitch variable within the k1 and k2 time windows and the dn by Fitch variable within the k1 time window are statistically significant at the 5% level or less. The MEs indicate that upgrades by Fitch increase (decrease) the probabilities of positive (negative) outlook/watch actions by S&P for all time windows, while actual downgrades by Fitch increase (decrease) the probabilities of negative (positive) outlook/watch actions by S&P in the subsequent 15 days. It is clear that S&P has a tendency to lead Fitch outlook/watch actions to a greater extent than it follows them. This is supported by more significant independent variables when S&P is a leader than a follower to Fitch actions, and also by greater values of MEs when S&P is a leader than when it is a follower, particularly during the 1–15 days window (which has an implication for market reactions). S&P positive outlook actions increase the probabilities of Fitch positive watch adjustments within 15 days to a greater extent than vice versa (22.5% versus 13.9%). S&P negative watch actions increase the probabilities of Fitch negative watch adjustments within 15 days to

54

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

Table 6 Leads and lags between S&P and Fitch. Coef

t-Val

ME % Avr |Chg|

−2

−1

1

2

Panel I: Fitch as outlook/watch follower to S&P actions, Eq. (3) op by S&P — h1: 1–15 days before 1.06** op by S&P — h2: 16–180 days 0.81** op by S&P — h3: 181–540 days 0.51* on by S&P — h1: 1–15 days before − 1.38** on by S&P — h2: 16–180 days − 0.82** on by S&P — h3: 181–540 days − 0.73** wp by S&P — s1: 1–15 days before na wp by S&P — s2: 16–180 days na wn by S&P — s1: 1–15 days before − 2.61** wn by S&P — s2: 16–180 days − 1.69** up by S&P — k1: 1–15 days before 0.45 up by S&P — k2: 16–180 days 0.47** dn by S&P — k1: 1–15 days before − 0.58 dn by S&P — k2: 16–180 days − 0.53* Pseudo R² 21.1%

3.65 19.1 4.25 15.6 2.41 10.0 − 4.09 19.5 − 4.04 14.4 − 3.01 12.7 na na na na − 5.37 39.4 − 4.50 24.9 1.28 8.8 2.72 9.3 − 1.35 10.4 − 2.24 9.7 No. of observations 317

− 6.6 − 6.9 − 4.9 37.2 16.2 14.8 na na 78.8 49.9 − 4.2 − 4.6 10.8 9.3

− 31.7 − 24.4 − 15.2 1.9 12.5 10.6 na na − 31.3 − 8.2 − 13.4 − 13.9 10.1 10.0

15.8 17.7 12.6 − 33.4 − 23.3 − 20.8 na na − 41.1 − 36.2 11.1 11.8 − 16.8 − 15.4

22.5 13.6 7.4 − 5.6 − 5.4 − 4.5 na na − 6.4 − 5.5 6.5 6.7 − 4.0 − 3.9

Panel II: S&P as outlook/watch follower to Fitch actions, Eq. (4) op by Fitch — h1: 1–15 days before 1.03** op by Fitch — h2: 16–180 days 0.57** op by Fitch — h3: 181–540 days 0.85** on by Fitch — h1: 1–15 days before − 0.54 on by Fitch — h2: 16–180 days − 0.36 on by Fitch — h3: 181–540 day 0.23 wp by Fitch — s1: 1–15 days before na wp by Fitch — s2: 16–180 days 1.70 wn by Fitch — s1: 1–15 days before − 1.94** wn by Fitch — s2: 16–180 days − 0.94 up by Fitch — k1: 1–15 days before 0.75** up by Fitch — k2: 16–180 days 0.60** dn by Fitch — k1: 1–15 days before 0.68* dn by Fitch — h2: 16–180 days 0.55 Pseudo R² 11.5%

9.04 18.6 3.58 11.2 6.67 16.1 − 1.92 9.8 − 1.62 6.7 1.05 4.5 na na 1.82 25.2 − 3.15 31.3 − 1.83 15.3 4.18 14.2 3.97 11.8 − 2.40 12.0 − 1.83 10.0 No. of observations 349

− 7.7 − 6.3 − 7.5 11.5 6.7 − 3.1 na − 8.6 62.5 24.3 − 6.9 − 6.4 15.4 11.4

− 29.4 − 16.0 − 24.6 8.2 6.8 − 5.9 na − 41.8 − 19.1 6.4 − 21.6 − 17.1 8.5 8.6

23.3 17.5 22.5 − 18.0 − 12.1 7.5 na 15.3 − 41.0 − 28.5 20.7 18.2 − 22.0 − 18.2

13.9 4.8 9.6 − 1.7 − 1.4 1.4 na 35.1 −2.4 − 2.1 7.8 5.3 − 1.9 − 1.8

This Table reports the results of ordered probit estimations of Eq. (3) and Eq. (4) using data from Fitch and S&P for 10 August 1994 to 31 December 2009. The dependent variables are: yitF in Panel I (Eq. (3)), referring to an outlook/watch status change by Fitch (follower agency) for sovereign i on day t, and yitSP in Panel II (Eq. (4)), referring to an outlook/watch status change by S&P (follower agency) for sovereign i on day t. Four different classes of outlook/watch status changes are employed: − 2, − 1, 1, and 2 (see Table 4). The independent variables are: opi, h, oni, h, wpi, s, wni, s, upi, k, and dni, k (see Table 4 for the definitions). We apply Huber–White robust standard errors. We also estimate and report the impact of each variable on the probability of an outlook/watch status change (marginal effect (ME)). **Significant at 1% level; *significant at 5% level. The estimates of the three threshold parameters are significant at the 1% level in all estimations, and are not shown here. ‘na’: no data available/observed.

a greater extent than vice versa (78.8% versus 62.5%). S&P tends not to follow negative outlook actions and positive watch adjustments by Fitch. The Pseudo R² values are 11.5% when S&P is a follower and 21.1% when S&P is a leader. 6. Conclusion Moody's, S&P and Fitch assigned 97.1% of the ratings reported by NRSROs and employed 90.2% of the credit analysts and supervisors reported by NRSROs at year-end 2010 (SEC, 2011). Using a rich dataset from these CRAs, this paper is the first to compare the behaviour of outlook and watch signals for sovereign issuers across CRAs. SEC (2011) illustrates that 79.4% of the ratings reported from these three CRAs relate to government securities. The current financial crisis brought increased interest in the performance of CRAs, and sovereign rating actions, in particular, have recently attracted considerable attention. The analysis conducted in this paper is particularly important, given that outlook and watch signals are the source whereby CRAs reveal more private information to the markets, and therefore have real economic consequences. We highlight that CRAs employ different policies. S&P has more emphasis on short-term accuracy than other CRAs, while Moody's policy places more weight on stability. S&P reverses its outlook actions much more frequently than Moody's. As matter of policy, S&P tends not to place sovereigns on watch for possible upgrade. The number of positive watch and outlook changes exceeds the number of negative

actions in the case of Moody's, and vice versa in the cases of S&P and Fitch. We use ordered probit modelling to examine (i) momentum in outlook and watch signals, and (ii) the lead–lag relations across CRAs regarding outlook and watch announcements. We find momentum in negative outlook actions, but no evidence of momentum in watch nor in positive outlook signals. There is evidence of strong interdependence among the CRAs regarding outlook and watch actions for sovereign issuers. However, Moody's is a leader in positive outlook and watch actions. It is the most common “first mover”, with S&P and Fitch likely to follow Moody's positive outlook and watch actions within 180 days. Moody's may follow (by placing a sovereign on positive watch, which is a strong signal) positive news released by Fitch and S&P, only in the case of upgrades (not positive outlook or watch adjustments). In contrast, there is a stronger mutual interdependence among the three CRAs for negative outlook and watch actions. This could relate to a strong negative reputational effect for a CRA which is tardy in the case of negative actions. To some extent, S&P shows the strongest lead in negative actions. S&P is the most independent CRA, while Fitch is the most dependent. S&P has the tendency to lead Moody's negative outlook/watch adjustments and both positive and negative Fitch outlook and watch actions to a greater extent than it follows them. Further, S&P tends not to follow negative outlook actions and positive watch adjustments by Fitch. Positive and negative watch actions by Fitch have an insignificant influence on future outlook/watch adjustments by Moody's.

R. Alsakka, O. ap Gwilym / International Review of Financial Analysis 21 (2012) 45–55

Acknowledgment We are very grateful for the comments and suggestions from the anonymous referee and the editor. References Alsakka, R., & ap Gwilym, O. (2009). Heterogeneity of sovereign rating migrations in emerging countries. Emerging Markets Review, 10, 151–165. Alsakka, R., & ap Gwilym, O. (2010). Leads and lags in sovereign credit ratings. Journal of Banking and Finance, 34, 2614–2626. Altman, I., & Rijken, H. (2006). A point-in-time perspective on through-the-cycle ratings. Financial Analysts Journal, 62, 54–70. Bangia, A., Diebold, F., & Schuermann, T. (2002). Ratings migration and the business cycle, with application to credit portfolio stress testing. Journal of Banking and Finance, 26, 445–472. Bannier, C., & Hirsch, C. (2010). The economic function of credit rating agencies — What does the watchlist tell us? Journal of Banking and Finance, 34, 3037–3049. Breitenfellner, B., & Wagner, N. (2010). Government intervention in response to the subprime financial crisis: The good into the pot, the bad into the crop. International Review of Financial Analysis, 19, 289–297. Brooks, R., Faff, R., Hillier, D., & Hillier, J. (2004). The national market impact of sovereign rating changes. Journal of Banking and Finance, 28, 233–250. Cantor, R., ap Gwilym, O., & Thomas, S. (2007). The use of credit ratings in investment management in the US and Europe. Journal of Fixed Income, 17, 13–28. Cantor, R., & Mann, C. (2007). Analysing the tradeoff between ratings accuracy and stability. Journal of Fixed Income, 16, 60–68. Cantor, R., & Packer, F. (1996). Determinants and impact of sovereign credit ratings. Federal Reserve Bank of New York Economic Policy Review October, 1–15. Chambers, J., & Ontko, J. (2007). Outlooks: The sovereign credit weathervane, 2006 update. New York: Standard and Poor's Research (April). Duggar, E., Emery, K., Gates, D., Paulo, S., Lemay, Y., & Cailleteau, P. (2009). Emerging market corporate and sub-sovereign defaults and sovereign crises: Perspective on country risk. Moody's Investors Service February. ECB (2011). Financial Stability Review — June. : European Central Bank. Ferreira, M., & Gama, P. (2007). Does sovereign debt ratings news spill over to international stock markets? Journal of Banking and Finance, 31, 3162–3182. Fons, J., Cantor, R., & Mahoney, C. (2002). Understanding Moody's corporate bond ratings and rating process. Moody's Investors Service May. Fuertes, A., & Kalotychou, E. (2007). On sovereign credit migration: A study of alternative estimators and rating dynamics. Computational Statistics and Data Analysis, 51, 3448–3469. Gande, A., & Parsley, D. (2005). News spillovers in the sovereign debt market. Journal of Financial Economics, 75, 691–734. Güttler, A. (2011). Lead–lag relationships and rating convergence among credit rating agencies. Journal of Credit Risk, 7, 95–119. Güttler, A., & Wahrenburg, M. (2007). The adjustment of credit ratings in advance of defaults. Journal of Banking and Finance, 31, 751–767. Hamilton, D., & Cantor, R. (2004). Rating transitions and default rates conditional on outlook. Journal of Fixed Income, 14, 54–71. Hand, J. R., Holthausen, R. W., & Leftwich, R. W. (1992). The effect of bond rating agency announcements on bond and stock prices. Journal of Finance, 47, 733–752. Hill, P., & Faff, R. (2010). The market impact of relative agency activity in the sovereign ratings market. Journal of Business Finance and Accounting, 37, 1309–1347.

55

Hooper, V., Hume, T., & Kim, S. (2008). Sovereign rating changes — Do they provide new information for stock markets? Economic Systems, 32, 142–166. Hull, J., Predescu, M., & White, A. (2004). The relationship between credit default swap spreads, bond yields, and credit rating announcements. Journal of Banking and Finance, 28, 2789–2811. International Monetary Fund (2010a). The uses and abuses of sovereign credit ratings — Chapter III in IMF Global Financial Stability report: Sovereigns, funding, and systemic liquidity, October, 85–122. International Monetary Fund (2010b). Resolving the crisis legacy and meeting new challenges to financial stability — Chapter I in IMF Global Financial Stability Report: Meeting new challenges to stability and building a safer system, April, 1–61. Ismailescu, I., & Kazemi, H. (2010). The reaction of emerging market credit default swap spreads to sovereign credit rating changes. Journal of Banking and Finance, 34, 2861–2873. Johnson, R. (2004). Rating agency actions around the investment-grade boundary. Journal of Fixed Income, 13, 25–38. Kaminsky, G., & Schmukler, S. (2002). Emerging markets instability: Do sovereign ratings affect country risk and stock returns? The World Bank Economic Review, 16, 171–195. Kim, S. J., & Wu, E. (2008). Sovereign credit ratings, capital flows and financial sector development in emerging markets. Emerging Markets Review, 9, 17–39. Kim, S. J., & Wu, E. (2011). International bank flows to emerging markets: Influence of sovereign credit ratings and their regional spillover effects. Journal of Financial Research, 34, 331–364. Klaar, A., & Riley, D. (2005). Fitch sovereign rating outlook and watch study. Fitch ratings November. Lando, D., & Skødeberg, T. (2002). Analyzing rating transitions and rating drift with continuous observations. Journal of Banking and Finance, 26, 423–444. Livingston, M., Naranjo, A., & Zhou, L. (2008). Split bond ratings and rating migration. Journal of Banking and Finance, 32, 1613–1624. Löffler, G. (2004). An anatomy of rating through-the-cycle. Journal of Banking and Finance, 28, 659–720. Löffler, G. (2005). Avoiding the rating bounce: Why rating agencies are slow to react to new information. Journal of Economic Behavior and Organization, 56, 365–381. Manzoni, K. (2004). Modeling Eurobond credit ratings and forecasting downgrade probability. International Review of Financial Analysis, 13, 277–300. Mathis, J., McAndrews, J., & Rochet, J. -C. (2009). Rating the raters: Are reputation concerns powerful enough to discipline rating agencies? Journal of Monetary Economics, 56, 657–674. Metz, A., & Donmez, N. (2008). Moody's credit transition model: A summary of the watch/Outlook extension. Moody's Investors Service June. Norden, L., & Weber, M. (2004). Informational efficiency of credit default swap and stock markets: The impact of credit rating announcements. Journal of Banking and Finance, 28, 2813–2843. Pukthuanthong-Le, K., Elayan, F., & Rose, L. (2007). Equity and debt market responses to sovereign credit ratings announcements. Global Finance Journal, 18, 47–83. SEC (2011). Summary report of Commission staff's examinations of each Nationally Recognized Statistical Rating Organization. U.S Securities and Exchange Commission. Sundmacher, M., & Ellis, C. (2011). Bank ‘rating arbitrage’: Is LGD a blind spot in economic capital calculations? International Review of Financial Analysis, 20, 6–11. Sy, A. (2004). Rating the rating agencies: Anticipating currency crises or debt crises? Journal of Banking and Finance, 28, 2845–2867. Vazza, D., Leung, E., Alsati, M., & Katz, M. (2005). Credit watch and ratings outlooks: Valuable predictors of ratings behaviour. Standard and Poor's May.