Regulating rating agencies: A conservative behavioural change

We investigate whether the European regulatory reforms of the credit rating industry have been successful in improving the quality of financial institutions’ credit ratings. A shift to more conservative rating behaviour rather than rating quality improvement is identified, which is attributable to increased regulatory scrutiny. This change leads to a reduction in rating inflation and an increase in the number of unwarranted downgrades and false rating warnings in the post-regulatory period. A significant decrease (increase) in the informativeness of rating downgrades (upgrades) is evident. Our findings contrast with prior evidence for US corporates where reputational effects dominated.


Introduction
The US sub-prime crisis led to increased public and regulatory scrutiny of the quality of ratings issued by credit rating agencies (CRAs) (e.g.Bae et al., 2015, Flynn andGhent, 2018).High quality ratings are vital for the proper functioning of the financial system, given that credit ratings are heavily used by regulators, debt issuers, investors and financial institutions (Becker and Milbourn 2011;EC 2016;Jackowicz et al. 2020).In response to the sub-prime crisis, the EU acted promptly to establish new regulations for CRAs operating in Europe.The key aim of this study is to investigate the impact of the EU regulatory reforms on the quality of ratings.We focus on two dimensions of rating quality: (i) the ability of ratings to classify risk, and (ii) the ability of ratings to transfer information to market participants.Ratings that can correctly classify the future probability of defaults and are closely correlated with current market prices fulfil their expected functions.Inflated ratings (overstatements of creditworthiness) mislead the market about the true financial condition of a debt issuer.It is now evident that inflated ratings (especially in structured finance products) were prevalent prior to the global financial crisis, with the most notable example being Lehman Brothers' AAA rating months before its financial collapse.Steps to discourage rating inflation could therefore potentially enhance ratings quality.However, the increased regulatory scrutiny, liability and penalties could induce more conservative rating behaviour (Bannier et al., 2010).
The initial stage of EU CRA regulation was established in September 2009 (No 1060/2009, known as CRA I) and sought to address conflicts of interest in the rating process by requiring comprehensive disclosures by CRAs of their rating models, historical performance and annual transparency reports.In July 2011, the newly created European Securities and Markets Authority (ESMA) assumed responsibility for supervising and certifying CRAs operating in the EU (CRA II).ESMA sought to mitigate mechanistic reliance on credit ratings by market participants, and thereby reduce the potential for market overreactions to rating actions (EC, 2014).These regulatory reforms mark a shift from the pre-crisis scenario of CRA selfregulation and towards stringent regulation enforced by ESMA.Prior to this, the scope for legal and regulatory fines on CRAs was much more limited and no entity had direct responsibility to ensure that the regulation was implemented.This is the most significant factor that should contribute to a decrease in rating inflation.The May 2013 regulatory update (CRA III) strengthened the regulation with the instigation of a new civil liability regime and expansion of the transparency and monitoring requirements.Overall, the key aims of the regulation are to increase the quality of ratings by reducing rating inflation, to increase the informativeness of rating upgrades, and to reduce mechanistic market reactions to rating downgrades.This paper contributes to knowledge in many respects.Firstly, while previous related studies have focused on the impact of US regulatory reforms on corporate ratings and structured finance ratings (see Section 2), this paper fills a significant void in the literature regarding both the impact of the regulatory changes on the financial institution (FI) rating segment and in the European setting.Secondly, this study furthers the debate surrounding the most appropriate mechanisms for regulating CRAs in the future.Third, our paper investigates whether the EU regulatory reforms have achieved their stated objectives.Fourth, it sheds light on the question, initially raised by Baghai et al. (2014), of why CRAs have become increasingly conservative.Finally, our paper reveals how FI ratings have been affected in recent years, given their pivotal role before and during the global financial crisis.FIs are somewhat opaque and subject to a range of different risks, which make them more difficult to rate by CRAs compared with firms in other industries (Flannery et al., 2013;Morgan, 2002). 1 This study provides evidence on FI ratings behaviour in response to changes in CRA regulation, an aspect which is neglected in the earlier literature.Our sample includes ratings from the largest three CRAs (Moody's, S&P and Fitch) for 758 FIs across 27 European countries during the period January 2006 to June 2016.
Three hypotheses on the impact of the regulatory change on credit ratings are tested, namely the disciplining, rating conservatism and reputation hypotheses (see Section 2).We test three key indicators: rating levels, the number of false warnings and the informational content of rating signals.The precise testable predictions arising from each hypothesis are detailed in Section 4.
The disciplining hypothesis proposes that the regulation motivates CRAs to invest in improvements to their methodologies, due diligence and performance monitoring (Bae et al., 2015;Dimitrov et al., 2015).The regulation also promotes enhanced disclosure of conflicts of interest within the rating process, strengthening of CRAs' internal control structures and increased methodological transparency.These improvements in CRAs rating practices can enhance rating quality and accuracy (Hirth, 2014;Cornaggia et al., 2018).
Rating conservatism implies that CRAs will lower their ratings (under-rate) to avoid incurring fines, penalties and scrutiny introduced by the more stringent new regulations.A rating that is too generous is more likely to incur scrutiny and criticism than a rating that is too low, and thus CRAs may choose to err on the side of caution.Further, we argue that conservatism is more likely to be observed in FI ratings, since FIs have greater information opacity/asymmetry than firms in other industries (Flannery et al., 2013;Morgan, 2002).Bannier et al. (2010) find that the strength of the conservatism increases when the issuers' creditworthiness is more uncertain (i.e. more opaque).Atilgan et al. (2015) also show that information asymmetry is a key reason for increases in conservative rating bias.
The reputation hypothesis stems from the notion of "reputational capital" (Flynn and Ghent, 2017), whereby CRAs may enhance their reputation by rating accurately, so that they can benefit in the future from opportunities to inflate their ratings to increase their market share and hence their revenues.Reputational shocks deplete CRAs' reputational capital and trigger a subsequent period of reputation building which is characterised by conservative ratings with less informational impact in financial markets (Bedendo et al., 2018).Crucially, the effect of the reputation hypothesis is expected to be stronger in regions where CRAs are more concerned about preserving their reputational capital (Becker and Milbourn, 2011;Dimitrov et al., 2015).
The results reveal that EU regulatory actions have largely been successful in reducing rating inflation and have led to a significant decrease in rating levels, as predicted by the regulators surveyed in EC (2016).2However, the increased regulatory scrutiny has changed CRA behaviour whereby ratings are increasingly conservative (in line with the rating conservatism hypothesis). 3This leads to an increase in unwarranted downgrades or false warnings, which in turn contribute to an observed decrease in the market reactions to negative credit signals (less informative negative rating actions).There is some evidence that rating upgrades are more informative in the post-regulatory period, particularly those by S&P and Fitch.This is a consistent outcome of increased rating conservatism because CRAs expend greater effort to ensure that each upgrade is warranted.The findings also show that the EU regulatory update in May 2013 acted to strengthen the existing impact of the prior regulation.
Our results contrast with those reported by Dimitrov et al. (2015) for the US corporate rating market following the Dodd-Frank Act (DFA).They study the impact of the DFA on US corporate ratings (excluding FIs) and find no evidence of increased disciplining or rating conservatism, but that CRAs become more protective of their reputation (i.e.consistent with the reputation hypothesis).Our findings imply that there are unique effects in the EU context.
The EU and US CRA regulations have some similar objectives, but they differ in the details and the execution. 4ESMA has been more active in taking enforcement actions under its new regulatory regime than has the US Security and Exchange Commission (SEC) during the same period.ESMA has issued several fines to CRAs for breaches of the new regulation, while the SEC has appeared to be more reluctant. 5Our results are robust to consideration of the DFA timing, and there is a clear incremental effect of the additional EU regulation when CRA II and CRA III are implemented in July 2011 and May 2013 respectively.
The remainder of the paper is organised as follows.Section 2 reviews prior research on the impact of regulation on CRAs and discusses the development of hypotheses.Section 3 describes the data sample and Section 4 discusses the methodology and the testable predictions based on the hypotheses.Section 5 analyses the empirical results and Section 6 concludes the paper.

Literature review and development of hypotheses
The business model adopted by CRAs is predominantly the "issuer pays" approach, whereby the issuer is charged for receiving a rating on a debt issuance.Issuers can be assumed to prefer favourable over truthful ratings and, since it is the issuer who pays fees to the CRA, there exists an inherent conflict of interest.This could be even more problematic in a context of competition for rating business, as discussed later in this section.CRAs argue that the main incentive for them to provide honest and accurate ratings is their concern for their reputation (Bar-Isaac and Shapiro 2013).Some researchers propose that CRAs possess "reputational capital" (Flynn and Ghent, 2017), whereby CRAs may enhance their reputation by rating accurately, so that they can subsequently benefit from future opportunities to inflate ratings to increase revenues.Bedendo et al. (2018) argue that reputational shocks, such as the sub-prime crisis and the lawsuit against S&P,6 cause the depletion of CRAs' reputational capital and thus trigger a period of reputation building which is characterised by more conservative ratings with less informational impact in financial markets.Baghai and Becker (2020) also confirm that CRAs which suffer reputational damage can (re)gain market share by issuing optimistic ratings.
Therefore, the reputation hypothesis argues that CRAs lower their ratings to rebuild their depleted "reputational capital" following a reputational shock.Previous studies (e.g.Becker and Milbourn, 2011;Dimitrov et al., 2015) show that the effect of the reputation hypothesis is crucially stronger in markets where CRAs are more concerned with preserving their reputational capital, particularly in markets where CRAs are less concerned about competition.
Competition in the rating industry could potentially impact upon the quality of ratings issued.This proposition is tested by Becker and Milbourn (2011) who examine the entry of a third CRA (Fitch) into the US corporate bond rating market.They find that increased competition from Fitch coincides with lower quality ratings from incumbents (Moody's and S&P), which is attributed to inflated corporate rating levels.In addition, Dimitrov et al. (2015) empirically analyse the impact of the DFA on corporate bond ratings, using Fitch market share across industries as a proxy for reputational concerns (drawing from Becker and Milbourn (2011)).They find that CRAs issue lower, less accurate and less informative ratings following the DFA, especially in circumstances where their reputational costs are greater, which is consistent with the reputation hypothesis.
Similar findings are reported for structured finance ratings.Cohen and Manuszak (2013) investigate the competition effects on AAA-rated tranches of over 300 commercial mortgage-backed securities.With similar findings to Becker and Milbourn (2011), they provide evidence that competitive pressure from a third market entrant (Fitch) results in more lenient ratings assigned by the incumbents (Moody's and S&P).Such effects of competition were more pronounced when Fitch's market share was low, but disappeared after Fitch became more established.Flynn and Ghent (2017) analyse the entry of new CRAs into the structured finance rating market and find evidence to support Becker and Milbourn (2011).The evidence points to the fact that CRAs are more concerned about preserving their market share, by assigning more inflated ratings, than maintaining their reputational capital when competition is fierce.
Crucially, the strength of their desire to protect their 'reputational capital' will vary with their concern for their reputation, i.e. inversely proportional to the competition.
The disciplining hypothesis argues that the increased rating discipline promoted by the regulation leads to improved rating quality.The EU regulatory reforms contain many clauses to motivate CRAs to invest in improving their methodologies and having a strong framework for due diligence and performance monitoring.They also require CRAs to fully disclose any conflicts of interest, strengthen their internal control structures, and increase transparency of rating processes and performance.Hirth (2014) finds that the implementation of performance monitoring by a regulator rather than by investors can lead CRAs to become more honest.Cornaggia et al. (2018) argue that improved rating processes and increased rating transparency can enhance rating quality, leading CRAs to increasingly assign ratings free from inflation.
The rating conservatism hypothesis stems from Bannier et al. (2010) who show that CRAs are exposed to more severe scrutiny and penalties by over-rating (being less conservative), rather than by under-rating (being more conservative).The global financial crisis highlighted the detrimental role of rating inflation, which then became a focus of increased regulatory scrutiny (Baghai et al., 2014).Although the regulation discourages optimistic ratings bias, it does not equally punish pessimistic rating bias.As a result, increased regulatory stringency, fines and liability can cause a shift to more conservative rating behaviour.Opp et al. (2013) develop a theoretical framework which predicts that the DFA would result in a systematic downward shift in the distribution of ratings from CRAs, caused by lower regulatory advantages for higher ratings.In addition, Baghai et al. (2014) show that CRAs became more conservative from 1985 to 2009, with average rating levels dropping three notches over the period, which is at odds with declining default rates during their sample period.Their evidence suggests that capital markets do not perceive the corresponding increase in conservatism to be warranted.Atilgan et al. (2015) also highlight that CRAs are more likely to be conservative when the cost for over rating is high.Therefore, the rating conservatism hypothesis states that in an attempt to avoid incurring such fines, penalties and scrutiny, CRAs will lower their ratings, i.e.rate more conservatively. 7verall, our three hypotheses are summarised as follows:

Hypothesis Summary
Disciplining Improvements in rating process and methodology, stimulated by the regulation, lead to better quality ratings.

Rating conservatism
CRAs rate more conservatively to avoid incurring regulatory fines, scrutiny and penalties.

Reputation
Following a reputational shock, CRAs enter a period of reputation building where they rate more conservatively, and the effect is stronger in markets where CRAs care more about their reputation.
Each hypothesis makes distinct testable predictions on the way in which the regulation will impact three key areas: (i) rating levels, (ii) false warnings and (iii) the informational content of credit rating signals.These will be discussed in Section 4.

Data
The  2016) and others).
Table 1 presents the descriptions and summary statistics for the variables, which are selected following the literature on the determinants of FI ratings (e.g.Huang and Shen (2015)).10 The credit ratings are mapped to a 52-point comprehensive credit rating (CCR) scale: 11, 12 Then, for positive (negative) watch we add +2 (-2) and for positive (negative) outlook we add +1 (-1).There are 1108 negative rating, outlook and watch events and 430 positive events (Table 4).S&P issues more downgrades during the sample period (398), than Moody's (379) and Fitch (331).Moody's issues the most upgrades (191), compared to S&P (142) and Fitch (97).

S&P market share
To distinguish between markets with greater and lesser reputational concerns, it is necessary to utilize a proxy.A suitable proxy is derived from Becker and Milbourn (2011) and Dimitrov et al. (2015) but is adapted to the European FI context.Two CRAs with a dominant market share will consider a strengthening presence of a third CRA with a smaller pre-existing market share as a competitive threat.Consequently, they will behave increasingly competitively (caring less about their reputation and being more likely to inflate ratings) in seeking to stave off continued incursion into the market by the competitor.Becker and Milbourn (2011) and Dimitrov et al. (2015) chose Fitch market share as a proxy for reputational concerns in the US corporate rating market, because Fitch has a relatively weaker presence in that market.This study's sample consists of European FI ratings, where the three large CRAs have substantially varying market shares across countries.Fitch is a relatively stronger participant in Europe than in the US and stronger in the FI sector than in corporate bond ratings.The long-established strength of Fitch in the European FI rating sector is influenced by: (i) having their global headquarters in London during the relevant time period; (ii) historical acquisitions of IBCA Limited (thereby achieving a strong European presence) and Thomson Financial BankWatch (thereby strengthening their position in FI ratings).
Calculated at the issuer level, S&P has the lowest market share in the European FI rating market and thus its market share serves as a better proxy for reputational concerns.Further, S&P has the lowest rate of growth in market share in FI ratings during the sample period, while Fitch has the fastest rate (see Fig. 1). 13ae et al. (2015) argue that there are two problems with the measure used by Becker and Milbourn (2011) and Dimitrov et al. (2015).First, that the results are driven by an endogeneity problem caused by unobservable industry effects and second, that the positive relation between credit ratings and Fitch market share does not hold when only firms in nonregulated industries are included in the analysis.We address the first issue by limiting our sample to a single industry and calculating market share variation on the country level, while controlling for country level variation using country*year fixed effects as well as FI characteristics.The second issue is addressed by considering a single industry, whereby the regulation is therefore applied homogenously across the sample (as all countries are affected equally and simultaneously by the regulation).
S&P market share (S&PMS) is calculated by dividing the number of S&P issuer ratings (assigned to FIs) in country j in year t by the total number of FI issuer ratings assigned by the big three CRAs in country j in year t (the resulting market share is lagged by 1 year in estimated models).Fig. 2 shows that the average S&P market share varies substantially across all countries in the sample and across time.S&P market share in the sample ranges from an average of 21.4% in 2005 to 24.1% in 2016.S&P market share also varies across countries with Estonia having no S&P FI ratings and Luxembourg having an average S&P market share of 40.1%. 14good proxy for reputation as there is no positive relationship with rating levels and therefore competition.CRAs have increased reputational concerns in markets where there is less competition (Becker and Milbourn, 2011).As S&P has the smallest market share, its increased presence in the market triggers increased competitive behaviour from the other two incumbent CRAs.Because Fitch has a more established presence, an increment in its market share does not trigger more competitive behaviour from the other two CRAs.We further check, using Eq.(A1) (see Appendix A), that S&P market share has a positive relation to rating levels both before and after the regulation, in addition to the entire sample (those results are available upon request). 14In Appendix A, we illustrate how S&P market share (S&PMS) can be used as a proxy for reputational concerns.The inference is that Moody's and Fitch assign higher ratings in countries with higher S&P market share.

Rating levels
A key aim of the EU regulation is to reduce rating inflation.All three hypotheses (see Section 2) predict that rating levels decrease in the post regulatory period.The disciplining hypothesis argues that the improvements in CRA methodology and rating process lead to a reduction in rating inflation.Rating conservatism argues that CRAs tend to under-rate issuers to reduce their susceptibility to fines, scrutiny, and liability.The reputation hypothesis suggests that CRAs assign lower ratings to safeguard their reputation.To examine whether rating levels decreased in the post regulatory period, the following ordered logit15 model is estimated: CRi,j,k,t is the credit rating of a FI i in country j assigned by CRA k at time t based on a 52-point CCR scale (see Section 3).Post is a dummy variable that takes the value of one after the new regulation and zero before.Eq. ( 1) is first estimated using July 2011, when the regulation became more strongly enforced by the newly established ESMA, as the start of the postregulatory period. 16Eq.( 1) is then estimated using two separate post-regulatory dummies.
Post1 takes the value one during the period July 2011 to May 2013, and zero otherwise.Post2 takes the value of one after May 2013 and zero otherwise to capture the latter regulatory update that increased the stringency of the rules and introduced a new liability regime.BANK is a set of variables that control for FI-specific characteristics (see Table 1).Moody's and Fitch are dummy variables that distinguish between ratings assigned by Moody's, Fitch and S&P (both dummies are zero in the latter case).CF *YF is a full set of interacted country and year dummy variables.In line with Acharya et al. (2013) and Philippon and Reshef (2013), we use country and time fixed effects, along with a dummy variable for regulatory change.17 Crucially, the reputation hypothesis makes a different prediction to the other two hypotheses, namely that the effect should be stronger in countries where CRAs care more about their reputation.To detect the presence of reputational effects, the model is expanded to consider whether the FI is in a country with stronger or weaker reputational concerns.We use S&P market share as a proxy for reputational concerns (see Section 3.1).In countries with a greater presence of the third CRA, the other two CRAs care less about their reputation due to the stronger competition (Becker and Milbourn, 2011;Dimitrov et al. 2015) The sample is split into two sub-groups, the lower quartile of S&PMS and the upper three quartiles of S&PMS (similar to Dimitrov et al. (2015)).The variable S&PMSj,t is a dummy with a value of one if in the first group and zero if in the second. 18The addition of Post*S&PMS allows for the extraction of the effect due to variations in reputational concerns in the postregulatory period and thus Post represents the change arising solely from the regulation.

False warnings
This section addresses the question of whether lower credit ratings in the postregulatory period are warranted by changing FI creditworthiness.If any change in rating levels is fully justified, there will be no significant increase in false warnings.If the observed lower ratings are not fully justified, an increase in false warnings would be identified (i.e.unjustified downgrades).The following logit model of false warnings is estimated: FWi,j,k,t is a dummy variable taking the value of one for a FI i rated BB+/Ba1 or lower in country j by CRA k at time t that does not face financial distress within one year, and zero otherwise (see Dimitrov et al. (2015)). 19FI failures are rare in Europe and therefore defining when a FI faces distress can be challenging.Betz et al.'s (2014) method is adopted here, whereby FIs are examined for potential distress events, including: (i) default/liquidation, (ii) government intervention/support and (iii) forced merger.The incidence of false warnings in our sample is shown in Fig. 3, and there is a clear increase in false warnings from 2010 to 2014.
The three hypotheses make different predictions with regards to false warnings.The disciplining hypothesis predicts no increase in the number of false warnings, because the regulation has acted to improve rating methodology and reduce rating inflation.Rating conservatism predicts an increase in the number of false warnings, as greater risk of regulatory intervention causes CRAs to under-rate, thereby inducing an increased incidence of unwarranted downgrades.The reputation hypothesis predicts that any increase in false warnings is more apparent in countries with stronger reputational concerns in the postregulation period.The following model is estimated: A positive and significant coefficient on Post would indicate an increase in false warnings and unwarranted downgrades in the post-regulatory era.Post*S&PMS captures the difference in impact between countries with stronger and weaker reputational concerns.

Informational content of ratings
Jackowicz et al. ( 2020) argue that credit rating downgrades represent one of the most prolific types of economic shocks influencing both issuers and investors, given that credit ratings are inherent in regulatory requirements and internal investment policies.ESMA seeks to reduce financial markets' mechanistic reliance on credit ratings and hence to reduce market overreactions to downgrades, which should consequently reduce the market reaction to negative rating signals.Improving rating quality would increase the informational content of (hence greater market reaction to) positive rating news.We use two methods, which are commonly applied in previous ratings-related literature, to examine the information content of CRAs' bank credit signals during periods prior and subsequent to the regulatory reforms of the rating industry: (i) Event study methodology, and (ii) Fixed effects model.First, in the event study, the market reaction to a credit signal on day t is measured by the abnormal stock return, calculated using a technique widely adopted in the literature (e.g.Correa et al., 2014;Jackowicz et al., 2020): The FI stock return is calculated over a 2-day period (t-1, t+1).α and β are the intercept and slope coefficients, respectively, of an OLS regression of FI i's stock returns on the market return.This is estimated using daily data from an event window of 230 days prior to 30 days prior [-230, -30] each rating announcement and a constant. 20  Second, a fixed effects model of rating announcements is constructed (positive and negative credit rating events are considered separately) as follows: Rating Eventit is a dummy variable equal to 1 on a credit signal date t for FI i and zero otherwise).AR is the abnormal stock return and is calculated as in Eq. ( 5).
The disciplining hypothesis predicts that negative and positive credit signals will become more informative, because improved methodologies, reduced rating inflation and greater diligence by CRAs will result in improved rating quality.Rating conservatism predicts that negative credit signals will become less informative, because CRAs tend to deflate their ratings to protect themselves against increased regulatory intervention.In addition, the EU regulation aims to mitigate the mechanistic market reaction to rating downgrades, which may potentially reduce the stock price reactions to negative signals.Conversely, positive credit signals may become more informative, as over-rating exposes CRAs to greater potential penalties and liability.This incentivises CRAs to expend greater effort to ensure that each positive signal is warranted.The reputation hypothesis stipulates that negative credit signals 20 Stock market data for 107 listed FIs and their respective country indices is collected from DataStream.
may become less informative, and positive credit signals may become more informative because CRAs wish to avoid the perception of biased ratings and therefore expend greater effort when issuing rating upgrades.Any effect due to the reputation hypothesis would differ between countries with greater and lesser reputational concerns.

Testable predictions
The testable predictions of our three hypotheses (see Section 2) on rating levels, false warnings and the informational content of rating upgrades and downgrades are summarized below:

Rating levels
In this sub-section, we analyse whether rating levels have changed following the introduction of the EU regulation of CRAs.To preview the findings, we show that: (i) rating levels are lower following the regulation, (ii) the effect does not differ with reputational concerns, and (iii) the May 2013 regulation update strengthens the regulatory/conservatism effect.
Eq. ( 1) is estimated twice using different dates for the start of the post-regulatory period, with the results reported in Table 2. Credit ratings are lower following the regulatory change.
First, Eq. ( 1) is estimated using July 2011 (when ESMA was established) as the start of the post-regulatory period.The coefficient of the regulatory change Post is -0.304, and thus the odds that a FI is rated as non-investment grade are 1.36 (1/ −0.304 ) times greater following the regulation. 21The results are consistent with the disciplining hypothesis, whereby rating quality improves and there is a reduction in inflated ratings, and with rating conservatism, whereby CRAs are induced by greater regulatory scrutiny to issue more conservatively biased ratings.
The results are also in line with the reputation hypothesis, whereby CRAs issue lower ratings following a reputational shock in order to protect their reputation.
Eq. ( 1) is then estimated using two separate post-regulatory dummies.Post1 takes the value one during the period July 2011 to May 2013, and zero otherwise, to capture any effects caused by the enforcement of the initial regulation by ESMA.Post2 takes the value of one after May 2013 and zero otherwise to capture the latter regulatory update.Eq. ( 1) produces the same inferences as reported above for the July 2011 handover of responsibilities to ESMA.The regulatory update in May 2013 then acts to strengthen this effect with a further decrease (Post2 coefficient is -0.413 and the odds of being rated non-investment grade are 1.51 times greater).
Consistent with the rating conservatism hypothesis, this additional decrease could arise from the increased stringency of the rules introduced by the 2013 regulatory update.This primarily introduced a new liability regime (Article 35a), giving investors and issuers the right to sue for damages, and strengthening existing disclosure and transparency requirements.
To investigate the difference further, Eq. ( 2) is estimated to take account of differences between countries with different reputational concerns, with the results reported in Table 2. that there is no difference in the impact of the regulation between countries where CRAs have stronger or weaker reputational concerns.The implication is that there are no reputational effects present and only the disciplining effect of the regulation remains.This acts through either the discipline channel or by stimulating increased rating conservatism, thus supporting the regulators' views expressed in EC ( 2016).This finding contrasts strongly with US evidence that reputational effects are strongly connected to the reductions in corporate ratings levels.22

False warnings
This sub-section aims to determine whether rating conservatism is driving the decrease in rating levels.To preview the findings, we show: (i) an increase in false warnings in the post regulatory period, (ii) the increase does not differ with reputational concerns, and (iii) the May 2013 regulation update strengthens the effect.
The results from Eq. ( 3) are reported in Table 3.After July 2011, there is a significant increase in false warnings (Post coefficient is 0.383).This implies that the odds that a CRA would issue a false warning after July 2011 are 1.47 ( 0.383 ) times greater than before.This increase in false warnings implies that not all rating downgrades are warranted.There are two potential reasons for this.First, increased rating conservatism caused by CRAs' concerns about potentially greater regulatory intervention in cases of over-rating.Second, CRAs issue more downgrades to protect their reputation and build reputational capital.Eq. ( 3) is then estimated using two separate post-regulatory dummies.The results show a strengthening of the result from Post1 to Post2 (the coefficient is 0.694, which doubles the odds of a false warning).This increase in unwarranted downgrades following the strengthening of the regulation in May 2013 and the introduction of the civil liability regime is highly suggestive of an increase in rating conservatism by CRAs as they respond to the increased potential cost for over-rating.
To differentiate between the two possibilities, Eq. ( 4) is estimated (see Table 3).
Following July 2011, there is an increase in the incidence of false warnings (Post coefficient is 0.464).The coefficient on Post*S&PMS is negative and is not significant, implying that countries in the bottom quartile of S&P market share do not show different outcomes from those in the top three quartiles (i.e.countries with lesser reputational concerns, greater competition).This evidence supports the notion that increased rating conservatism induced by regulation is driving the increased incidence of false warnings, rather than CRAs protecting their reputation.In other words, CRAs are downgrading FI ratings to avoid potentially exposing themselves to increased regulatory interventions.This is not dependent on reputational concerns because regulatory penalties would be applied to CRAs irrespective of their reputation.This result again contrasts with evidence from US corporate ratings, whereby the DFA's impact on false warnings is significantly stronger for industries where CRAs had stronger reputation concerns.
On estimating Eq. ( 4) with Post1, Post2 and S&PMS, the coefficients of the interaction terms are both insignificant, i.e. there are no different effects for countries where CRAs have weaker or stronger reputational concerns.This reinforces the hypothesis that rating conservatism drives the rating changes rather than CRAs protecting their reputation.The May 2013 regulatory update exacerbates the effect, as we see an increase in the number of unwarranted downgrades (i.e.false warnings) and no difference between countries with different reputational concerns.

Informational content of ratings
This sub-section compares stock market reactions to rating announcements before and after the establishment of ESMA in July 2011.To preview the findings, we reveal a decrease in the informational content of rating downgrades and an increase in informational content for rating upgrades, which are both consistent with increased rating conservatism.
The event study results, reported in  One of the intended aims of the regulation is to reduce the mechanistic market reaction to negative credit signals and it could therefore be argued that this has been successful.
However, this change may be also due in part to an increase in rating conservatism induced by the new regulation's discouragement of over-optimistic ratings.Following the regulatory reforms, there is an increase in unwarranted negative signals (false warnings, see Section 5.2).
It follows logically that unwarranted negative signals hold less information for the market.The results of the fixed effects model (Eq.( 6)) for upgrades demonstrate that, prior to the 2011 regulatory change, no significant reaction to rating upgrades is observed.Following the establishment of ESMA, a 0.445% reaction in stock prices is observed in response to rating upgrades (see Table 5).There is therefore some evidence for a limited increase in the informational content of upgrades.This is consistent with increased rating conservatism, in the sense that CRAs will expend more effort to ensure that rating upgrades are justified and those rating actions will thereby typically become more informative.
Lastly, the impact of reputational concerns is also considered.The results (available upon request) of both the event study and Eq. ( 7) show no significant stock market reaction to FI rating downgrades in groups of countries with greater and lesser reputational concerns following the regulatory change of July 2011.This indicates that reputational effects are not driving the decrease in the informational content of rating downgrades.These results support the overall findings of the negligible relevance of the reputation hypothesis in the European FI rating context.In contrast, the US corporate rating market demonstrates strong evidence of reputational effects, with downgrades in industries with stronger reputational concerns exhibiting a stronger stock market reaction (Dimitrov et al. (2015)).
The impact of the May 2013 regulatory update upon the stock market reaction to negative rating signals is also examined (see Panel B of Table 4).The event study results show a clear reduction in the informational content of the negative credit signals following the regulatory update (1.146% decrease in the market reaction, see Panel B of Table 4 -All signals sample).
We also find that, following the regulatory change, all types of negative credit signals did not induce negative and significant abnormal return.The results from the fixed effects model (Eq.( 6)) corroborate those of the event study because once again a significant negative reaction to rating downgrades is observed (-0.483%, see Table 5) prior to July 2011.This then disappears and a positive reaction (which indicates a lack of information) is observed following the May 2013 update (Table 5).For rating upgrades, the fixed effects model shows no significant reaction to rating upgrades prior to July 2011, a significantly stronger market reaction after July 2011 and then an insignificant reaction following the May 2013 update.These results are consistent with the rating conservatism hypothesis.
crisis (the collapse of Lehman Brothers in September 2008), (ii) the EU sovereign debt crisis (April 2010, the date that S&P downgraded Greece to junk status) and (iii) the S&P litigation case (February 2013).To control for the impact of reputational shocks during the sample period, Eq. ( 1) and ( 4) are estimated with an additional dummy RepShocki,j,t.that captures periods of reputational shock for CRAs and takes the value of one for a period of one year after the reputational shock and zero otherwise.
The results of Eq. ( 1) in Table 7 show a significant reduction in rating levels in the year following a shock and there also remains a significant impact from the regulation (Post coefficient is -0.303, therefore the magnitude of the rating reduction due to Post has barely decreased at all compared to the previous estimation).Thus, while reputational shocks may contribute to decreased rating levels, they are not solely responsible.The results of Eq. ( 3) show a significant increase in false warnings following both the reputational shock and the regulation.This is attributable both to CRAs seeking to protect their reputation after any shock and to the role of regulation.
The European sovereign debt crisis was characterised by a particular concentration of rating downgrades in peripheral Euro-zone countries, namely Greece, Ireland, Italy, Portugal and Spain (GIIPS).Our sample is dominated by FIs in other (core) countries.Yet, as a robustness test, Eq. (1) to Eq. (4), Eq. ( 6) and Eq. ( 7) are estimated with a sub-sample excluding the GIIPS countries.The inferences (results available upon request) are similar to those reported earlier in the paper.This indicates that our findings are not driven by the EU sovereign debt crisis.Dilly and Mählmann (2016) show that rating quality is counter cyclical and ratings quality should be higher in an economic downturn.We would then expect that during our sample period (economic downturn) that ratings quality should increase.This would then predict a reduction in false warnings and an increase in the informational content of ratings announcements.We find, however, that there is an increase in false warnings and a reduction in the informational content of rating downgrades.We can conclude that our results cannot be driven by cyclical effects.
Finally, the recent EU bank bail-in regulations (starting from January 2016 but variable timing across countries) are an additional factor to consider.Because these laws shift some of the responsibility for bank resolution from the government to shareholders and creditors, they could potentially impact FI rating levels.A dummy variable is included on a country-bycountry basis to take account of the period when the law is introduced in that country (based on ISDA ( 2016)).The results (available on request) of Eq. (1) to Eq. ( 4) are consistent and robust to the inclusion of this bail-in dummy.The bail-in variable is not significant in any estimated model.

Conclusions
This unique study investigates whether the EU regulatory reforms of the rating industry in response to the global financial crisis have been successful.Our paper is also unique in its focus on the quality of FIs' ratings following the regulatory reform.A sample of 758 financial institutions across 27 European countries rated by S&P, Moody's and Fitch during January 2006 to June 2016 is used.We examine the impact of EU regulation on rating levels, the incidence of false warnings and the responsiveness of stock markets to credit rating signals (rating informativeness).
We contribute to the literature by demonstrating that the EU regulatory reforms act to promote more conservative rating behaviour, leading to a reduction in the levels of European FI ratings.Overly generous ratings are much more likely to incur scrutiny and criticism, thus CRAs err on the side of caution.This has led to an increased incidence of unjustified downgrades (false rating warnings) and with it a corresponding decrease in the informational content of (and stock price reactions to) rating downgrades.The latter decrease in informational content may also be driven in part by a declining reliance on CRAs by market participants, which reduces the mechanistic reactions to rating signals in financial markets (a key aim of ESMA).There is evidence of increased stock price sensitivity to rating upgrades (mainly those by S&P and Fitch) following July 2011.This is consistent with the increased presence of rating conservatism, i.e. within an environment of increased regulatory scrutiny and potential legal repercussions, CRAs spend more effort and resources to ensure that upgrades are justified.
These results are robust to the inclusion of reputational shocks, the more recent EU bail-in laws and to alternative definitions of false warnings and of the rating scale.
Our results contrast with evidence from US corporate bond ratings where it appears that reputational effects have driven changes in CRA behaviour subsequent to the DFA.Becker and Milbourn (2011) and Dimitrov et al. (2015) propose that incumbent CRAs have greater reputational concerns in markets with the presence of a third CRA with a smaller market share (markets with less competition).In contrast to the US, we find no evidence of variation in effects for EU FI ratings across countries with differing reputational concerns.The EU regulatory update of May 2013 strengthens the existing impact of the regulation on rating conservatism by further reducing rating levels and increasing unwarranted downgrades.
Although the EU and US CRA regulatory reforms have some similarities, there are substantial differences in the details and execution.ESMA has been more active in enforcing the regulatory amendments than the US SEC.We consider the incremental effect of the EU regulation, alongside the earlier introduction of DFA to regulate CRAs in the US.The results are robust to the consideration of DFA and we find that the EU regulation has a far more significant impact, as would be anticipated.
This paper furthers the discussion on suitable mechanisms for regulating CRAs in the future.While the regulation has been successful in reducing rating inflation, the evidence indicates that this is a by-product of a behavioural shift towards increased rating conservatism, in line with Baghai et al. (2014), rather than a direct result of increased rating quality.This has come at the cost of an increased incidence of false warnings and reduced rating downgrade informativeness, but there is evidence of reduced mechanistic market reactions to rating downgrades.This is not the first illustration of CRA regulation producing some unintended consequences (Behr et al., 2018)).
Several other policy recommendations arise.Credit ratings are an important source of information for market participants and therefore regulators should reflect on the need to alleviate both overly optimistic and conservative biases.Promoting improvements within the rating process should continue as a central tenet of the regulation in order to mitigate the conservative rating bias.Regulators should also consider the potential costs to market functioning and informational efficiency which arise from a reduced informativeness of rating downgrades.Further, regulators should more explicitly consider the structured debt-rating sector separately from the FI rating segment, given that we find evidence that increased competition among CRAs leads to more inflated FI ratings.
Rating signals are restricted to those of Moody's and Fitch and the estimated model includes the S&P market share variable.A strengthening of the impact of the regulation is observed in all countries following both July 2011 and May 2013, implying the strong presence of either disciplining effects or increased rating conservatism.Using the July 2011 regulatory start date, there is no variation in effect between countries with greater or lesser reputational concerns (insignificant Post*S&PMS) and countries in the bottom quartile of S&P market share reveal no differences compared with countries in the top three quartiles.Second, the significant coefficients on both Post1 (-0.345) and Post2 (-0.427) imply lower ratings following the regulation.Post1 * S&PMS and Post2 * S&PMS coefficients are not significant, indicating The impact of the regulatory change in July 2011 on stock market reactions to positive signals is also examined.Panel A of Table4shows that abnormal stock returns for positive credit news are statistically insignificant before the regulatory change and remain insignificant after the regulation (see All signals sample).This is consistent with the findings of prior literature (e.g.Correa et al., 2014) that the responses to CRAs' positive credit signals are muted given that positive credit signals are usually anticipated by market participants.Prior to the regulatory change, all types of positive credit signals did not induce a significant increase in stock prices.However, following the regulation, rating upgrades which were not preceded by watch/outlook signals, and therefore less anticipated by the market participants, elicit positive and significant abnormal returns (1.650%).Examining signals by Moody's and Fitch only, unreported results (available on request) reveal that, following the regulation, rating upgrades by both Moody's and Fitch elicit positive and significant abnormal returns (0.734%).

Fig. 1 .
Fig. 1.S&P and Fitch market share over time.The Figure displays the variation of average S&P and Fitch market share over time in the sample of 758 rated European FIs during the period from January 2006 to June 2016 in the 27 EU countries.

Fig. 2 .
Fig. 2. S&P market share distribution.Variation of S&P market share over country and year in the sample of 758 rated European FIs during the period from January 2006 to June 2016 in the 27 EU countries.

Fig. 3 .
Fig. 3. Incidence of false warning.The Figure displays the count of periods in which a CRA had issued a false warning to a FI from the sample of 758 rated European FIs during the period from January 2006 to June 2016 in the 27 EU countries.
sample consists of 758 rated FIs in 27 EU countries, 8 of which 378 are rated by S&P, 468 by Moody's and 494 by Fitch, during the period from 1 st January 2006 to 1 st June 2016.FI ratings and accounting variables are obtained from BankScope.

Table 4
watch signals have a stronger impact on financial markets because they are less anticipated by market participants, and CRAs disclose more private information to the markets via the

Table 5 )
. However, after the regulatory change, rating downgrades no longer do so (insignificant Post * Rating downgrade).

Table 1 . Distribution and summary statistics for the control variables
The Table reports the variables used in the regression models.The sample consists of 758 rated European FIs during the period January 2006 to June 2016 in the 27 EU countries.The data of these financial variables is trimmed at the top and bottom 1% to remove outliers.

Table 2 . Rating level
The Table presents the results of the ordered logit regressions for the sample of European FIs during the period January 2006 to June 2016 rated by S&P, Moody's and Fitch in Eq. (1), and by Moody's and Fitch in Eq. (2).Two different regulatory start dates are included.First, July 2011 when ESMA was established and second, May 2013 when the regulatory update was released.The dependent variable is  ,,, : the credit rating level of FI i in country j by CRA k at time t based on a 52-point CCR rating scale.Post is a dummy variable that takes the value of 1 after July 2011 (establishment of ESMA) and zero otherwise.When both regulatory changes are considered, Post1 takes the value of one between July 2011 and May 2013, zero otherwise.Post2 takes the value of one after May 2013 and zero otherwise.S&PMS is a dummy variable that takes the value of 1 in countries in the bottom quartile of S&P market share and zero in the top three quartiles.Moody's and Fitch are dummy variables that take the value of 1 if the rating is issued by them and zero otherwise (if both are zero this indicates a rating by S&P).For control variables' definitions, see Table1.Standard errors are clustered by FI and a full set of year*country dummies are included.

Table 3 . False warnings
The Table presents the results of logit regressions for the sample of rated European FIs during the period January 2006 to June 2016 rated by S&P, Moody's and Fitch in Eq. (3), and by Moody's and Fitch in Eq. (4).Two different regulatory start dates are included.First, July 2011 when ESMA was established and second, May 2013 when the regulatory update was released.The dependent variable  ,,, , a dummy representing false warnings, takes the value of 1 if an FI with a rating of BB+ or below does not default after one year and zero otherwise.Post is a dummy variable that takes the value of 1 after July 2011 (establishment of ESMA) and zero otherwise.When both regulatory changes are considered, Post1 takes the value of one between July 2011 and May 2013, zero otherwise.Post2 takes the value of one after May 2013 and zero otherwise.S&PMS is a dummy variable that takes the value of 1 in countries in the bottom quartile of S&P market share and zero in the top three quartiles.Moody's and Fitch are dummy variables that take the value of 1 if the rating is issued by them and zero otherwise (if both are zero this indicates a rating by S&P).For control variables' definitions see Table1.Standard errors are clustered by FI and a full set of year*country dummies are included.***, **, * represent significance at 1%, 5% and 10% levels respectively.

Table 6 . Incremental effect of the regulation
The Table presents the results of the ordered logit regressions for the sample of European FIs during the period January 2006 to June 2016 rated by S&P, Moody's and Fitch in Eq. (1) and (3), and by Moody's and Fitch in Eq. (2) and (4).Three different regulatory start dates are included.First, July 2011 when ESMA was established, second May 2013 when the regulatory update was released and third, July 2010 when Dodd-Frank Act was implemented in the US.The dependent variable in Panel A is  ,,, : the credit rating level of FI i in country j by CRA k at time t based on a 52-point CCR rating scale, and in Panel B is  ,,, , a dummy representing false warnings, takes the value of 1 if an FI with a rating of BB+ or below does not default after one year and zero otherwise.Post is a dummy variable that takes the value of 1 after July 2011 (establishment of ESMA) and zero otherwise.When both regulatory changes are considered, Post1 takes the value of one between July 2011 and May 2013, zero otherwise.Post2 takes the value of one after May 2013 and zero otherwise.Post Dodd-Frank takes the value of 1 after July 2010 and zero otherwise.S&PMS is a dummy variable that takes the value of 1 in countries in the bottom quartile of S&P market share and zero in the top three quartiles.Moody's and Fitch are dummy variables that take the value of 1 if the rating is issued by them and zero otherwise (if both are zero this indicates a rating by S&P).For control variables' definitions, see Table1.Standard errors are clustered by FI and a full set of year*country dummies are included.***, **, * represent significance at the 1%, 5% and 10% levels respectively.

Table 7 . Reputational shocks
This Table shows the results of ordered logit regressions for Eq.(1) (rating levels) and Eq.(3) (false warnings) using a sample of 758 rated European FIs during the period January 2006 to June 2016 in the 27 EU countries.Post takes the value of one after 1 st July 2011 and zero otherwise.Reputational shock is a dummy that takes the value of one in the following a reputational shock and zero otherwise.Reputational shocks take place in September 2008, April 2010 and the February 2013.Moody's and Fitch are dummy variables that take the value of 1 if the rating is issued by them and zero otherwise (if both are zero this indicates a rating by S&P).For control variables' definitions, see Table1.The standard errors are clustered by FI. ***, **, * represent significance at 1%, 5% and 10% levels respectively.

Table A1 . Impact of S&P market share
The Table reports the results of the ordered logit model -Eq.(A1).The dependent variable is the FI credit rating (based on the 52 point CCR scale).The key independent variable is S&PMSt-1, S&P market share (lagged by 1 year), defined as a dummy variable with a value 1 for FIs in countries within the lower quartile of S&P market share and zero within the upper three quartiles of S&P market share.The sample includes 758 rated European FIs during the period January 2006 to June 2016 in the 27 EU.See Table1for the definitions of control variables.Standard errors are clustered by FI and a full set of country*year dummies are included.***, **, * represent significance at the 1%, 5% and 10% levels respectively.