Are credit rating agencies information providers or certifiers? A textual and readability analysis of rating reports

Transcription

1 Are credit rating agencies information providers or certifiers? A textual and readability analysis of rating reports by Florian Kiesel 1 August 24, 2017 We thank Kelly Cai, Sascha Kolaric, Ian Marsh, and Dirk Schiereck for their helpful comments and suggestions. We also thank Jianan He, Paul Monigatti, Thomas Pekar, Maximilian Rust, Patrick Schirmer, Kristin Schreiber, and Simon Schild von Spannenberg for their valuable research assistance. 1 Department of Business Administration, Economics and Law, Technische Universität Darmstadt, Germany, kiesel@bwl.tu-darmstadt.de.

2 Are credit rating agencies information providers or certifiers? A textual analysis of rating announcements Abstract This paper analyzes the informational content of 3,365 Moody s credit rating reports, such as rating changes, watchlist announcements, and outlooks, and their impact on financial markets. We test whether the tone or the length of credit rating report have an immediate impact on the equity or credit default swap (CDS) market and therefore provide new market information. Moreover, we analyze whether credit rating agencies (CRAs) have a monitoring type in financial markets. Our results show that the tone of the rating report has an significant impact on stock returns and CDS spread. This is particularly driven by the negative words in the reports, resulting in negative stock market reaction and increasing CDS spreads. These results document the importance of credit rating reports and the role of CRAs as informational providers. Moreover, the results underline their monitoring function for financial markets. Keywords: Credit rating, Stock market, Credit default swap, Textual analysis, Textual tone, Readability 1

3 1 Introduction Credit rating agencies (CRAs) are well-established institutions in financial markets. Most of bond issuing corporates have at least one issuer rating by a leading CRA, many of them have two or even three. CRAs aim to mitigate problems of asymmetric information between firms and capital market participants. They communicate their assessments of the firm s creditworthiness in credit ratings and detailed reports that are released with the rating update. However, the value added of CRAs has been questioned in recent years. The financial crisis has triggered a debate on the real economic function of CRAs and their credibility. CRAs have faced criticism for primarily adjusting a firm s rating when they detect credit risk changes through the firm s business cycle rather than on the firm s current condition (Löffler, 2004). Yet, there is an ongoing discussion whether rating announcements by CRAs provide new information in a timely manner to financial markets. In this paper, we investigate whether CRAs provide new information to the financial markets or just act as information certifier. Prior studies show that rating action reports have a significant impact on financial markets. The vast majority of previous studies shows a negative equity market reaction following rating downgrades (e.g. Bannier and Hirsch, 2010; Goh and Ederington, 1993, 1999; Hand, Holthausen and Leftwich, 1992; Holthausen and Leftwich, 1986) and that the entity s credit risk increases (e.g. Finnerty, Miller and Chen, 2013; Galil and Soffer, 2011; Hull, Predescu and White, 2004; Norden and Weber, 2004). These studies show that rating reports are important for financial markets, but do not address whether CRAs provide new information to the financial markets. The impact of credit ratings may also be triggered by the certifier role of CRAs. Kisgen (2006, 2009) shows that CRAs are important as information certifier and that their ratings play a crucial role in the firm s financing and capital structure. Galil and Soffer (2011) document that bad news and negative rating announcements tend to cluster, and they therefore argue that rating changes have little informational value for market participants. In order to analyze the information content of rating reports we measure the tone through the frequency of negative and positive words and the capital market reaction depending on the tone. We develop our hypotheses in the spirit of the existing empirical literature, which argues that the tone contains new valuable information (e.g. Arslan-Ayaydin, Boudt and Thewissen, 2015; Henry, 2008; Price, Doran, Peterson and Bliss, 2012; Tetlock, 2007; Kothari, Shu and Wysocki, 2

4 2009). The impact of the tone in announcements is particularly shown for news articles (Tetlock, 2007), corporate filings (Kothari et al., 2009), and earnings press releases (Henry, 2008; Arslan- Ayaydin et al., 2015). In addition, our study relates to Boot, Milbourn and Schmeits (2006), who show in their theoretical model that a CRA s is not limited to the role of an information provider, but includes a monitoring function when new developments potentially affect the firm s creditworthiness. To the extent that CRAs act as information provider, one would expect that financial market reactions are related to the positive and negative information enclosed in the rating reports. Despite its relevance and potential implications, the role of CRAs as informational provider remains relatively unexplored. In this study, we focus our analysis on the credit default swap (CDS) market for two reasons. First, the CDS market is still a comparatively unregulated OTC market, dominated by large institutional investors, such as banks, insurance firms, and hedge funds. It is therefore possible that CDS market participants possess informational advantages that stem from trading in an opaque market and reveal new information faster than the equity or bond markets (Acharya and Johnson, 2007). Second, CDS contracts represent a pure form of default insurance (Finnerty et al., 2013). A firm s CDS spread is the cost to buy protection against the default of the firm. Therefore, CDS spreads reflects the pure default risk of the firm. It is reasonable to assume that if CRAs act as information providers, its impact should be more evident on CDS spreads than equity returns or bond yield spreads, because they are affected by a variety of other factors (Finnerty et al., 2013). In order to contribute to the prior literature, credit rating action reports are obtained from the Moody s website between 2004 and After controlling for illiquid CDS and stock data, 3,365 reports are included in our analysis. The information content of rating action reports is measured with tone. The tone of the rating action report is classified by negative and positive words used in each rating report. Negative and positive words are accordingly to Loughran and McDonald (2011) database and compute as the ratio of negative and positive words to the total number of words in a rating action report. This paper thereby contributes to the ongoing discussion on the economic roles of CRAs, and demonstrates that CRAs act as information providers. We present evidence that the content of rating reports strongly affect the CDS market. The CDS spread increase is stronger for rating reports that contain more negative words. However, positive words do not lead to a reduction of 3

5 the CDS spread. It is also related to the monitoring role of CRAs. Boot et al. (2006) argue that watchlist placements serves as a coordination mechanism for investors beliefs. We empirically show that CRAs play an important role to resolve coordination failures among investors, particularly in the case of deteriorating credit quality. Therefore, we start our analysis where the study of Agarwal, Chen, Sim and Zhang (2016a) ends. They find that negative tone is related to sovereign CDS spread changes. The results indicate that the content of rating reports is important for CDS market investors. In addition, we analyze the equity market reaction. Our results extends the findings of Agarwal, Chen and Zhang (2016b) who show that the tone of S&P rating reports affect equity returns and Löffler, Norden and Rieber (2016) who show that stock returns overreact to the tone of negative change reports. The remainder of this paper is organized as follows. Section 2 presents the related literature of the effect of credit rating action reports on financial markets and develops empirical hypotheses. Section 3 describes the data and defines the key variables used in this study. Section 4 reports our empirical findings. Section 5 offers multiple robustness checks and Section 6 concludes the paper. 2 Related literature and hypothesis development Research on the effect of rating action reports on firms largely focuses on the impact of rating changes on stockholders and bondholders, but neglects the fact that the content of rating reports provided by the CRA differs. Several studies also examine these fields together in order to analyze whether rating announcements lead to wealth transfers between stockholders and bondholders. However, most of the studies analyze the overall effect of rating actions on stock prices and CDS spreads, rather than a textual analysis of the rating content given concurrently with the credit rating actions by CRAs. The vast majority of previous studies finds a negative equity market reaction following rating downgrades (e.g. Bannier and Hirsch, 2010; Goh and Ederington, 1993, 1999; Hand et al., 1992; Holthausen and Leftwich, 1986). The early studies of Holthausen and Leftwich (1986), as well as Hand et al. (1992) and Goh and Ederington (1993) show that rating downgrades are associated with significant negative stock market reactions. In line with stock prices, Hand et al. (1992) also 4

6 document that bond prices show a significant negative reaction to rating downgrades. Yet, they also show that the significant negative market reaction to rating downgrades is driven by rating changes due to poor financial performance of the downgraded firm. Rating downgrades as a result of changes in a firm s leverage are not associated with significant negative stock returns. However, in contrast to rating downgrades, the findings on the reaction to rating upgrades is not conclusive. Holthausen and Leftwich (1986), Goh and Ederington (1993, 1999), as well as Bannier and Hirsch (2010), find no significant equity market reaction to rating upgrades. On the other hand, Jorion and Zhang (2007) and Dichev and Piotroski (2001) document weak positive reactions for stock and bond markets. Furthermore, Jorion, Liu and Shi (2005) observe significant positive market reactions following rating upgrades after the introduction of the SEC Regulation Fair Disclosure in October Goh and Ederington (1993) conclude that rating downgrades provide new information to market participants, whereas rating upgrades do not. In addition to rating downgrades, negative watchlist placements also lead to significant negative stock market reactions (Hand et al., 1992; Holthausen and Leftwich, 1986). Holthausen and Leftwich (1986) document significant positive equity market reactions to rating reviews for upgrade, while Hand et al. (1992) find the opposite to be true, as they find significant negative stock market reactions for negative and positive watchlist placements. Hand et al. (1992) investigate the effect of watchlist placements and rating changes on bond prices. They find that negative watchlist placements lead to significant negative bond market reactions. However, in contrast to the equity market, a weakly significant positive reaction can be observed following rating upgrades. Overall, the results suggest that there is an asymmetric reaction to positive and negative rating actions, as rating downgrades and negative watchlist placements lead to more pronounced market reactions than rating upgrades or positive watchlist placements. A part of the literature on the impact of rating changes on stock prices also examines whether the different reasons for watchlist placements and rating changes lead to different capital market reactions. In their announcement, CRAs give different reasons why a firm s rating is placed on review or why its credit rating has been changed. Goh and Ederington (1993) are among the first to analyze whether the equity market reactions differs to rating change announcements, depending on the reason the CRA gives for its rating decision. They find that only rating downgrades as a result of a deterioration in a firm s earnings or financial prospects lead to significant stock market 5

7 reactions, while rating downgrades due to changes in a firm s leverage or due to other reasons are not associated with significant stock market reactions. Goh and Ederington (1993) consequently conclude that only rating announcements with regard to a firm s financial prospects provide new information to market participants, whereas other rating changes do not and are therefore already reflected in the stock price. For rating upgrades, on the other hand, the CRA s reasoning does not matter. In a later study, Goh and Ederington (1993) confirm that rating downgrades due to deteriorating financial prospects lead to more pronounced equity market reactions, not only on the announcement day, but also during the time period preceding the announcement, particularly for non-investment grade rated firms. Bannier and Hirsch (2010) also note that rating reviews are likely to be triggered by certain corporate events, such as mergers and acquisitions (M&As) or assets sales, or by trends due to a firm s operational or financial performance. The results of their analysis with regard to the underlying reason of a rating review for downgrade or rating downgrade suggests that both, distinct corporate events as well as changes in the firm performance, lead to negative stock returns. The negative market reaction is slightly more pronounced for discrete firm events, if the firm s rating was on review for downgrade prior to the downgrade. Chung, Frost and Kim (2012) likewise distinguish between rating announcements due to discrete firm events, such as M&As and restructuring, and announcements resulting from trends in the issuers operations or financial strength or from changes in the condition of the firm s industry or regulatory environment or the macroeconomic climate in general. In line with Bannier and Hirsch (2010), they find that reviews for downgrade lead to significant negative market reactions, irrespective of the CRA s reason of the review. However, reviews for downgrade as a result of negative financial trends show a more negative equity market reaction than other reasons. It is also noteworthy that rating reviews for upgrade due to distinct corporate events lead to significant positive reactions in the equity market, whereas reviews based on changes in the industry or market environment do not. Agarwal et al. (2016b) is the first study analyzing the tone of credit rating reports and the effect on the stock market. They use the Loughran and McDonald (2011) word list and a Bayesian approach to analyze whether the tone of credit rating reports, defined as the ratio of negative and positive words, influence the stock market participants behavior. The main result is that in particular the negative tone has a significant effect on the stock market. 6

8 For the CDS market, a similar picture emerges. Rating downgrades are usually associated with a significant increase in CDS spread levels (e.g. Finnerty et al., 2013; Galil and Soffer, 2011; Hull et al., 2004; Norden and Weber, 2004), while the effect of rating upgrades is less clear. With the exception of the recent studies of Finnerty et al. (2013), Imbierowicz and Wahrenburg (2013), and Galil and Soffer (2011), prior studies fail to observe a significant CDS market reaction to rating upgrades (e.g. Hull et al., 2004; Norden and Weber, 2004). It is, however, noteworthy that the early studies on the effect of rating changes on CDS markets frequently suggest that negative rating changes are anticipated by CDS markets, at least to a certain extend (Hull et al., 2004; Norden and Weber, 2004). The study of Acharya and Johnson (2007) suggests that this may be due to CDS markets being a preferred channel for informed trading. Hull et al. (2004) document a stronger CDS market reaction to watchlist negative placements than for actual downgrades. In particular, they find a significant announcement effect only during the three day event window surrounding the announcement day for negative watchlist placements, and therefore they argue that specifically the negative watchlist placement contain new information for credit market participants, whereas downgrades and negative outlooks do not. In contrast, Norden and Weber (2004) find a significant negative effect for downgrades for the CDS and stock market. Their results also suggest that there are anticipation effects with regard to negative rating actions in the stock and CDS market. Placements on watchlist negative have a larger economic impact than rating downgrades themselves. This applies to both, the stock and CDS market. Both, Hull et al. (2004) and Norden and Weber (2004), fail to observe any significant CDS market reaction to positive watchlist placements or rating upgrades. Galil and Soffer (2011) also document that watchlist placements, both negative and positive, have a stronger impact than rating changes, thereby giving further support to the results of Norden and Weber (2004). However, they also show that bad news and negative rating announcements tend to cluster, and they therefore argue that in those cases the actual rating changes have little informational value for market participants. Finnerty et al. (2013) conduct an analysis of S&P rating actions between 2001 and In contrast to prior studies, they find a significant CDS market reaction to rating upgrades, but also confirm that downgrades have a greater impact on CDS spreads than upgrades. Rating reviews, on the other hand, are not associated with any discernable reaction in the CDS market. Imbierowicz and Wahrenburg (2013) offer a comprehensive analysis of 7

9 the effects of the reason of rating reviews and rating changes by Moody s on stock prices and CDS spreads. They document a negative effect on CDS and stock markets for reviews for downgrade and downgrades. In addition, their results suggest that rating downgrades due to nearly all reasons have a significant effect on the CDS market, with the exception of downgrades attributable to changes in the capital structure, which is in line with the results of Goh and Ederington (1993). Furthermore, they find evidence for wealth transfers from bondholders to stockholders, which is particularly pronounced if an M&A announcement is the reason for the rating review or change by the CRA. Agarwal et al. (2016a) are among the first who analyze the content of rating reports and the impact on the CDS market. However, Agarwal et al. (2016a) analyze the content of sovereign credit rating reports and its effect on the CDS market. Therefore, they use a Bayesian approach and analyze Moody s rating reports for 62 countries from 2003 to They show that in particular the negative tone is significantly related to an abnormal sovereign CDS spread. Hence, we state the second hypothesis as follows: Hypothesis 2: The content of credit rating action reports, such as credit rating changes, watchlist placements, and rating outlooks, has a significant impact on CDS spreads. While the research on the equity market reaction to rating review and rating change announcements appears to be well established, only few studies focus on the rating watchlist process and the resulting rating change or rating affirmation. Wansley and Clauretie (1985) are one of the first to investigate the impact of the S&P rating review process in its entirety on equity and bond prices. They document a significant equity market reaction only in those cases where the placement is followed by an actual rating change. The announcement of a rating review for downgrade as well as the downgrade lead to significant negative returns, while the positive watchlist announcement and the rating upgrade lead to significant positive returns. For the bond market, a similar picture is observed: negative watchlist placements lead to a significant reduction in bond prices, as do the subsequent rating downgrades. For upgrades, however, there is no observable reaction in the bond market. In addition, Wansley and Clauretie (1985) document that the announcements of rating reviews, which are not followed by a rating change, do not lead to significant market reactions, neither in the equity nor in the bond market. The finding suggests that credit watchlist 8

10 announcements contain sufficient information for capital market participants so that they are already able to differentiate between firms whose rating will be changed and those whose rating will be affirmed. Holthausen and Leftwich (1986) also show that the resolution of a watchlist placement, either through the affirmation of the initial rating or through a rating upgrade, does not lead to significant market reactions. A downgrade following a negative watchlist placement, however, still leads to negative stock market reactions, even though the reaction is less pronounced than for a downgrade without prior review. Boot et al. (2006) construct a theoretical framework in which the rating review process serves as a coordination mechanism for investors beliefs. According to this model, CRAs play an important role as a focal point to resolve coordination failures among investors, particularly in the case of deteriorating credit quality. Therefore, placing a rating on review for downgrade allows CRAs to influence firm s risk choices by threatening firms with a rating downgrade if they are not able to lower their risk exposure within a certain time frame. Thereby, the CRA and the rated firm enter into an implicit contract, which implies that if a firm fails to change its risk exposure, a rating downgrade will take place. This in turn will in all likelihood lead to an adverse investor reaction and increase the refinancing costs of the firm. Thereby, CRAs add a monitoring-type element to financial markets. With regard to positive watchlist placements, Boot et al. (2006) argue that CRAs have little incentive to place firms on watchlist for possible upgrade, as markets will likely have incorporated positive information prior to the CRA making its announcement. This argument is backed by older empirical evidence that shows that markets show little to no reaction to positive watchlist placements or rating upgrades (e.g. Hull et al., 2004; Norden and Weber, 2004). However, more recent evidence shows positive market reactions, in the form of a reduction in CDS spread levels, to positive watchlist placements (Galil and Soffer, 2011; Imbierowicz and Wahrenburg, 2013) and rating upgrades (Finnerty et al., 2013; Galil and Soffer, 2011; Imbierowicz and Wahrenburg, 2013). These results may imply that CRAs also enter into a Boot et al. (2006) style implicit contract for rating upgrades as well. Here, CRAs would engage in a form of positive monitoring where firms are rewarded with a rating upgrade, in case the company continues to perform well and reduce its risk exposure. In this context, a rating affirmation would send a negative signal to market participants that the firm is not able to fulfill the targets set by the CRA. 9

11 Bannier and Hirsch (2010) and Chung et al. (2012) examine the merits of the model proposed by Boot et al. (2006) with regard to reviews for downgrade and rating downgrades. Both find evidence for the validity of the model, but are not able to confirm all of the model s predictions. Bannier and Hirsch (2010) show that rating changes by Moody s following the introduction of its institutionalized watchlist in 1991 lead to stronger equity market reactions than during the years prior to the introduction. They are also able to partially confirm the model of Boot et al. (2006) with regard to the implicit contract between the CRA and the firm, particularly for NIG rated firms. Bannier and Hirsch (2010) conclude that the introduction of the rating review process has enhanced the role that CRAs play in financial markets. CRAs now not only supply and certify information, but they also engage in the active monitoring of a firm s credit risk. Yet, for investment grade rated firms other factors appear to dominate a CRAs decision to place a rating under review. In this case, it appears as if investor demand, rather than risk monitoring, is driving the decision to initiate a watchlist process (Bannier and Hirsch, 2010; Chung et al., 2012). Chung et al. (2012) arrive at similar results to those of Bannier and Hirsch (2010). They also show that the rating review process leads to an implicit contract between the CRA and the rated firm in case of a deterioration of the firm s credit quality. In line with prior results, Chung et al. (2012) find significant negative stock market reactions to rating reviews for downgrade and significant positive stock market reactions to rating reviews for upgrade. In addition, they also document significant negative stock market reactions to rating downgrades, irrespective of whether the downgrade was preceded by a watchlist placement or not. Rating upgrades, on the other hand, do not lead to distinct stock market reactions. This indicates that a CRA s role as information supplier, who helps to facilitate the use of credit ratings in contracting, may be more prevalent than its monitoring role. Our paper thereby significantly contributes to the literature and understanding of the effects that rating actions have on credit markets, the perception of CRAs in capital markets, and their ability to provide valuable and timely information to market participants. Hence, we state the third hypothesis as follows: Hypothesis 3: The content of watchlist placements has a significant impact on the subsequent rating decision of the CRA. 10

12 3 Variable definitions and sample description We now define the dependent variables equity return and CDS abnormal spread change, and the content specific variables used in the tests. In addition, this section describes the sample selection procedure and the final sample of rating reports. 3.1 Measurement of abnormal CDS spread changes We employ a similar empirical set up as Hull et al. (2004), Galil and Soffer (2011), and Finnerty et al. (2013) to measure the abnormal CDS spread changes around rating reports. The observed CDS spread changes are adjusted by changes of a CDS spread index of the same rating class as the company s initial rating: ASC it = (CDS it CDS it 1 ) (I t I t 1 ) (1) where ASC it is the abnormal CDS spread change of firm i on day t, CDS it is the observed CDS spread for firm i on day t, I t is the relevant CDS spread index for the rating class on day t. 1 Daily CDS spread index levels correspond to the equally weighted cross-sectional mean of all CDS spreads for each of the six letter rating classes Aaa/Aa, A, Baa, Ba, B, Caa. 2 We thereby follow the majority of the prior literature (e.g. Galil and Soffer, 2011; Hull et al., 2004) by keeping the index the same as prior to the rating change. This approach better captures any abnormal spread changes as we test the null hypothesis that rating changes have no effect on CDS spread changes, for which it should be assumed that the spread remains adjusted to the old rating Measurement of content specific variables The key independent variable is the lingustic tone. We follow Agarwal et al. (2016b) and Arslan- Ayaydin et al. (2015) and measure the tone of the credit rating report through a content analysis in which the net tone is defined as the spread between the percentage of negative and positive words, 1 We forego the common practice of linearly interpolating the daily mid CDS spreads between missing observations (e.g. Blanco, Brennan and Marsh, 2005; Finnerty et al., 2013; Hull et al., 2004; Norden and Weber, 2004), as singlename CDS have generally become more liquid instruments. 2 Due to the small sample size of companies rated Aaa and Aa are combined into one. 3 This choice in method only affects the calculation for rating changes across letter classes (e.g. Aa3 to A1). Watchlist placements, outloooks, and changes within a letter class are unaffected. 11

13 relatively to the total number of words: NET T ONE i = 100 NEG T ONE i P OS T ONE i T OT W ORDS i (2) where T OT W ORDS i is the total number of words in the rating report of firm i and NEG T ONE i and P OS T ONE i are the number of positive and negative words, respectively. Positive and negative words are accordingly to the Loughran and McDonald (2011) database. As a robustness check, an alternative aggregation of words proposed by Henry (2008) is considered. Besides the tone of the rating report, we also measure the length of the report. The variable W ORDS is defined as the natural logarithm of the total number of words in the rating report. Longer reports include further information about the financial situation of the firm and the new firm situation. This variables tests the hypothesis that CRAs act as informational providers. If the information in rating action reports is a new information, the length of the rating report, and therefore the higher news content, should have an influence on capital markets. The variable CLOSENESS is a dummy variable defined as 1 if the Moody s office released the rating report is in the same country than the firm s headquarter. This controls for the public debate, whether the location of the CRA has an impact on the rating decision. Similar to Agarwal et al. (2016b), we removed the regulatory disclosure before the performing the textual analysis. However, only 36.1% of the rating reports have a regulatory disclosure in the article. The regulatory disclosure is standardized, so we control with a dummy variable defined as 1 if the rating report includes a regulatory disclosure. 3.3 Measurement of other variables The control variables are divided in rating specific and firm specific variables. The rating specific variables consists of DOW NGRADE, UP GRADE, NEG OL, and P OS OL, which are dummy variables indicating the type of action. The base category is watchlist announcements. P rior rating is the Moody s rating prior the rating action announcement. The rating is based on a 20-step numerical scale (Aaa=1, Aa1=16,..., Ca=20). Hite and Warga (1997) show that rating changes from investment grade to non-investment grade have a fatal effect on bond prices. The dummy variable IG GRADE controls for investment grade. In addition, changes from investment grade to 12

14 non-investment grade and vice versa are captured by the two variables F ALLING ANGEL and RISING ST AR. The firm specific variables include T A, defined as the natural logarithm of the total assets of firm in million USD on the last trading day prior to the rating action, DEBT, the ratio of total debt to total assets on the last trading day in the year prior to the rating action, EU, a dummy variable defined as 1 if the firm s headquarter is located in the EU, and V IX, the change of options exchange market volatility index in the month of the rating action. All variables used in our analyzes are defined in Table Measurement of abnormal returns The relationship between the content of rating action reports and stock returns is analyzed using the standard market model event study. The abnormal returns (ARs) of stock j at time t are calculated by: AR jt = R jt ( α β R mt ) (3) where R jt is the market return of stock j on day t, R mt is the Datastream value-weighted national total return index of the country of the event firm, α and β are the regression estimates from an OLS regression using a 252-trading-day (one year) estimation period that ends two day prior to the Moody s rating announcement (t=0). [Place Table 1 approximately here] 3.5 Sample selection This study is based on an international sample of U.S. and European listed firms with available CDS spread data and Moody s long-term issuer rating. The CDS data is retrieved from Thomson Reuters Composite EOD and covers the time period from January 2004 to December Banks, financial services, and insurance companies (SIC ) are excluded due to their unique capital structure. In addition, Moody s announced rating actions for several banks simultaneously in one report. Including their rating action reports may lead to a distortion of our results and should be 13

15 analyzed separately. In total, we were able to obtain CDS data for 530 firms via Thomson Reuters, 504 of which had a Moody s long-term issuer rating. This selection procedure implies that we use the CDS data for all non-financial U.S. and European firms available in Thomson Reuters EOD, resulting in the largest possible sample for this analysis. In a next step, all Moody s press releases were obtained from the Moody s website. In total, 3,487 rating reports between January 2004 and December 2015 were collected. All events for which CDS or stock data is not available in sufficient quality or not available on one of the three days around the rating action report had to be dropped. Table 2 shows the sample selection procedure. The final sample consists of 3,365 rating action reports: 800 downgrades, 578 upgrades, 655 watchlist negative placements, 226 watchlist positive announcements, 483 negative outlooks, and 623 positive outlook changes. [Place Table 2 approximately here] 3.6 Sample description Table 3 provides the summary statistics of the independent and dependent variables. Similar to Agarwal et al. (2016b), there are more negative rating actions than positive actions. The abnormal returns are on average negative and the abnormal CDS spread changes positive, indicating a deterioration in the firm s credit quality. The average net tone is positive, indicating that there are more negative words than positive words in the rating report accordingly to the Loughran and McDonald (2011) word list. The ratio of negative words to the total number of words is on average 5.4%, whereas the ratio of positive words to the total number is is 3.7%. Figure 1 provides an overview of the net, negative, and positive tones across the years in the sample. The percentage of negative words is very steady between 5 to 6%, whereas the percentage of positive tones is between 3 and 4%. During the financial crisis, particularly in the fourth quarter of 2008, the net tone rapidly increased, indicating more negative words in the rating reports. However, in the aftermath of the financial crisis, the percentage of positive words relatively increased. Since approximately the third quarter of 2011, the net tone normalized to the value before the financial crisis. The text length of rating reports varies between 133 and 3,988 words without header, regulatory disclosure and contact details. The average words in a rating action report is without header, 14

16 regulatory disclosure and contact information. 75.8% of all rating reports are released in country of the firm s headquarter. However, this indicates that 24.2% are released from offices outside the firm s headquarter country. This is particularly driven by European firms. [Place Table 3 approximately here] [Place Figure 1 approximately here] 4 Results This section presents the empirical results of the three main hypotheses with respect to the rating action content and its impact on credit markets and the predictability of watchlist decisions. 4.1 The impact of rating content on abnormal CDS spread changes In this section, the impact of the content of rating action reports on the CDS market is tested as proposed in Hypothesis 2. Therefore, we use the CASC [ 1; +1] as dependent variable and repeat the previous regressions. The results for CASC [ 1; +1] as dependent variable are provided in Table 4. Analyzing the impact of textual tone, a similar picture emerges. NET T ONE i is in all three models highly significant and positive. This indicates that the tone of rating reports has a strong effect on the CDS market. More negative words proportional to positive words leads to a significant CDS spread increase. The results show that CDS market participants rely on CRAs in the same way investors on the equity market do and that CRAs have a monitoring role. In Model 4 to 6, we use the percentage of negative and positive words again. The results are similar to the stock market, suggesting that the NEG T ONE is the primary driver of the the NET T ONE variable. The variable NEG T ONE is in all models significant at the 5% level, indicating a CDS spread increase as the number of negative words increases. we only find a weak significant evidence for the opposite effect when we focus on positive words. The variable P OS T ONE is only significant at the 10% in Model 4 and 5 and lacks in significance in Model 6. Positive words do not lead to a reduction in CDS spread. 15

17 Furthermore, the regression results present a highly significant effect for the coefficient of W ORDS. The number of words in a rating report reduces the abnormal CDS spread. If the rating report is longer, this indicates a good sign for the CDS market. This strongly support the hypothesis, that CRAs act as an information provider. The more information CRAs presents in the rating report, the lower is the CDS spread change. Longer rating action reports contain more financial evaluations and recommendations for the firm s management, including possible determinants for a rating downgrade or rating upgrade. CLOSENESS or DISCLOSURE do not affect the CDS spread. IG GRADE has a negative sign and is significant at the 5% level. This indicates that the abnormal spread changes are stronger for firm s that are non-investment grade. This is in line with the findings of Kiesel, Kolaric and Schiereck (2016). They show that investment-graded firms have lower CDS spread changes than firms rated non-investment as the original CDS level is larger for these firms. [Place Table 4 approximately here] On the whole, the content of rating reports strongly affect the capital markets. Stock and market reaction depend on the textual tone of rating reports. Therefore, we can confirm our Hypothesis 1 and Hypothesis 2. We find that, in particular, the percentage of negative words leads to a reduction in the firm equity and an increases in CDS spreads, both indicating a deterioration of the firm s credit quality. The results also support the hypothesis that CRAs have a monitoring role. The informational provider role is particularly pronounced for the CDS market as the number of words and therefore more information has a significant effect on the CDS spreads. 4.2 The impact of rating content on the outcome of a watchlist placement Boot et al. (2006) argue that particularly the credit rating watchlist allows CRAs to extend their traditional role of information providers to one of credit risk monitors. While a firm s rating is being reviewed, the analysts of the CRA collect additional information, which usually involves some form of interaction with the firm s management, in order to obtain more information on the firm s current financial situation. As a consequence, particularly for negative watchlist placements, the firm and 16

18 the CRA enter into an implicit contract, in which the firm can adjust its risk exposure in a timely manner or face a rating downgrade and the ensuing reaction by equity and debt investors. In this section, we focus on watchlist placements and whether the content of the watchlist report has an effect of the outcome of the watchlist process. In order to analyze the outcome of the watchlist, we match all watchlist placements with Moody s subsequent rating decision. The applied methodology reduces the watchlist sample to 823 watchlist placements from 881 watchlist placements in the original sample due to a finalized Moody s decision until the end of the investigation period. The dependent variable is 1 if the outcome of a watchlist placement is a change in the firm s rating and 0 if the prior rating is affirmed. In order to assess whether the content specific variables increase or decrease the likelihood of a subsequent change, a probit model is estimated using the same variables as for the OLS regression. The results are presented for negative and positive watchlist placements separatly. Table 5 presents the probit regression results for negative watchlist placements. The variable NET T ONE increases the probability of a subsequent rating downgrade. The tone of the rating report has a considerably impact on the rating outcome. In Model 4 to 6, the NET T ONE is separated into NEG T ONE and P OS T ONE. Similar to the prior results for the capital market reactions, NEG T ONE is the main driver for the increased probability of a subsequent downgrade. P OS T ONE has a negative sign, indicating that a subsequent rating change is less likely, but the variable lacks in significance. Neither the number of words in the rating report nor the closeness between the firm s headquarter and the Moody s office increase or decrease the probability of a subsequent downgrade. P RIOR RAT ING is weak significant in all models. This suggests that firms with a worse rating are placed faster on the watchlist, but a subsequent downgrade is less likely. In addition, the coefficient of V IX increases the probability of a subsequent downgrade. V IX measures the market expectations of near-term volatility. If the market expectations increase, the probability of a rating downgrade increases in the same way. [Place Table 5 approximately here] The results for positive watchlist placements is shown in Table 6. NET T ONE do not increase or decrease the probability of a subsequent rating upgrade. This result is robust using NEG T ONE 17

19 and P OS T ONE. However, the number of words has a significant impact on the subsequent rating decision. Moody s announce that the firm receives an upgrade if it fulfills the conditions Moody s announces in the watchlist report. If the report is longer, more conditions have to be fulfilled and the probability to meet all expectations is less likely. [Place Table 6 approximately here] Overall, the results for watchlist placements show that the content of the watchlist report has a significant impact on the outcome of the watchlist decision. For negative watchlist the NET T ONE, and in particular the NEG T ONE, increases the probability of a subsequent downgrade. For positive watchlist placements, the number of words used in the watchlist report is highly significant. The results support the finding that the content of rating reports has a significant impact about the firm s future credit quality. 4.3 The impact of rating content on abnormal returns To test the impact of the content of rating reports on the stock market, which was defined in Hypothesis 1, we regress CAR [ 1; +1] on the content specific variable. We include rating specific and firm specific variables to minimize other effects. In Model 4 to 6, we use the variables NEG T ONE i and P OS T ONE i instead of NET T ONE i to analyze the effect of rating action content on abnormal returns and Model 3 and 6 include dummy variables to analyze the effect of the different rating action reports. In addition, Model 2, 3, 5 and 6 include year and industry fixed effects. Table 7 shows the results for the abnormal return regression. The first three models include NET T ONE i as independent variable. The variable is highly significant in all three models, indicating that the abnormal returns are dependent from the tone in the rating report. Rating action reports that contain more negative than positive words lead to stronger negative abnormal returns. This result supports the findings of Agarwal et al. (2016b). They show a similar relation between S&P rating action reports and the stock market reaction. In Model 4 to 6, the variables NEG T ONE i and P OS T ONE i are used instead of NET T ONE i. The results show that primary the percentage of negative words has a significant effect on the abnormal returns. However, the coefficient for the positive tone is also weakly significant in Model 6. The number of words, in- 18

20 dicating further advices for the firm by Moody s do not have an influence on the abnormal returns. The coefficient CLOSEN ESS and DISCLOSU RE also do not impact the abnormal return. The variable DOW NGRADE is highly significant and negative in all six models. This indicates that downgrades have the strongest effect on the abnormal return. This is also evident when we include OL NEG and OL P OS in the regression to control for the different rating actions. In addition, the prior rating has a weak significant and negative influence, indicating that firms with a lower rating are affected more strongly by rating actions. [Place Table 7 approximately here] Overall, the results provide evidence that the equity market reaction strongly depends on the content of the rating report. The results show that in particular the tone of the rating report has a significant effect on the stock market. This supports the theoretical framework of Boot et al. (2006). They argue that CRAs have a monitoring role and the rating report is an important indicator about the future firm s credit quality. Therefore, investors use rating reports to make more informed decisions. 5 Robustness checks In order to verify the results, we achieved in the prior sections, we conduct two robustness tests. The first one is an analysis whether the impact of the tone on stock returns and CDS spreads is robust to the choice of the library used. we will use the Henry (2008) database instead of the Loughran and McDonald (2011) dictionary. For the second robustness test, we use the an alternative methodology for calculating abnormal CDS spreads in order to test whether the results are robust to different empirical approaches. Table 8 shows the impact of the rating action content on the abnormal returns using the Henry (2008) database to define positive and negative words. Table 3 include the descriptive statistics for NET T ONE Henry (2008), NEG T ONE Henry (2008), and P OS T ONE Henry (2008). The average NET T ONE Henry (2008) changed to negative. This indicate that this text analysis identifies more positive than negative words. However, Table 8 shows that the choice of the of the word database does not change the main results for the NET T ONE. NET T ONE Henry (2008) is highly signifi- 19

21 cant and negative. Next, NET T ONE Henry (2008) is again separated into NEG T ONE Henry (2008) and P OS T ONE Henry (2008). The variables are included in Model 4 to 6. Whereas the results for NEG T ONE Henry (2008) is similar to the prior results, P OS T ONE Henry (2008) becomes also highly significant. This supports the prior findings that negative words have a significant impact on the stock market, and positive words at least a significant impact. The results for the other variables are approximately the same. [Place Table 8 approximately here] The impact of the rating action content on the CDS market using the Henry (2008) database is shown in Table 9. The net tone, defined as negative minus positive tone of the Henry (2008) word list, is highly significant. The results of the tone impact on CDS spreads is also robust to the choice of the dictionary. The tone of the rating action reports provide new information to the CDS market participants. The coefficient for the negative tone increases from 1.7 to 2.2, but it is only significant on the 10%. The positive words have now a stronger influence on the CDS market and the coefficient decreases from 1.0 to 2.1. Overall, using the Henry (2008) database confirms the prior results with regard to the tone in rating action reports. [Place Table 9 approximately here] Summarizing, the results are therefore robust to different dictionaries defined the positive and negative words. Furthermore, as an additional robustness test, we use an alternative methodology to calculate abnormal CDS spread changes. Instead of the mean CDS benchmark, we use the median CDS benchmark. This approach has previously been used by Micu, Remolona, Wooldridge et al. (2004); Micu, Remolona and Wooldridge (2006) as well as Galil and Soffer (2011). alternative index calculation procedure minimizes the effect of outliers in the benchmark. This The results of this robustness test is presented in Table 9. This robustness tests largely confirm our prior results that the tone of the rating action significantly impacts the abnormal CDS spread. In addition, the number of words have a significant impact on the CDS spread. The number of words reduce the abnormal CDS spread, indicating an improvement in the firm s credit quality. V IX is also significant, indicating that the spread changes are higher during times of uncertainty. 20

22 Overall, the results of this analysis give again additional support for the previous findings that the content of the rating information has a significant impact for equity and debt market participants. [Place Table 10 approximately here] 6 Conclusion CRAs play a prominent role in financial markets but the empirical evidence still does not arrive at clear conclusions whether rating announcements actually provide new information to market participants. Boot et al. (2006) argue that CRAs potentially take on a monitoring type. This paper analyzes whether credit rating reports contain new information to equity and CDS market participants and and whether CRAs take on a monitoring role. Therefore, a textual analysis of 3,365 rating reports and their impact on the financial markets have been analyzed. The results show that the linguistic tone affect equity and CDS markets in the same way. Negative words result in negative abnormal returns and increasing CDS spreads, while positive tons leads to positive abnormal returns and decreasing CDS spreads, indicating an improvement of the firm s quality. This paper extends the findings of Agarwal et al. (2016b) who show that the tone of S&P rating reports affect equity returns. This paper confirms the prior results for Moody s rating and the CDS market. In addition, this study find that the number of words in a rating report reduces the abnormal spread change. The tone of watchlist reports also impact the final rating decision by Moody s. These results help to understand the economic function of CRAs for the financial markets. we confirm the theoretical model of Boot et al. (2006) and empirically verify that CRAs have a monitoring role and that the information content of rating reports has a significant impact on equity as well as CDS markets. References Acharya, V.V., Johnson, T.C., Insider trading in credit derivatives. Journal of Financial Economics 84, doi: /j.jfineco