Credit rating agencies (hereafter CRAs) are companies that issue ‘qualified opinions’ on the creditworthiness of a variety of financial products. This evaluation aims to reduce information asymmetry on markets and to give an objective view on how risky is the rated product, leading to higher confidence from market players. The fact that ratings reduce uncertainty and have an informative function gave them an important role in regulatory frameworks in both the EU and the USA. A good example of this reliance on ratings from regulators can be found in the ongoing asset purchase programmes of the European Central Bank, in which qualified bonds are determined by their ratings.

The credit rating industry can be viewed as an oligopoly market, with three main agencies (the so-called Big Three) accounting together for 95% of total market shares [1]. These Big Three are Standard & Poor’s (S&P), Moody’s (each of them accounting for 40% of market share) and Fitch ratings (15% of market share).

During the last decades, both investors and regulators tended to have an excessive reliance on ratings with noxious consequences in some cases. A representative case is the Enron one (2001): 4 days before its bankruptcy the company still had an investment-grade status from S&P and Moody’s [2]. This case and the more recent debacle of the subprime financial products shed light on the risk implied by an over-reliance on ratings, and called for a new and better-designed regulation of CRAs activities. The next section will present a historical overview of regulatory regimes in the USA and the EU, followed by a comparison between these regimes. It will shed light on structural differences across regions prior to the 2007 crisis, explaining why a massive regulatory change had to be made in the EU, while it was not the case in the USA. This historical review will also point out how CRAs and ratings became a cornerstone of financial regulations, making changes quite hard to implement.

In order to allow an easy and unequivocal comparison of regulatory frameworks, we decided to adopt a chronological view. US and EU regulations will be presented together and finally compared at the end of the section. A timeline of the most important reforms is presented in “Appendix 1”.

Credit rating agency reforms before the subprime crisis (1934–2007)

Ratings were already informally used by the Federal Reserve since 1930 in order to evaluate the quality of banks’ assets [3]. The very first reference to CRA in a legally binding text can be found in the Securities Exchange Act [4] (hereinafter SEA) (1934), aiming to set a governance framework for markets and to protect the investors. The original document was amended in 1936, prohibiting any investment from banks in bonds rated below investment grade. Although the original aim of this regulation (i.e. improving the global quality of banks’ balance sheets) is clearly positive and beneficial, some imperfections or perverse effects can also be pointed out. First of all, including ratings and CRA’s activities in law allowed them to acquire a crucial role in both market behaviour and future regulatory frameworks. This point is clearly underlined by Partnoy [5] with the so-called regulatory license phenomenon. From this time, CRAs have the power to unlock access to the market, leading even small issuers to have an incentive to have their debt rated by the rating agencies. Second, even if the SEA clearly mentions ratings, it does not give any precise reference on who can issue ratings. A simple unclear reference to ‘recognized rating manuals’ is made, leading White [3] to conclude that probably only the Big Three (S&P, Moody’s and Fitch) were allowed to issue ratings at that time.

This last imperfection was partially curbed in 1975 by a Securities and Exchange Commission (hereafter SEC) rule targeting capital requirements for broker-dealers (Rule 15c3-1 [6]). This rule linked capital requirements to ratings for each product detained by market players and, more remarkably, created the concept of ‘Nationally Recognized Statistical Rating Organizations’ (NRSRO) whose ratings were the only ones that could be used in order to determine capital requirements. Despite this step, the Rule did not totally address concerns about CRAs, as it led de facto to a greater reliance on ratings from market players [5]. Moreover, as it was pointed out by Conte and Parmeggiani [7], the selection procedure of NRSROs was very opaque and not formally designed. The SEC only sent ‘no-action letters’ to the recognised CRA, without giving any further detail on why they were admitted in the NRSRO category. In 1997 the SEC proposed a clear framework on how to designate NRSROs, but this framework was ‘mercifully’ [3] not implemented due to inherent contradictions in the proposal (to be nationally recognised was a prerequisite for the candidates who wanted to obtain the NRSRO status and to be nationally recognised).

Bond raters came back at the centre of attention in the early 2000s after the debacle of Enron, this company being granted an investment-category rating by two of the Big Three until a few days before its bankruptcy on 2 December 2001. This event led the US Congress to include a special section in the Sarbanes–Oxley Act [8] constraining the SEC to shortly issue a report analysing the role and function of credit rating agencies in the operation of the securities market (section 702). Such a report was released in 2003, but it failed to address concerns raised by the Sarbanes–Oxley Act and even shed light on new issues about NRSROs’ actual and future regulation.

In 2006 the USA adopted the Credit Rating Agency Reform Act of 2006 [9], containing new regulatory regimes for CRAs. The main targets of this Act were to increase the quality of ratings, to formalise the NRSRO designation procedure and to give to the SEC the status of sole enforcement authority over all NRSROs. This regulation gave clear steps for the designation of NRSROs, but also put a great emphasis on the possible abuses made by these NRSROs.

Before the presentation of binding rules that were adopted in the EU in 2009, we now turn to the regulatory framework which was previously built. Before a specific regulatory framework was set out at the European level in 2009, CRAs were nationally regulated and had to do self-regulation following the IOSCO (International Organization of Securities Commissions) Code Of Conduct Fundamentals For Credit Rating Agencies [10] under the ‘comply-or-explain’ principle: if a CRA decided not to follow a specific point of the Code or adopted a different approach, it had to publicly explain why it did so. In addition to that, CRAs had to submit an annual report to the Committee of European Securities Regulators (CESR), and this procedure was considered as being sufficient to ensure a good conduct of business from CRAs.

The first (indirect) reference to CRAs in a European legislation text can be found in the ‘Market Abuse Directive’ of 2003 [11], which particularly targeted insider trading and market manipulation by forcing companies to disclose the name of their staff members having access to privileged information. CRAs were also cited the same year in the Directive 2003/125/EC [12] underlying the fact that ratings are not recommendations to invest but only ‘opinions’, and separating CRAs from institutions providing investment advisory. CRAs were, moreover, supposed to underline this point and had to disclose any potential conflict of interest that could have occurred during the rating process. The MiFID Directive [13] indirectly impacts CRAs consultancy activities, but does not say anything about ratings issuance, which is their main activity. A 2005 report [14] from CESR adopts a ‘wait-and-see’ position, arguing that no specific regulation is needed and that the effects of the application of the IOSCO code seem to be sufficient. This report also underlines the fact that the issues raised by CRAs have to be regulated within a global framework.

Ratings gained importance in the EU legislation in March 2006 through the ‘Capital Requirements Directives’ (2006/48/EC & 2006/49/EC, hereafter CRD) [15, 16], translating the Basel II framework (2004) into legally binding texts. The CRD directive gives a fundamental role to ratings in the computation of capital requirements imposed to banks, the amount depending on how each component of the balance sheet is rated. In order to ensure the quality of ratings a list of recognised CRAs is created (as it was previously done in the USA), reinforcing de facto the oligopolistic character of CRAs industry by making market entry more difficult for potential new players. No further regulation on CRAs was planned at that time, except if it was strictly necessary for the achievement of public policy objectives [17]. In December of the same year, a report from the CESR to the European Commission [18] points out that punctual deviations from the IOSCO code of conduct are observed in each of the major CRAs. A particular emphasis is made on ancillary services proposed by CRAs, services that are supposed to be fully separated from rating activities. Nevertheless, the report underlines that all major CRAs consider the ‘rating assessment service’ (study of the potential impact on a rating of an event, such as a merger) as being an ancillary service, while the IOSCO code does not do so. The report shows that this specific point is particularly important for the structured finance products leading to potential risks, and calls for an improvement of the definition of ancillary services in the IOSCO code.

It turned out during the so-called subprime crisis in 2007–2008 that the CESR warnings were justified and consistent. These events allowed regulators to become aware that regulations were not sufficient nor well designed, and that a change had to occur.

Credit rating agency reforms after the subprime crisis (2007–present)

In May 2008, a CESR report [19] points out that a stricter regulation of CRAs is needed at the European level and that the IOSCO code has to be modified, and calls for the creation of a European monitoring body. Nevertheless, this report does not consider self-regulation as being harmful and does not plan specific measures against it.

Following a 2008 proposal from the European Commission [20], the very first real and powerful common regulatory framework at the EU level was set up in 2009 with Regulation (EC) No 1060/2009 on Credit Rating Agencies [21], also known as ‘CRA I’. This text has the double objective to restore confidence and ensure the stability of markets, by targeting four main areas:

  1. 1.

    Avoidance of conflicts of interest, notably by ensuring the independence of administrative/supervisory boards and by preventing CRAs to give consultancy services to rated entities or to related third parties (article 6; Annex I-A and Annex I-B).

  2. 2.

    Improvement of the quality of ratings, notably by ensuring the quality of methodologies, models and key assumptions that are used (article 8).

  3. 3.

    Increased transparency, by setting disclosure requirements with the publication of a yearly transparency report showing detailed information, notably on the legal structure of the CRA, on internal control mechanisms and on the sources of revenues of the CRA (articles 11 to 13 and Annex II). As underlined by De Haan et al. [22], this regulation is also the first one to make mandatory the disclosure of methodologies and quantitative assumptions used by CRAs.

  4. 4.

    Increased efficiency of registration and surveillance frameworks: the enforcement of the rules is supervised by CESR and national authorities (Title III).

Despite these improvements, it is important to notice that some aspects of the regulation are present in the preamble but not in the text of the regulation, and then are not enforced. For example, a call for more competition is made in point 55 of the preamble, but is not present in the legally binding text.

Covering numerous aspects of financial markets, the US Dodd–Frank Wall Street Reform and Consumer Protection Act [23] (2010, hereinafter DFA) also tightens CRAs regulation in several ways. An increase in the SEC powers, linked with the creation of the Office of Credit Rating (OCR) which monitors and examines registered NRSROs and verifies their compliance with statutory and SEC requirements, is one of the improvements allowed by the DFA. Nevertheless, this text also showed the great power that CRAs had at this time on financial markets. In its section 939G, the DFA intended to remove rule 436(g) of the Securities Act (1933) [24] and then to make possible lawsuits against CRAs in case of adverse practices. The CRAs obviously did not appreciate such a possibility and retaliated by deciding to refuse the use of their ratings in the SEC fillings for asset-backed securities, leading to a freeze of this market. The removal of rule 436(g) was promptly suspended by the SEC, but this episode allowed CRAs to demonstrate their power and to continue to push against the DFA [25]. The DFA finally intended to remove reliance on NRSROs in new regulations (section 939). The idea was to create independent creditworthiness standards and to reduce the over-reliance on CRAs that was observed in the previous regulatory frameworks. It nevertheless turned out that creating new models was a difficult task, and the fact that many financial activities (such as collateral agreements) still rely on ratings from CRAs gave them a crucial role and made a complete removal in future legislations almost impossible. Then only section 939A is applied, removing all reference to NRSROs ratings in the future legislation.

As demonstrated above, the enforcement of the DFA ‘CRA aspects’ was not an easy task and was not completed. Indeed, the DFA allowed regulators to become aware of the fact that CRAs have a great place in the financial system but also in existing regulations, leading them to be able to freeze markets. This implication of CRAs could constitute a hindrance to the development of future legally binding rules, both in the EU and in the USA.

In line with what was done in 2009, the EU reinforced its powers by creating the European Securities and Markets Authority (ESMA) in 2010, replacing the CESR and aiming to centralise all aspects of CRA regulations in the EU. This point was confirmed by the Regulation (EC) No 513/2011 [26], known as ‘CRA II’ (May 2011). The major improvement made by CRA II is to give to the ESMA the status of only supervisory power in the EU for registering and supervising CRAs, in order to benefit from advantages of centralisation (no risk of bias from national authorities and homogeneity of rules across countries). Other regulation aspects present in CRA I are kept and remain unchanged in CRA II.

The excessive reliance from investors on ratings was pointed out by various market players and institutions representatives during a roundtable which took place at the European Commission on 6 July 2011 [27]. This point was targeted in the latest EU regulation, known as ‘CRA III’.

The Regulation (EU) No 462/2013 [28], known as ‘CRA III’ (May 2013), is a new big step forward made by the EU. Consumers and investors protection is a clear objective of the text, which includes shareholders and members of CRAs in its scope. The regulation calls for a reduction in the reliance on external ratings, preventing all European Authorities, the European Parliament and the European Council from using reference to external rating in ‘their guidelines, recommendations and draft technical standards where such references have the potential to trigger sole or mechanistic reliance on credit ratings’ (Article 5b-1). CRAs are also told to put in place and to document an ‘effective internal control’, and to clearly underline that their ratings are only an opinion. Finally, tougher rules in the rating process of structured finance products are included in the text, making their rating recognised only if the issuer appointed at least two independent CRAs.

Intermediate conclusion

It is clear from what was exposed previously that, prior to the disastrous situation faced during the 2007–2008 financial crisis, the USA had a quite effective regulatory framework, while the EU had an obviously inadequate set of rules targeting CRAs. The financial crisis, followed by the sovereign debt crisis faced by the EU, pushed authorities to take further and stronger actions in order to better regulate the industry of CRAs. If the goal of the newly introduced texts, namely to restore confidence in markets and reduce the reliance on CRAs ratings, is shared between the EU and the USA the areas targeted differ between the two blocs. Indeed, the USA call for more transparency in the registration requirements and for the removal of references to external ratings from existing rules, while the EU mainly focuses on the quality and the clarity of ratings and ratings computations. This wish of better quality of ratings is probably induced by the sovereign debt crisis that some EU members had to face since 2010, and during which the accuracy and the objectivity of ratings were questioned (a more detailed study of this phenomenon can be found in Gärtner et al. [29]). Despite these differences, John Coffee Jr. underlines in a 2011 paper [30] that the EU and the USA tend to converge on some areas of their regulation, e.g. in targeting conflicts of interest.

It also turns out that the implementation process of the new binding rules is relatively slow in both the EU and the USA. This process also highlights the unexpected power that CRAs effectively have on markets and the paradoxical aspect of the current situation, because this power partly comes from previous regulations such as the CRD directive. This phenomenon, referred to as ‘regulatory license’ by Frank Partnoy [5], is a great hindrance to a smooth and swift implementation of the regulations in the USA and in the EU, although this phenomenon is less pronounced in Europe where there is historically a lower variety of debt issuers.

Despite this difficult path towards full implementation, the European Securities and Markets authority (ESMA) points out some tangible progress through time. After having underlined failures in a 2012 report [31, 32], the regulator launched a consultation paper in December 2012 to better address these issues and implement the new rules more efficiently [33]. As highlighted in a 2013 press release [34], the CRAs made some improvements, but some areas still had to be improved, such as the monitoring and the surveillance of ratings. These issues were addressed by CRAs, as the ESMA does not point them out in a 2015 press release [35], although other issues are underlined in this document. Moreover, in an October 2015 press release [36], the authority recognises that the EU CRA Regulation has improved the governance and operation of CRAs. It can be said from this institutional literature that the implementation of regulations on CRAs was indeed tough, but some objective and relevant progresses were made since the implementation of CRA I.

Finally, it seems interesting to underline that, since the implementation of regulation 1060/2009, the number of registered credit ratings agencies has remarkably increased to reach 27 in the EU by the end of 2011 according to the ESMA website [37]. This point could appear as being problematic, as Becker and Milbourn highlight in a paper [38] that increased competition leads to a decrease in ratings quality. Conversely, Behr et al. [39] underline that a lower level of competition associated with a stronger regulatory framework also leads to a deterioration in ratings accuracy. These mixed conclusions certainly explain why competition is not included in the scope of European regulations. However, in spite of the great increase in the number of players since 2009, it is important to note that competition has not been affected due to the persisting unequal distribution of market shares (ESMA [40], p. 10), allowing us to make a study without distortions in the data.

To conclude, it can be said that the recent events demonstrated that there is room for a tightening of rules regulating the CRA industry. Even if undeniable efforts are made in order to correct mistakes that were previously made, the way towards a global and efficient regulation of CRAs seems to be long and difficult.

Data collection and variables

The aim of this paper is to determine whether the EU regulation had an influence on the rating process of Moody’s. It is important to focus on that specific point because Dimitrov et al. [41] pointed out that the US regulation that entered into force in 2010 (Dodd–Frank Act) had no impact on CRA discipline or accuracy. Thus, to conduct a study at the EU level appears to be necessary in order to see whether the European regulation is better designed than the one in the USA.

We focus only on ratings issued by Moody’s in order to assess the efficiency of the first EU implemented regulation on credit rating agencies, given that this CRA is the only one giving free access to all rating actions that were taken. Moreover, it is reasonable from our point of view to assume that Regulation 1060/2009 had a similar impact on the three main rating agencies, allowing to study its repercussions on a single entity. To curb the influence of the global crisis faced in 2007–2008, we collected both rating actions (up- and downgrades) and announcements (reviews, first-time assignments) between 1 January 2009 and 31 December 2011. We, moreover, only focused on Eurozone firms publishing financial statements and balance sheet labelled in Euro, this in order to avoid any bias induced by currency conversion during the collection of the data.

We kept three types of rating in our sample: the corporate family rating (CFR) targeting the entire company and considering it as having a single class of debt, and the ratings of senior secured and unsecured debt, these types of debts being the first ones to be reimbursed in case of bankruptcy of the firm. For each type of debt, we only considered ratings covering the overall category and not targeting a specific emission.

At the company level, we took into consideration specific types of entities and cases. The head of a group was always included in the database through the CFR. For subsidiaries, we only considered rating announcements or actions specifically targeting the subsidiary and not the consequences of a CFR modification. As a result, subsidiaries having their ratings modified through the CFR were not included in our sample. We finally decided not to include financial-related firms and banks in order to avoid fear-driven rating actions or announcements.

For each company we compiled profitability, solvency and liquidity indicators as well as information derived from the balance sheet (such as total assets); almost all these dataFootnote 1 come from AMADEUS (Analyse MAjor Databases from EUropean Sources—Bureau van Dijk), a database compiling accounting data of European companies. This allows us to avoid any bias induced by regional and/or country-specific definitions of accounting indicators and gives a necessary homogeneity across observations of the sample.

Given that Moody’s does not give a specific rating when a company defaults, we also checked the current status of each firm appearing in our database with a speculative grade. This will allow us to measure the repercussions of the 2009 regulation on what is called false warning, i.e. a company does not default within a year after being rated in the speculative category.

We obtained a sample containing 338 rating observations (97 before CRA I and 241 after the regulation) linked with various accounting indicators, each of them being the one appearing in the latest balance sheet of the company prior to the rating action. This relatively small sample is explained by the shortness of the analysed period in order to obtain data not ‘contaminated’ by the financial crisis or the second European regulation on credit rating agencies (‘CRA II’), which was implemented in May 2011. However, it is important to underline that, even if not very large, our sample contains a sufficient number of observations to obtain robust results from an econometric point of view. Moreover, the limitation of the period allows our database to avoid ‘contamination’, which certainly reinforces the robustness of the results showed later in this study.

We decided to follow as closely as possible the quantitative methodology adopted by Moody’s during its rating process, which is based on three main categories of balance sheet items.Footnote 2 The first one is the scale of the company, which is measured by the total amount of assets detained by the company. The second category encompasses profitability and efficiency indicators, with four variables: profit margin (in %), cash-flow variation (in % compared to the previous year, indicates the ability of the company to generate cash), liquidity ratio (showing how a company is likely to face short-term debt obligations) and solvency ratio (in %, measures the ability of a company to face both short-term and long-term debt obligations). We also included the leverage by computing the long-term debt leverage, defined as total long-term debt divided by total assets.

We finally created a dummy variable in order to assess the impact of reviews for downgrade, this variable equalling one if the company’s rating was placed under review for downgrade at most 6 months before the action (upgrade, downgrade or confirmation of the rating). A similar dummy associated with reviews for upgrade was not created as no observation of such reviews was made before CRA I, leading to the impossibility to assess the impact of the regulation on this type of review (Table 1).

Table 1 Descriptive statistics of variables
Full size table

A correlation matrix is presented in “Appendix 2”. It can be noticed that a great majority (425 out of 435, i.e. 97.7%) of the observed correlations are below 0.3 in absolute value, which can be considered as a threshold for problematic correlations. Moreover, 9 of the 10 remaining pairs are correlated below 0.4 in absolute value, which is not econometrically problematic.

CRA I influence on ratings

Our study has similar aims as the one of Dimitrov et al. [41], namely studying the influence of a regional regulation on credit rating agencies. However, our contribution is different from an econometric point of view: Dimitrov et al. measure the overall impact of the Dodd–Frank Act through a single dummy variable, while we decided to use time-varying coefficients to test change in behaviour from Moody’s. This will allow a finer study of the impact of CRA I on ratings.

This paper can also be related to Alsakka et al. [42], as this study measures the influence of the establishment of the ESMA (induced by CRA I) on the reactions of stock returns to banking rating actions. Although the main goal of our work and Alsakka et al.’s work is to measure the influence of CRA I, our study clearly distinguishes from Alsakka et al. on three mains points. First, we focus on all industrial sectors but the financial one, i.e. the exact opposite of Alsakka et al. Second, these authors use an event-study methodology, while we adopted an ordered logistic configuration. Finally, we do not focus on market reactions, but on the determinants of ratings. In spite of these differences, it could be interesting to compare our results with the results of Alsakka et al., as they do not find clear evidence on the effects of the new regulatory regime. This will be done later in this paper.

One of the aims of our study is to determine whether CRA I had an impact on the way rating agencies decide to assign or to modify ratings of companies. As we use grades assigned to corporate bonds we decided to estimate an ordered logistic model in this part, using a numerical conversion of ratings ranging from 1 to 20 (i.e. from Ca to Aaa, see Table 2) as dependent variable.

Table 2 Numerical conversion of ratings
Full size table

As a first approach, we decided to run an OLS model using all variables presented previously. We also created new variables, based on the existing ones and obtained by multiplying each existing variable by zero if the rating action was operated before the entry into force of CRA I, and by one otherwise. The newly created variables capture the potential breaks in data before and after CRA I, leading to the possibility to run a Chow test on the model. This procedure tests each coefficient of the new variables against zero and leads in our case to the conclusion that CRA I had an effect on at least one parameter of our model. Following this, it seems interesting to have a finer look at this impact and on its direction for each parameter.

An efficient way to do so is to estimate an ordered logistic model using the original variables and their interaction with the CRA I dummy. The first advantage of this procedure is to allow a finer treatment of subsets (in our case before and after CRA I) with different variances into a single model compared to the pooled specification we used previously. It secondly allows to disentangle the effects of CRA I on each variable and to see precisely their direction and significance. The model is presented on the left-hand side of Table 4 and follows the equation presented below:

$$\begin{aligned} {\text{Current}}\,{\text{grade}}_{i} & = \left( {\alpha + \delta_{1,i} } \right) * {\text{Profit}}\,{\text{margin}}_{i} + \left( {\beta + \delta_{2,i} } \right) * {\text{Cash{-}flow}}\,{\text{var}} ._{i} \\ & \quad + (\gamma + \delta_{3,i} ) * \log \left( {{\text{total}}\,{\text{assets}}} \right)_{i} + (\varepsilon + \delta_{4,i} ) * {\text{LT}}\,{\text{debt}}\,{\text{leverage}}_{i} \\ & \quad + (\zeta + \delta_{5,i} ) * {\text{Solvency}}_{i} + (\eta + \delta_{6,i} ) * {\text{Liquidity}}_{i} \\ & \quad + \theta_{i} * D_{\text{CRAI}} \times D_{\text{reviewed}} + \iota_{i} * D_{\text{country,period}} \\ \end{aligned}$$

with

  • α,…, η coefficients of variables before CRA I,

  • δ 1,i ,…, δ 6,i measuring the impact of CRA I on the influence of each variable (equalling 0 if the rating action/announcement of observation i occurred before CRA I),

  • θ i capturing the interaction between CRA I and the fact of being reviewed for downgrade (baseline: before CRA I, not reviewed), and

  • ι i coefficient of the dummy variable combining the country of the observed company and the period of the observation (before or after CRA I) (baseline: Germany, before CRA I).

In this setting, coefficients measure the impact of a variation of each indicator on the probability that the assigned rating belongs to a higher level, holding everything else constant. Thus, it can be said that each coefficient can be interpreted as measuring the overall influence of the variable on the rating. As an example, it can be inferred from Model 1 (Table 4) that a higher profit margin tends to have a significant and positive influence on ratings in both periods. A specific attention will be given to the interaction between dummy variables later in this part.

After having looked at the influence of continuous variables on ratings, an interesting question is to see whether CRA I had an impact on the way Moody’s uses these indicators. A way to do so is to test whether the coefficients are statistically different before and after the implementation of CRA I. The results presented in Table 3 clearly show that CRA I modified the way Moody’s uses some accounting and financial indicators. It turns out from our model that Regulation (EC) No 1060/2009 led to a more cautious use of ‘profit margin’ and ‘long-term debt leverage’, as we observe a significant decrease in the absolute value of the coefficient.

Table 3 Comparison between coefficients—before CRA I (97 obs.) versus after CRA I (241 obs.) (Model 1)
Full size table

A new point has to be made about the interaction between dummies, presented just above geographical indicators in Table 4. The baseline for comparison is a company observed before CRA I which was not reviewed for downgrade. Table 4 confirms the impact on companies of being reviewed before CRA I, as we observe a significantly negative coefficient for this interaction. However, due to collinearity issues, it is not possible to entirely measure the influence of CRA I on reviews for downgrade.

Table 4 Model 1: Ordered logistic regression of ratings (lhs) and influence of variables across rating categories (rhs) (Pseudo R2 = 0.1407)
Full size table

Finally, it turns out from Model 1 that CRA I had some repercussions on how the domicile of a company influences its rating, as some coefficients of country dummies are statistically different across the two periods. As an example, coefficients associated with Belgium underline that Moody’s tends to be more severe with Belgian companies after the implementation of the regulation. It can, moreover, be highlighted that all coefficients associated with these dummies are negative for the post-CRA I period, leading us to assert that Moody’s tends to be more severe since CRA I is applied, independent of the nationality of the company.

Results shown in Tables 3 and 4 demonstrate that CRA I had repercussions on how Moody’s employs some quantitative indicators and geographical information. It can, however, be surmised that the relative influence of each variable is not linear and varies across rating levels. This hypothesis is tested on the right-hand side of Table 4 by considering four levels of ratings: B3 (low grade), Ba1 (middle grade), A2 (high grade) and Aa2 (really high grade). In this table, each coefficient measures the influence of variables on the probability of belonging to the considered grade. Significances are reported with stars: (***) for the 1% level, (**) for the 5% level and (*) for the 10% level.

As it was asserted, the relative influence of indicators is not linear among rating categories. A cogent example is the one of log(total assets) (after CRA I), as we observe a switch in coefficient sign across categories and a greater influence of this variable for ‘non-extreme’ ratings. The first observation is not surprising at all, the negative (positive) sign noticed for ‘bad’ (‘good’) grades meaning that a greater amount of total assets reduces (increases) the probability of belonging to such a category. The greater influence of variables on ‘non-extreme’ ratings, also observed for the profit margin and solvency and liquidity ratios after CRA I, tends to prove that Moody’s relies more on other indicators (such as expectations on the future of the considered company) for very bad and very good grade levels.

Even if only a few coefficients associated with countries are significant, it is, however, possible to draw some conclusions about the geographical influence on ratings. Firstly, for all countries and all periods, a reversal of sign is observed when the rating goes from Ba1 to A2. This leads us to assert that Moody’s considers each country as being beneficial or detrimental for its companies, depending on their overall economic health. Second, the impact of nationalities is almost always greater for ‘non-extreme ratings’, following the observation made previously.

From what was previously shown, it can be said that CRA I constrained Moody’s to modify its rating methodology as we observe significant variations after CRA I compared to the baseline situation. Even if some of the underlined effects are not statistically significant, it can be said that CRA I attained some of its targets. A further research question is now to determine whether CRA I also had effects on the magnitude of rating movements operated by Moody’s.

CRA I influence on magnitudes

In this part, we restrict the sample to rating actions (upgrades and downgrades), ratings announcements being associated with a magnitude of 0. The retained variables and the methodology used in the previous part remain the same. The Chow test leads us to the conclusion that CRA I had a significant impact on magnitudes of moves.

This intuition is confirmed by Table 5 and by Model 2 (Table 6), based on the same methodology as the one used for Model 1.

Table 5 Comparison between coefficients—before CRA I (50 obs.) versus after CRA I (95 obs.) (Model 2)
Full size table
Table 6 Model 2: Ordered logistic regression of magnitudes (Pseudo R2 = 0.3303)
Full size table

Interesting conclusions can be derived from Model 2 and Table 5. The first one is that, after CRA I, not reviewed companies tend to face greater magnitudes of changes in line with what could be expected. It also appears that CRA I had an effect on the influence of reviews for downgrade on rating actions. A major increase in the coefficient is observed for reviewed firms when moving from before to after CRA I, meaning that reviewed companies are more likely to face a greater magnitude after the entry into force of the regulation.

As this result might seem counter-intuitive at the first sight, we have to remind that ‘− 1’ (downgraded by one notch) is greater than ‘− 2’ (two-notch downgrade), leading to the conclusion that CRA I reduced the absolute value of downgrades, but did not turn them into upgrades.

This point is confirmed in Table 7, as no structural difference arises between the two periods. It can then be concluded that CRA I led Moody’s to adopt a less severe attitude when it downgrades companies previously under review.

Table 7 Distribution of magnitudes—before and after CRA I (reviewed companies only)
Full size table

Our second model also sheds light on some influence of the country on the way Moody’s assigns grades. It is important to notice that, in this setting, dummy variables without any observation were dropped in order to limit collinearity issues faced in Model 1. These variables are the ones for Austria, Estonia, Finland, Ireland and Luxembourg.

It is also important to underline that some of the impacts of CRA I highlighted in this part might be strengthened by other factors, such as the economic cycle. However, the conclusions derived from Model 2 are statistically robust and can be imputed (at least partially) to the regulation.

The first observation derived from this model is that Moody’s changed quite heavily its mind on how to take nationalities into account before and after the reform. It can indeed be noticed in Table 6 that all coefficients of dummies ‘before CRA I’ are not significant, while almost all geographical variables after CRA I are significant at the 1% level.

Secondly, it turns out from Model 2 that Regulation 1060/2009 led to a tougher way of modifying grades from Moody’s, as a sharp decrease in coefficients is noticed for each country when moving from before to after the reform. This means that, in case of modification of the existing rating, the probability of being upgraded or not severely downgraded has been reduced by the entry into force of CRA I.

Finally, Table 5 shows that our conclusions about the influence of countries are robust, as a majority of the associated dummies are significantly different between the two periods. This leads us to assert that CRA I had some impact on how Moody’s uses the nationality of a company during the grading process.

Contrary to Alsakka et al. [42], our study points out some effects of CRA I on the way Moody’s assigns and modifies its ratings. It can be inferred from Tables 3 and 5 that CRA I had some impact on a great majority of the variables that are used in our models, but also that some of them influence ratings, while the others have repercussions on how ratings evolve. Even if the observed effect is quite low, it can be said from our study that CRA I had an influence on the way Moody’s uses almost all accounting parameters during the rating process. It can finally be concluded that CRA I led Moody’s to smooth its downgrades and reduced the relative impact of some variables on ratings.

CRA I influence on false warnings

Another way to measure the influence of CRA I on Moody’s is to have a look at false warnings, which can be viewed as a proxy measuring the quality of the ratings. In order to measure these warnings, we only took companies with a non-investment grade (i.e. below Baa3, or strictly smaller than 11) into account and we observed whether each firm defaulted within a year after this non-investment grade was assigned. If this situation did not occur we considered the corresponding dummy variable as equal to 1, and to 0 otherwise. As we use a dummy as dependent variable, a test of differences in proportions between the two periods is used, the null hypothesis being that this difference is equal to zero. The following results are shown in Table 8.

Table 8 Proportions of false warnings, before and after CRA I
Full size table

As we want to know whether a reduction in false warnings occurred, we focus on the difference in the respective proportions. If CRA I effectively had an impact on the issuance of false warnings, this difference should be significantly greater than 0, i.e. prop(before) > prop(after), which is not the case according to Table 8. Moreover, this hypothesis is the most strongly rejected among the three alternatives, leading to the conclusion that CRA I was inefficient in reducing the proportion of false warnings released by Moody’s.

Conclusions

The historical review presented in the first part of this paper showed how a regulatory framework targeting rating agencies was necessary to be put in place, but also demonstrated that such a change would be uneasy to make. In addition, this review sheds light on the great role acquired by CRAs in both financial markets and existing regulations prior to the 2007–2008 global crisis, underlying the great need to reform this type of companies.

Based on a sample of about 350 observations, our models allow to conclude that CRA I had some influence on Moody’s ratings. It first turned out from Chow tests that the regulation modified both the grading process and magnitudes of moves that occurred. The first specification of ordered logistic regression demonstrates that CRA I had a significant impact on how Moody’s uses some quantitative variables and geographical information in order to determine which grade will be granted to a company. This model also underlines a more cautious use of indicators after the implementation of the regulation.

Similar conclusions can be derived from Table 5 on the consequences of CRA I on magnitudes of rating changes. Model 2 highlights the fact that Regulation (EC) No 1060/2009 led to more limited notches moves when reviewed companies are up- or downgraded, this effect being statistically significant at the 5% level. Following this regression, the test on false warnings clearly shows that CRA I failed to address this issue, as we do not observe a significant difference between the two periods.

It is certain that, in spite of the robustness of the models we used, our conclusions could be strengthened, notably by using data from the three major CRAs and by considering a greater range of years. This would certainly help to obtain even more robust results and to conduct a finer study on the impact of the recent EU regulations on ratings.

It is finally important for us to underline that the findings of this paper are not only relevant within the theoretical debate on CRAs, but also consistent from a practical point of view. Indeed, our study of the impact of the first important legally binding text on the regulated entities in the EU can directly be used by legislators in order to improve further the credit rating market.