Do Credit Watch Procedures Affect the Information Content of Sovereign Credit Rating Changes?

Transcription

1 1 Do Credit Watch Procedures Affect the Information Content of Sovereign Credit Rating Changes? Paula Hill* and Robert Faff** JEL Classification: G32 Keywords: Credit rating, rating change, credit watch, information, sovereign * University of Bristol, UK ( Paula.Hill@bristol.ac.uk) ** Monash University, Australia & University of Leeds ( robert.faff@buseco.monash.edu.au) Corresponding author: Professor Robert Faff Department of Accounting and Finance Faculty of Business and Economics Building 11E Monash University Vic 3800 AUSTRALIA Telephone: robert.faff@buseco.monash.edu.au Acknowledgements: We would like to thank Rob Brooks, Louis Ederington, David Hillier and staff at the Finance Department of the University of Melbourne for their helpful comments and suggestions.

2 2 Do Credit Watch Procedures Affect the Information Content of Sovereign Credit Rating Changes? Abstract Very few studies to date have considered the market reaction to the announcement of credit watch procedures and the impact of credit watch procedures upon the market reaction to credit rating changes. This paper addresses this gap in the literature by focusing on credit watch procedures at the sovereign level. Our primary finding is that re-ratings which follow watch procedures are neither more nor less informative, and we conclude therefore that the credit watch procedure does not impact upon the private information of credit rating agencies. In addition we provide evidence on the types of sovereign state for which credit watch procedures are invoked and of the relative activities at the sovereign level of three major rating agencies.

3 1. Introduction As investment portfolios have become increasingly diversified across national boundaries, research interest has intensified in improving our understanding and assessment of sovereign risk and ratings (see, inter alia, Erb et al., 1996; Harvey and Zhou, 1993). Brooks, Faff, Hillier and Hillier (2004) argue that a key information event, such as a change in a sovereign rating, might instigate substantial re-weighting of international portfolios, particularly if such a re-rating resulted in a change in investment grade status. Accordingly, they examine the impact of sovereign rating changes upon the aggregate stock market, thereby building upon earlier work which examines the impact of individual company rating changes upon stock and bond prices (see, for example, Holthausen and Leftwich, 1986; Goh and Ederington, 1993; Dichev and Piotroski, 2001). A recent paper by Boot, Milbourn and Schmeits (2006) highlights the importance of considering the way in which rating changes come to pass. Ratings are often put under review prior to a change, and during this process, known as a credit watch, the credit rating agency (CRA) is involved in gathering additional information and monitoring the rated firm/government. Fitch Ratings, a leading credit rating agency, indicate that at the corporate level such information gathering would involve dialogue between the CRA analysts and senior executives, while at the sovereign level dialogue would typically be between the CRA analysts and key policymakers and senior representatives of various public sector institutions, such as the finance ministry and the central bank. 1 Kaminsky and Schmukler (2002) and Boot et al. (2006) argue that the announcement of a review process (i.e. a credit watch) and the information gathering involved in this review process, are likely to impact upon the information content of rating changes. To the best of our knowledge, there is only one (limited) study at the corporate level which specifically examines the news content of rating changes conditional upon the presence or absence of a prior review process. 2 Accordingly, our paper aims to address this significant gap in the extant empirical literature. As far as we are aware no prior study examines credit watch procedures at the sovereign level. 1 See page 2, The Rating Process, FitchRatings, July Holthausen and Leftwich (1986). Further discussion of this paper follows.

4 2 Studies at both the corporate level (see, for example, Goh and Ederington, 1993; Dichev and Piotroski, 2001; Purda, 2007) and at the sovereign level (see, for example, Brooks, Faff, Hillier and Hillier, 2004 and Gande and Parsley, 2005) find that credit rating downgrades do provide news to the market, although most studies find that rating upgrades do not. Various explanations have been posited for this asymmetric effect. For example, Holthausen and Leftwich (1986) suggest that CRAs might have an asymmetric loss function, which leads to upgrades being less timely, an hypothesis which is supported by the findings of Kim and Nabar (2007). Conversely, Gande and Parsley (2005) suggest that rating agencies may be less keen to implement rating downgrades, for fear of foreign governments denying access to vital information, and the news element of a downgrade is therefore greater. 3 Holthausen and Leftwich (1986) and Ederington and Goh (1998) suggest that firms/governments may have an incentive to leak positive information to the market prior to a credit rating upgrade, but have no incentive to do so prior to a downgrade. The news element of a downgrade is therefore greater. Kim and Nabar (2007) however present evidence that pre-rating disclosures do not differ across upgrades and downgrades. Ederington and Goh (1998) also suggest that CRAs expend more effort researching negative developments as their reputation relies on identifying credit problems, and this increased effort produces information. Boot et al. (2006) argue that the CRA is only at an informational advantage during the credit watch procedure and the private information of the CRA is released to the public at the conclusion of the credit watch via the rating change. As such, the only rating events which will be informative are those which follow credit watch procedures. Boot et al explain that credit watch procedures are more likely to precede downgrades than upgrades and the observed empirical evidence in relation to downgrades and upgrades in fact reflects the relative proportions of informative events (i.e. those associated with watch procedures) within each sub-sample (i.e. downgrade/upgrade). Conversely, Kaminsky and Schmukler (2002) argue that where sovereign bonds are placed on credit watch prior to 3 This would appear to be a less likely explanation if we assume that news over long periods is biased neither in a positive nor a negative direction, given that sovereign rating changes seem to be equally split between upgrades and downgrades across various sample dates (see Brooks et al., 2004; Gande and Parsley, 2005 and Kaminsky and Schmukler, 2002).

5 3 a rating change, the rating change will be anticipated and thus the rating event will have a lesser news content. Holthausen and Leftwich (1986) specifically examines the information content of re-ratings which are preceded by credit watch additions. Whilst they argue that their results are compromised by small sample size, they tend to find that ratings which occur as resolutions of the credit watch process provide less information than re-ratings which are not preceded by a watch procedure. This finding would tend to support the argument of Kaminsky and Schmukler (2002), that rating changes which follow credit watch procedures are anticipated. Also consistent with this view, Hull et al. (2004) find that additions to the credit watch for a proposed downgrade are informative whereas rating downgrades themselves are not. The aims of our paper are as follows. First, we provide evidence on sovereign credit watch additions which can thus be compared with the findings of broadly counterpart work that has been conducted at the corporate level: Holthausen and Leftwich (1986); Hand, Holthausen and Leftwich (1992) and Hull, Predescu and White (2004). Second, we compare the information content of rating changes which are preceded by credit watch procedures with those which are not, in light of the competing arguments of Boot et al. (2006) and Kaminsky and Schmukler (2002). Third, to provide further insights into pre-, at and post- rating abnormal returns we analyse the types of sovereign issuer likely to be put on watch. Further, we comment on the relative information content reflected in the actions of different agencies. Whilst most studies of the information content of rating events focus on only one credit rating agency, we consider the rating events of the three major agencies. By pursuing the above aims, we contribute to the existing body of work on credit ratings in three major respects: (i) To the best of our knowledge there has not been any prior research into credit watch additions at the sovereign level. (ii) The theory of Boot et al. (2006), being very new, has not been subjected to empirical analysis, 4 and has much to say on the relative information content of 4 This is stated with the proviso that the study of Holthausen and Leftwich (1986), whilst obviously not motivated by the theory of Boot et al (2006) which came 20 years later, did in fact compare the information content of rating changes which follow watch procedures, with those which don t. The analysis of watch- versus non-watch-preceded rating changes in the current paper is wider ranging that that of Holthausen and Leftwich. In addition their study is undertaken at the corporate level, whilst in this paper the analysis is at the sovereign level.

6 4 different rating events. Our study provides evidence on the relevance of their theory at the sovereign level. (iii) There is limited research into the relative information content of different rating agencies. Our findings are easily summarized. First, as reported in prior studies, we observe that positive rating events are non-informative, and our significant findings relate therefore to negative rating events. We show that additions to the watch list for review for downgrading are anticipated by the market. Hull et al. (2004) also conclude that the market anticipates watch negative additions, although they find significant announcement day effects, which we do not. Second, we show that the re-ratings preceded by credit watch events are not more informative. Thus, we reject the explanation put forward by Boot et al. (2006) of the pervasive empirical finding that downgrades are informative whereas upgrades are not. Third, as reported elsewhere for downgrades, we find that credit watchpreceded downgrades have significant negative returns prior to the rating change, 5 however, we also report a new finding, that significant positive returns follow watch-preceded downgrades. Further analysis leads us to suggest that this is unlikely to be due to the types of firm put on watch, but rather that this is a function of the watch procedure itself. Finally, we document further evidence to support the earlier findings of Brooks et al. (2004) that Standard and Poor s and Fitch rating events are informative whereas Moody s are not. We suggest that this differential information effect links to the fact that Fitch and Standard and Poor s are both more timely and more discriminating in their rating behaviour. The remainder of the paper is structured as follows: Section 2 contains a description of the credit rating process and details of both theoretical and empirical research into the information content of announcements by credit rating agencies. This leads us to various hypotheses regarding the information content of sovereign rating events. Section 3 contains a description of the data and an analysis of the differences in rating behavior across three CRAs: Fitch, Moody s, and Standard and Poor s, and an analysis of the types of sovereign issuer put on watch. In Sections 4 and 5 we examine the relative impact of credit watch procedures on the information content of rating events via an event study (Section 4) and multivariate analysis (Section 5). Section 6 provides a concluding discussion. 5 A number of studies report negative abnormal returns prior to rating downgrades. Full details of these studies are provided in the discussion of our results in Section 4.

7 5 2. The Information Content of Sovereign Ratings Events Conditional upon Credit Watch Procedures 2.1 Sovereign ratings and the credit rating process In this paper we consider the rating activity of the three major credit rating agencies, Standard and Poor s, Moody s, and Fitch. Three types of event may be actioned by these CRAs which affect the current credit standing of a given rated entity: a change in the credit outlook, the instigation of credit watch procedures and a change in the rating itself. An outlook takes a longer term view of the credit worthiness of a bond issuer and is typically attached to all ratings. 6 A credit outlook typically covers a period of up to 2 years ahead. Outlooks may be either positive (signalling the possibility that at some stage over the two-year horizon a rating may be raised), stable (a rating is unlikely to be changed) or negative (a rating may be lowered). A credit watch is more short-term focused and is instigated where new developments become known which might affect the rating. For example, Fitch put a rating on review where new information comes to light which may affect the rating and this review of the existing rating is normally undertaken within 2 business days. If a review is not concluded, an entity will be put on credit watch typically as rating watch negative. 7 A company placed on credit watch does not have a ratings outlook during the credit watch period. The duration of credit watch status is typically 90 days. A credit watch is designated either positive (a rating may be raised), developing (a rating may either be raised or lowered but inadequate information is available for this to be currently assessed) or negative (a rating may be lowered). It is important to note that credit watch status need not lead to a ratings change and a ratings change need not be preceded by an entity being placed on credit watch. For corporate bonds, Moody s 6 See Standard and Poor s Primer on CreditWatch and Ratings Outlooks, 08/04/ See page 3, The Rating Process, FitchRatings, July 2006.

8 6 estimate that, historically, between 66% and 76% of all ratings have been changed in the same direction as indicated by the credit watch review. 8 As discussed in the introduction, the rating process involves dialogue between the rated entity and the CRA s analysts. At the corporate level this dialogue would typically involve senior executives and at the sovereign level it would involve key policymakers, senior representatives of various public sector institutions and the central bank. A number of alternative ratings are issued at the sovereign level, namely, country ceilings (for domestic and foreign currency denominated debt); issuer government ratings (for domestic and foreign currency denominated debt); and individual bond issue ratings. 9 We employ the rating attached to the sovereign issuer The information content of credit rating events conditional upon the presence of watch procedures We are aware of three studies, Holthausen and Leftwich (1986), Hand, Holthausen and Leftwich (1992) and Hull, Predescu and White (2004), which extend the work on the information content of rating changes to consider the announcement of credit watch procedures. All of these studies are at the corporate level and their theoretical motivation is simply to determine the extent to which the announcements of credit rating agencies are informative. A finding that credit watch additions are informative implies that CRAs possess private information other than that which is gathered during the credit watch procedure. Boot et al. (2006) argue that credit watch announcements are in response to publicly available information and should therefore, on average, be uninformative. Hull, Predescu and White (2004) examine the relationship between credit default swap (CDS) spreads and additions to the credit watch list and find evidence of significant announcement day 8 See page 7, Understanding Moody s Corporate Bond Ratings and Rating Process, Moody s Investors Service, May The following quote from Moody s literature illustrates the distinction between the country ceiling rating (formerly known as the sovereign ceiling rating) and a government issuer rating: The 12 countries currently comprising the Eurozone share a common currency and so all issuers located in those countries fall under the common Aaa Eurozone ceiling. The individual governments, however, are evaluated for their own fundamental creditworthiness, with ratings on their bonds ranging from Aaa to A2 (page 4, Sovereign Ratings History: Special Comment, Moody s Investors Service, January 2002). 10 Any recovery effort during credit watch procedures is dependent upon the actions of the issuer, and thus it is the sovereign issuer ratings which are key to investigating the theory of Boot et al. (2006).

9 7 effects in relation to watch negatives. Prima facie, this would appear to support the fact that CRAs generate private information outside the credit watch procedure thereby refuting the proposition of Boot et al. that CRAs are primarily at an informational advantage only during the credit watch period. However, Hull et al. employ a sample of investment grade bonds, 66% of which (75 companies) have a rating of A and above. Under the theory of Boot et al. (2006) credit watch procedures are superfluous (i.e. unexpected) for firms of high credit quality 11 and this may explain why the addition of these firms to the watch list is informative. Hull et al. suggest that evidence of even larger CDS spreads for the period prior to the credit watch announcement is evidence that the market anticipates CRA behaviour. However, rating agencies are careful not to take precipitate action which may then need to be revoked, owing to the implications this could have for market volatility. This makes it unlikely that news would induce an immediate reaction on the part of the CRAs. As such, it would be expected that the (instantaneous) reaction of the market to news would precede the actions of CRAs. Hand et al. (1992) find that, on average, credit watch additions generate no news (abnormal returns are not evident at the time of the watch addition). However, once the watch additions are divided into sub-samples of expected and unexpected watch additions, Hand et al. (1992) find that unexpected credit watch additions with indicated downgrade lead to negative excess bond and stock returns. The results for indicated upgrades again tend to show that unexpected additions to the watch list lead to more positive returns, however, the results are less conclusive. The theory of Boot et al. allows for (exceptional) unexpected credit watch announcements, where the CRA has private information about the recovery effort of a firm following a bad signal which has either a worse (better) chance of success than the market anticipated, thus having a negative (positive) affect on security returns. Holthausen and Leftwich (1986) report significant announcement date effects for additions to 11 Under the theory of Boot et al. (2006), the likelihood of a credit watch being instigated is a function of the extent to which asset substitution problems are faced. The incentive to engage in asset substitution is high (low) for low (high) quality firms. Firms of medium quality face multiple equilibria and conditional upon the financing cost faced, such firms will optimally select either a viable project or a high risk project, where the NPV of the viable project is higher than that of the high risk alternative. It is for these firms of medium quality that the watch procedure might provide a valuable role in steering the firm towards a good equilibrium. Thus watch procedures are expected for firms of medium credit quality with medium chance of recovery following a bad signal.

10 8 the watchlist for both upgrades and downgrades, which suggests that CRAs do generate private information other than during the credit watch procedure. Turning to the information content of re-ratings, Holthausen and Leftwich (1986) is the only study to date to examine the relative information content of watch-preceded and non-watch-preceded downgrades. They find that watch-preceded downgrades provide less information to the market than non-watch-preceded downgrades. This evidence is contrary to the theory of Boot et al (2006) but supportive of the theory of Kaminsky and Schmuckler (2002). Examination of the pre- and post-announcement impact of rating actions by credit rating agencies allows comment to be made upon the extent to which CRA actions are anticipated and the efficiency with which any new information is delivered to the market. The pre- and postannouncement returns may be affected by the credit watch procedure and/or the types of entity put on credit watch. Changes in the credit rating alter the risk profile of a sovereign issuer and that of firms located within the domestic market, and it would be expected that rating upgrades are followed by lower returns and rating downgrades by higher returns. Failure to adjust for this change in the risk profile would show up as positive abnormal returns following a downgrade and negative abnormal returns following an upgrade. Consistent with this, Brown et al. (1993) find that positive (negative) events which decrease (increase) the volatility of the returns of a firm provide negative (positive) CARs following the event. It is a reasonable assumption that sovereign issuers placed under credit watch might be characterized by information asymmetry, the watch procedure thus allowing the CRA a period over which to gather further information before making a rating decision. Zhang (2006) shows that greater stock return momentum would be expected in the presence of information asymmetry. As such, speculation regarding an imminent rating change of sovereigns characterized by greater information asymmetry might be expected to cause prices to deviate from fundamentals thus leading to larger preannouncement returns than for sovereigns which are associated with lower levels of information asymmetry. In turn, such deviations from fundamentals are likely to see a reversal of returns in the aftermath of the rating change announcement. Under the theory of Boot et al. (2006), credit watch

11 9 procedures are most valuable for medium quality firms with medium chances of recovery following a bad signal. Since credit quality and information asymmetry are likely to be negatively related, Boot et al do not predict that credit watch procedures are likely to be associated with sovereigns suffering from higher levels of information asymmetry. In summary, the extent to which credit rating changes convey new information and therefore impact upon asset returns is the subject of ongoing debate and investigation. Rigobon (1997); Radelet and Sachs (1998); Ferri, Lin and Stiglitz (1999) and Reinhart (2001 and 2002) find that ratings changes are procyclical which would suggest that they provide only a limited quantum of new information to the market. A number of authors we have cited above contradict this theory by providing evidence that the actions of credit rating agencies are informative. Boot et al. (2006) contribute to this debate by suggesting that credit rating agencies only generate private information during credit watch procedures. Existing empirical evidence on the information content of credit watch additions at the corporate level is inconclusive. Our first hypothesis allows us to make a comparison with the results at the corporate level of Holthausen and Leftwich (1986); Hand et al. (1992) and Hull et al. (2004): H1: If credit rating agencies are only at an informational advantage during credit watch procedures, there should be no stock market reaction to the announcement of such procedures. Kaminsky and Schmukler (2002) suggest that rating changes which follow watch procedures will be less informative since they are expected. Conversely, Boot et al. (2006) argue that rating changes will only be informative when they follow a watch procedure since it is at the conclusion of the watch procedure that CRAs release their private information to the market. Accordingly, our next two hypotheses are: H2: If credit rating agencies are only at an informational advantage during credit watch procedures (CWP), there should be no stock market reaction to any rating event which does not follow a CWP. H3: If credit rating agencies are at an informational advantage during credit watch procedures (CWP), and that this private information is released at the conclusion of the watch procedure, rating changes subsequent to CWPs should induce a market reaction. The implications of H2 and H3 taken together are that rating upgrades and downgrades which follow watch procedures will be more informative than those which don t.

12 10 A credit watch procedure need not lead to a change in the existing rating and indeed many watch procedures end with an affirmation of the previous rating. Only Boot et al. (2006) derive a theoretical explanation of the information content of affirmations, and they argue that they will be informative, for the same reason that rating changes which follow a credit watch procedure will be informative. Since Boot et al. (2006) only predict watch procedures will occur prior to downgrades we relate our fourth hypothesis specifically to downgrades. H4: Upon completion of credit watch procedures designated downgrade, security prices should react positively to ratings affirmations (i.e. confirmation of the previous rating). 3. Differences in Rating Activity across Credit Rating Agencies To test our hypotheses we employ government (sovereign) issuer ratings in respect of foreign currency denominated debt. We employ the ratings of Fitch, Standard and Poor s and Moody s and the ratings applied by these CRAs are shown in Table 1. Our period of analysis extends from January 1, 1990 through June 30, Table 2 provides summary statistics in relation to both rating changes and credit watch procedures across all 94 countries for which there was either a credit watch or rating change over our sample period. 13 Two features stand out from Panels A and B of Table 2. The first is that at the sovereign level Moody s employ credit watch procedures to a much greater extent than either Standard and Poor s or Fitch. This suggests that Moody s are prepared to commit considerably more resources to this activity than their competitors or that they devote less effort to each watch procedure (though the latter could lead to a potentially harmful credibility gap for them in a competitive market setting). The second notable feature is that Fitch watch negatives outweigh watch positives, and Standard and Poor s watches are virtually exclusively negative (they instigate only one non-negative case: a developing 12 It should be noted that Fitch did not rate sovereign issuers until August 10, Also, Moody s did not make use of credit watch procedures prior to the 1990s and there was very little sovereign re-rating activity prior to This lack of re-rating activity reflects the fact that sovereign ratings were only issued on a relatively select group of countries with higher credit ratings and few re-ratings were required. 13 All data relating to sovereign issuer credit ratings, credit watch procedures and ratings affirmations were provided by the rating agencies, Standard and Poor s, Moody s and Fitch Ratings.

13 11 watch procedure prior to an upgrade), however, Moody s watch procedures are equally common for rating upgrades as downgrades. Panel C of Table 2 reveals that watch procedures last for an average of 77 days (median = 62 days). For Moody s watch negative procedures consume about the same length of time as watch positive procedures at 70 days on average, but for Fitch watch negatives take nearly twice as long as watch positives (104.4 days versus 56.5 days). Across all CRAs the duration of the watch procedure is about twice as long before an affirmation, at an average length of days (median = 91.5 days), than before a rating change, when a credit watch lasts on average for 65.7 days (median = 56 days). Forty-five countries were re-rated by all 3 agencies during our sample period. 14 There were a total of 549 re-ratings across these countries of which 216 were by Standard and Poor s (192 post August 10, 1994, the date at which Fitch first issued sovereign ratings); 165 by Moody s (146 post August 10, 1994); and 168 by Fitch (all post August 10, 1994). It would appear that Moody s is slightly more reluctant to re-rate countries than either Standard and Poor s or Fitch. We also examine the extent to which agencies tend to lead or lag each other in terms of their ratings over the common 12-year period. Specifically, a rating is defined as the first in a series where no rating of the country has preceded the rating change within the previous 3 months, and another rating agency follows suit within 3 months. There were 95 re-rating events which can be defined as first in a series. Of these 95 cases, 43 re-ratings (45%) were by Standard and Poor s, 31 (33%) by Fitch and 21 (22%) by Moody s. This confirms a similar observation made in relation to sovereign ratings by Brooks et al. (2004) and Gande and Parsley (2005). We argued that if sovereigns put on watch are characterised by information asymmetry, this may affect the return pattern around CRA announcements. We examine the probability of a firm being put on watch dependent on its initial credit quality, with the latter being a proxy for information asymmetry (other suitable proxies, such as the size of the capital market, are not available for many of the rated countries). In view of our finding that Fitch and Standard and Poor s tend to employ watch procedures prior to downgrades, we undertake this analysis employing rating downgrades, of which 14 In this and the following paragraph we refer to untabulated descriptive information regarding these 45 countries to further examine relative rating behaviour.

14 12 there are 272 cases across our sample. Firms are divided into quintiles based upon their initial credit quality, with Quintile 1 firms suffering from the lowest credit quality, and by inference, from the highest information asymmetry. The results of this analysis are shown in Panel A of Table 3. These results provide no evidence to support the hypothesis that watch procedures are employed more frequently where sovereigns suffer from greater information asymmetry. The relative frequency of a watch procedure preceding a downgrade rating event is highest for medium quality sovereigns (Quintile 3) and lowest for high (Quintile 5) and low (Quintile 1) quality sovereigns. In the case of Moody s, whilst watch procedures are most likely to be implemented for medium quality cases - Quintile 3 - the probability of a watch procedure remains high for Quintile 5. Chi squared tests confirm that the proportion of low credit quality sovereigns (Q1) put on watch is significantly lower than medium quality sovereigns (Q2 through Q4) for Moody s and the proportion of high credit quality sovereigns (Q5) put on watch is significantly lower than medium quality sovereigns for Standard and Poor s. Evidence above suggests that Moody s credit watch procedures may differ in key ways from Standard and Poor s and Fitch. We therefore repeat this analysis employing only Standard and Poor s and Fitch data. We restrict our analysis to 107 downgrades where we have access to alternative proxies for information asymmetry, namely a sovereign corruption index (Corruption) prepared by Transparency International - Corruption takes a greater value, the lower the level of corruption - and the size of the stock market in US dollars at the beginning of the year (Market Cap), sourced from Datastream. 15 The results are presented in Panel B of Table 3. There is no evidence that sovereigns suffering from the greatest levels of information asymmetry (Quintile 1) are more likely to be put on watch, but there is, however, evidence that sovereigns suffering from the lowest levels of information asymmetry (Quintile 5) are less likely to be put on watch. Specifically, a chi-squared test confirms that the proportion of sovereigns of the highest credit quality put on watch is significantly lower (at the 1% level) than sovereigns of medium credit quality (Quintiles 2 through 4). This finding is compatible 15 A number of studies also employ stock return volatility to capture information asymmetry (see, for example, Zhang, 2006) however we cannot consider volatility to be an untainted measure of ex ante information asymmetry given that any ratings events which occur during the period in which we measure volatility may cause a change in volatility.

15 13 with the predictions of the Boot et al. (2006) who suggest that watch procedures are most likely for entities of medium credit quality. 4. Information Content of Ratings Events: Event Study Findings Boot et al (2006, page 101) argue that ratings and credit risk are obviously related and suggest that the information content of CRA announcements are best evaluated by examining their impact on security prices. Bond data are not available for many sovereigns and infrequent trading is a problem where such data is available. Further, Holthausen and Leftwich (1986) conclude that the signal to noise ratio may be more favourable for stock than bond data (page 59). As such, we follow Brooks et al. (2004) and we employ stock market returns to analyse the informativeness of sovereign CRA announcements, which has the significant advantage of greatly expanding the feasible sample due to the much greater availability of this data compared to non-equity data. 16 Daily market return indices for each country and for the world index (denominated in US dollars) were obtained primarily from Datastream (Thomson Financial). However, for certain countries a market return index denominated in US dollars was not available via Datastream in which case the following alternative sources were variously employed: S&P/IFCF, The Bank of New York, FTSE Group, and Morgan Stanley Capital International. We calculate the average cumulative abnormal returns (CAR) over a 2-day (0, +1) and 3-day (-1, 0, +1) window on the stock indices of each ratings event country. 17 We employ standard event study methodology. Daily abnormal returns (AR) in country i are derived using both the (i) market 16 Holthausen and Leftwich (1986) point out that if a rating downgrade (upgrade) is driven by a change in the cash flows of the firm then the value of equity will be negatively (positively) affected. However, if a rating change occurs owing to a change in the variance of the firm s cash flows, ceteris paribus, via option pricing theory, a downgrade may lead to an increase (decrease) in the value of equity. As with other studies we are unable to identify whether it is changes in expected cash flows or volatility (or both) which lead to rating changes. Prior studies which report a significant effect find that on average rating downgrades have a negative impact on share returns. 17 Gande and Parsley (2005) suggest minimising the event window to reduce spillover effects of rating changes of related countries (and in our model, the rating activity of other agencies) and accordingly select a 2-day window. A 3-day window is suggested by Jorion et al. (2005) to allow for announcement date uncertainty.

16 14 model and (ii) an index model (i.e. the raw return minus a world market index return). 18 The market is proxied by the Thomson Datastream World index. The market model is estimated using 200 daily return observations, t= -220 to t = -20. Average cumulative ARs (CARs) are calculated for each of 5 portfolios relating to different events: (i) watch negative announcements (H1), (ii) watch positive announcements (H1), (iii) rating downgrades watch vs. non-watch (H2 & H3), (iv) rating upgrades watch vs. non-watch (H2 & H3), (v) rating affirmations following a watch negative (H4). 19 Following Boehmer et al. (1991) we base our test statistics on the standardized abnormal return. Since some countries will be re-rated/put on watch by agencies within quick succession, we omit any same country events where there is an overlap within the event window, leaving only the first event in the sample thus, we achieve an uncontaminated sample. 20 For those cases in which a watch event occurs on the same day as a rating change or affirmation, since theory tells us that such events are unlikely to be economically significant, we omit them from our watch event analysis, but retain the rating changes and affirmation in their respective event study analyses. 21 After these adjustments there are 57 watch positives, 66 watch negatives, 160 rating downgrades, 246 rating upgrades and 26 affirmations following a watch negative across 44 countries for which stock index data are available over our sample period. Full details of the final sample are displayed in Table 4. The results of our event study analyses are shown in Tables 5 to 7, as follows: Table 5 (Watch Events), Table 6 (Rating Downgrades), and Table 7 (Rating Upgrades). 4.1 Market reaction to watch announcements In Table 5 we find that while the returns on the date of watch positives (negatives) are positive (negative), as expected under H1 neither case produces a significant impact upon stock market returns. 18 There are problems with the use of the world market index to replicate the market portfolio where national markets are segregated. Many emerging markets may have a low correlation with the world market portfolio simply because of market frictions. In addition, some emerging markets are relatively illiquid. However, in mitigation, the markets with more severe illiquidity and friction problems tend to be those markets for which no stock index data is available and are thus excluded from the event study analysis. 19 There were only 5 rating affirmations following a watch positive. 20 Where events occur simultaneously (5 cases) we deselect one of each pair of the rating downgrades to avoid double counting. Gande and Parsley (2005) point to the fact that in the event that spill-over effects are economically significant, studies of same country effects will be flawed where they fail to take account of spill-over effects in the event window being assessed. We further allow for the impact of spill-over effects in our multivariate analysis. 21 In all cases the rating change and watch event were in the same direction.

17 15 The significant negative abnormal returns which we report prior to a watch negative, (-10, -1), can plausibly explained by either the market anticipating the actions of CRAs (or alternatively preannouncement leakage), or with both the market and CRAs reacting to news, with the market reaction coming first. This latter scenario is consistent with our earlier argument that CRAs are keen to avoid precipitate action which may then need to be revoked, owing to the implications for market volatility. Since there are no positive abnormal returns shortly prior to watch upgrades, this might suggest that CRAs are slower to implement upgrades leaving a greater gap between any good news signal and the watch positive event. As discussed earlier, Kim and Nabar (2007) find evidence to suggest that downgrades are timelier than upgrades. These results at the sovereign level can be compared with those at the corporate level of Holthausen and Leftwich (1986), Hand et al. (1992) and Hull et al. (2004). Holthausen and Leftwich (1986) report significant announcement date effects for additions to the watch list for both indicated downgrades and upgrades. They also report some evidence of significant pre-announcement date effects for watch negative announcements. Hand et al. (1992) find that, on average, credit watch additions generate no abnormal returns at the time of the announcement, however, Hull et al. (2004) find evidence of significant announcement date effects and even larger pre-announcement date effects in relation to watch negatives (but not positives). We therefore report similar pre-announcement date effects to Holthausen and Leftwich (1986) and Hull et al. (2004), but since we do not find significant announcement date effects, in this regard our findings support those of Hand et al. (1992). 4.2 Market reaction to rating downgrades Turning to Table 6, in Panel A we report that returns are larger, (almost twice as large across the 3-day window) and significantly more negative for non-watch-preceded downgrades however, a t-test of the difference in mean CARs across watch and non-watch-preceded downgrades fails to be significant across any window. The Panel A results do not support the predictions of H2. The implications of H2 and H3 taken together are that rating upgrades and downgrades which follow watch procedures will be more informative than those which don t. It remains plausible that the CRAs release information during the credit watch procedure which would be expected to produce significantly larger cumulative

18 16 abnormal returns over a longer window, however, in this case a significant reaction to the rating change itself would not be expected since this would have been anticipated. With reference to the 3-day window results, only downgrades which are not preceded by watch procedures lead to negative CARs which are significantly different from zero. Kaminsky and Schmukler (2002) argue that where sovereign bonds are placed on credit watch prior to a rating change, the rating change will be anticipated and thus the rating event will have a lower news content. However, our results cannot be explained by the surprise element associated with non-watch downgrades, since the returns are significantly negative before the downgrade. The existence of significant pre-event abnormal returns for both watch and non-watch preceded downgrades is consistent with the findings of Wansley and Clauretie (1985), Steiner and Heinke (2001) and Purda (2007). However, these authors have provided no explanation of what may be driving these significant pre-event returns. Explanations include the possibility that the negative returns are due to the bad signal which precedes a rating event, or that pre-announcement speculation and/or information leakage/dissemination occurs. Given that the watch procedure, with an average duration of 77 days (see Panel C of Table 2), would take place between the bad signal and the rating change, and given that the negative returns in the 10 days prior to the rating change are equally significant for ratings preceded by a watch procedure, the pre-announcement leakage/dissemination and/or speculation explanation would appear to be a more credible explanation. A further notable finding revealed in Panel A of Table 6 is that sovereign debt under watch negative procedures are associated with significant negative returns prior to and at the rating downgrade, but then a reversal to significant positive returns in the period following the rating downgrade. Notably this reversal is equivalent in magnitude to the previous negative abnormal returns. No such reversal takes place for sovereigns who were not under credit watch. We provide further discussion and analysis of this finding at the end of Section 4. Brooks et al. (2004) examine sovereign rating changes for the period up to July 31, 2001 and find that while Fitch and Standard and Poor s downgrades lead to significant abnormal returns on the event day, Moody s downgrades do not. In Panels B and C of Table 6, we therefore repeat our analysis across two separate sub-samples, one for Moody s and one for combined Fitch/Standard and Poor s.

19 17 We confirm that Fitch/Standard and Poor s downgrades have a significant impact on stock returns, whereas Moody s downgrades do not. The observation made in Section 3, in which we find that Moody s have a lesser tendency to lead sovereign rating activity supports this finding. The fact that Moody s are also less active in undertaking sovereign re-ratings (see Section 3) may also suggest that they are more likely to miss the boat. We find that the pre-event returns for the Fitch/ Standard and Poor s sub-sample are significantly greater in the case of watch preceded downgrades, but also that the reversal effect mentioned above is concentrated in Fitch/Standard and Poor s downgrades. If CRAs disseminate information during the watch procedure rather than at the end of the watch procedure, then cumulative returns will be larger for watch preceded downgrades over the pre-event window, however this does not explain the reversal. 22 Further, across the pre-event, event and post-event windows, (-11 (max), +11 (max)), abnormal returns are smaller for watch preceded downgrades, rather than significantly larger, suggesting that the credit watch process does not involve a release of more information to the markets by the credit rating agencies. 4.3 Market reaction to rating upgrades In Table 7, we report that in line with previous studies there is no reaction to ratings upgrades. As such, this further undermines H2 and H3. The one significant result is that there are negative returns following rating upgrades which are associated with watch procedures, which mirror the reversal phenomenon (positive returns) following downgrades. Given the pattern of evidence found in relation to downgrades, it would be interesting to analyse whether this result holds for the Fitch/Standard and Poor s sub-sample, however, such analysis is disadvantaged by the relatively small number of Fitch and Standard and Poor s rating upgrades associated with a watch procedure. Thus, with caution, we present the results in relation to Fitch upgrades in Panel B of Table 6 and find that the negative returns following the rating upgrade are twice as large (and significant at the 10% level) once Moody upgrades are excluded We provide further discussion and analysis of this finding later. 23 There were no watch positives associated with upgrades for Standard and Poor s.

20 18 Our finding that rating upgrades are generally non-informative is supported elsewhere in the empirical literature. Even where watch procedures are present, rating upgrades fail to be informative. However, the majority of watch positives are associated with Moody s and it is possible that since Moody s are less discriminating in implementing watch procedures, fewer resources may be expended on each of these activities than is the case for Fitch and Standard and Poor s. In summary, since total cumulative abnormal returns over the pre-event, event and post-event windows are roughly the same for watch and non-watch preceded downgrades, it raises doubt whether CRAs have a greater informational advantage during the credit watch process. It should be noted however, that by stating there appears to be no increase in the informational advantage of CRAs during a watch procedure, we are not stating that CRAs are not at an informational advantage. Our findings lead back to the theories of Ederington and Goh (1998) and Gande and Parsley (2005) in which either CRAs expend more resources detecting rating downgrades, or alternatively, the theory of Holthausen and Leftwich (1986) in which downgrades are more timely. 24 Given that the market reaction to downgrade announcements is stronger than to credit watch negative announcements, it would seem that credit rating agencies expend more resources in gathering information prior to a rating change, as might be expected. 4.4 Market reaction to affirmations following watch negative events We examine rating affirmations which follow a watch negative. In unreported results, we find evidence of significant positive returns prior to the affirmation (CAR = 0.08 for the 10-day period (- 10, -1), significant at the 1 % level), but no response to the affirmation itself. Three possible explanations are that news of the affirmation leaked out (or was disseminated to) the market before the official announcement of the rating confirmation, or that good news which arrived was followed by an inevitable affirmation, or that once watch procedures are initiated, since affirmations on average take twice as long as rating changes, the market begins to anticipate an affirmation since a rating change has not been observed. Again this suggests that if the CRA does generate information during the watch 24 As discussed previously, Kim and Nadar (2007) present evidence at the corporate level which suggests that firms do not leak more information to the market prior to an upgrade thereby making downgrades more informative.

21 19 process, this is released to the market during the watch process and not at its conclusion. Nevertheless, our findings fail to support H Stock return reversals and credit watch procedures In this sub-section we further consider our finding that where a rating downgrade is preceded by watch procedures, stock returns are more negative prior to the rating change and more positive following the rating change, than the returns on the markets of sovereigns which undergo no watch procedures. Panel C of Table 6 shows that the average CAR for the watch preceded rating downgrades of Fitch/Standard and Poor s for the combined pre-event/event window (-10, +1) is %, compared to +8.23% for the 10-day post-event window (+2, +11), whilst the counterpart average CARs for nonwatch rating downgrades are -4.82% and +1.65%, respectively. The correlation between (i) combined pre-event and event period abnormal returns (-10, +1) and (ii) post-event period abnormal returns (+2, +11), across the sample of 109 Fitch/Standard and Poor s rating downgrades is -28.0% for watch preceded rating changes, compared to only -2.2% for non-watch preceded rating changes. 25 Following, Michayluk and Neuhauser (2006), we compare the extent of reversal according to the initial response to the rating downgrade (i.e. the CAR in window (-10, +1)). 26 Specifically, we partitioned the sample into quintiles based on the size of the initial CAR. Statistical tests confirm that CARs for sovereigns with the most negative initial response to the rating downgrade (Quintile 1) had significantly more positive post-event returns than other sovereigns (i.e. Quintiles 2 through 5). The incidence of watch procedures is also higher across those ratings which exhibited greater negative CARs in the initial response window (-10, +1). Reversals would seem to be a feature of rating changes associated with watch procedures. We suggested in Section 2 that changes in the credit rating alter the risk profile of a sovereign issuer and that of firms located within the domestic market, such that it would be expected that rating 25 These are the figures employing the market model and a 2-day event window. Correlations using alternative variations discussed earlier in the text are similar. 26 For this purpose we remove one outlier observation which had a post-event return of 94%. This related to a downgrade of the Dominican Republic by Standard and Poor s which was not preceded by a watch event. Interestingly, the pre-event CAR was also positive, however, the market subsequently declined in the longer term, suggesting that in this case Standard & Poor s led market opinion.