Information asymmetry and credit rating? Evidence from a Quasi-natural experiment in China

Xiaolu Hua Jing Shib

a School of Economics, Finance and Marketing, RMIT University, Australia

bSchool of Economics, Finance and Marketing, RMIT University, Australia

ABSTRACT

This paper examines how the information asymmetry of credit rating industry be alleviated by the entry of an independent rating agency, China Credit Rating (CCR), which utilize a combination of public utility and investor-pay business models. Using a difference-in-difference approach to compare the ratings by issuer-pay incumbents for CCR covered firms with uncovered firms. We find with respect to the uncovered group, the incumbent issuer-pay agencies significantly give more fair ratings for the CCR covered group. This result adds empirical evidence on literature documenting the influence of introducing a new rating agency with alternative business models differentiated from issuer-pay rating agencies. We further show that the more reputable issuer-pay rating agencies and firms with better information environment experience a greater reduction of information asymmetry after CCR entrant. .

Keywords: Credit ratings Information asymmetry Investor-pay rating agency Public rating agency

1. Introduction

Since the outbreak of sub-prime crisis of 2007-2009, criticisms on credit rating agencies (CRAs) such as Moody’s, S&P and Fitch have regularly made the headline. In particular, IMF (2009) estimated that the losses on structured financial products was around $4 trillion. A lot of arguments and reform proposals focus on CRA’s issuer-pay business model, which may cause conflict of interest then rating inflation. This study estimates how the entry of a CRA with a new business model alleviate the rating inflation problem of the incumbents, and to some extent correct the information asymmetry, based on a quasi-natural experiment in China.

From investor-pay to issuer-pay

Originally, with the purpose to avoid the reverse selection problem between investors and issuers, private CRAs emerged to solve the information asymmetry in the 1900s (White 2010), when investors paid CRAs for their rating reports. But from the early 1970s, this charging model began to change to the “issuer-pay”, which means the CRAs charged the issuers according to the size and type of the issue. There are two main reasons for this change. Firstly, the widespread use of photocopy machines leaded to the problem of “free-rider”, which decreased CRAs’ revenue. CRAs even did not have enough money to support their operation (White 2002). The second reason is the urgent demand of issuers for credit ratings from 1970. In 1970, Penn Central failed to pay back its $82 million commercial paper (CP). And together with the 1970 recession, investors started to doubt the solvency of a lot of companies’ CPs, which was booming from the 1960s. When faced by the liquidity crisis

1

caused by investors’ refusal to roll over their CPs, issuers wanted to reassure those panic investors by actively seeking credit ratings (White 2002). And then a regular market pattern was established that new debt issues must had at least one credit rating when they came to the market. This increasing rating demand let CRAs find they can impose charges on issuers. Moody’s and Fitch changed their charging models in 1970, and S&P followed in 1974. Nowadays, a majority of CRAs accept issuer-pay model and collect most of their incomes from the companies they rate. Among the current 10 NRSROs (nationally recognized statistical rating organization), there is only one exception, Egan-Jones Rating Company (EJR), still applies the investor-pay model.

Many researchers and policymakers argued that the current charging model may encourage CRAs to give favourable ratings to issuers (the conflict of the interest), which will lead to the rating inflation and descending rating accuracy, an abundantly studied moral hazard dilemma. As a result of their adoption of an issuer-pay model, CRAs have a preference for customers who have larger issues (Bolton, Freixas & Shapiro 2012; He, Qian & Strahan 2011) or who are repeat issuers (Bolton, Freixas & Shapiro 2012), resulting in inflated ratings for these issuers. The accusers point out that issuer-pay CRAs fear to lose the clients for their next issues. Then the delays in adjusting ratings to reflect the changing financial situation of the issuers and the bias to the large companies are criticized. Sean Egan of the Egan Jones investor-pay CRA views the issuer-pay model is irredeemable, citing evidence from specific past failures, such as Enron and Parmalat and certain structured products in the sub-prime crisis (Nielsen 2013).

Comparison between issuer-pay and investor-pay

A bunch of studies aim to compare different business models of CRAs. And the question of who pays the credit rating seems to matter. For example, Beaver, Shakespeare and Soliman (2006) compare the ratings between Moody’s and EJR, resulting that EJR’s ratings as an more independent rating from the recognition and regulatory purpose, provide more accurate and timeline information to investors than Moody’s. In Skreta and Veldkamp (2009)‘s model, the issuers are inclined to release the most favourable rating through rating shopping among different issuer-pay CRAs with a naïve investors assumption. They suggest this situation can be alleviated by changing to investor-pay model. Pagano and Volpin (2010)'s model advocates that the adoption of investor-pay model can improve the ratings accuracy. And they suggest a more modest method that keep the issuer-pay CRAs but require issuers to pay an upfront fee irrelevant to the rating results, to avoid the rating shopping. Jiang, Harris and Xie (2012) test S&P’s ratings before and after it adopted an issuer-pays model, taking Moody’s as the benchmark. They found that the ratings of S&P were higher afterwards, suggesting that the issuer-pay model leads to rating inflation. They also undertook cross-sectional analysis to identify whether ratings increased more for bonds that might generate more conflicts of interest, with the result that an increase in ratings is correlated with inherent conflicts of interest. Bongaerts (2013) sets a model with the rational agents’ assumption, showing the investor-pay model, the issuer-produced ratings or the mandatory co-investments model can all improve social welfare, and the very high degrees of regulatory intervention are essential to make the above alternative models take hold. Cornaggia and Cornaggia (2013) carry out a research on another subscriber-pay CRA, Rapid Ratings, to conclude that this agency gives more reasonable, timeline and accurate ratings for investors compared with Moody’s who is inclined to satisfy the issuers. The very recent study is from the theoretical model developed by Kashyap and Kovrijnykh (2016). They analyse optimal compensation schemes for the CRAs, and the results differ according to which party pay for the ratings, a

2

social planner, the issuers or the investors. Ratings are found to be of larger errors when issuers order the ratings than when investors order. Some further find the entry of a new issuer-pay CRA is inclined to deteriorate the rating quality of the incumbent, aggravating the rating inflation, such as Bongaerts, Cremers and Goetzmann (2012). They argue that Fitch, a new issuer-pays entrant, brought no additional useful information to investors, and its entry and the consequent direct competition drove existing agencies to raise their ratings in order to keep their customers. For the structured financial products such as CDS, because CRAs directly engaged in the designing process, the conflict of interest is argued to be even worse under issuer-pay model (Richardson & White 2009).

Many reasons are supposed to explain why issuer-pay model causes rating inflation and inaccuracy. Rating shopping is thought to be one essential reason, through which issuers can take advantage of the issuer-pay business model to increase the ratings for both the corporate bonds and the structured products. The issuers may choose the highest rating from the preliminary rating process, here issuers are assumed to always pursue higher ratings. This rating shopping may pressure a CRA to increase the rating, causing rating inflation (Skreta & Veldkamp 2009). Fennell and Medvedev (2011) also find the evidence of rating shopping from their interviews. Bongaerts (2014)’s model finds even with rational investors, rating shopping still exists. In principle, rating shopping can be exacerbated by competition, and competition itself is also argued to be responsible for rating inflation and inaccuracy. Bolton, Freixas and Shapiro (2012)’s model shows lower reputation cost, fierce competition and more naïve investors give CRAs incentives to inflate ratings, and monopoly should be more efficient to solve rating shopping problem. Bongaerts, Cremers and Goetzmann (2012) identify a tie-breaker role Fitch plays when Moody’s and S&P disagree with each other, and this role creates an incentive for issuers to seek a favourable rating from Fitch. Moreover, Becker and Milbourn (2011) suggest that the entry of Fitch is coincide with lower quality ratings from the incumbents such as Moody’s and S&P, so that rating levels increased, the correlation between ratings and market-implied yields fell, and the ability of ratings to predict default deteriorated. Their empirical study accepts how competition weakens the reputation mechanism thus leads to higher rating inflation severe, while they reject rating shopping as the mechanism.

Other proposed business models and debates

Through the above studies comparing issuer-pay with investor-pay models, the latter receives a high level of support. Apart from the investor-pay business model, investor-produced model, Public Utility model and platform model are also proposed by scholars. For example, Bongaerts (2014) introduced the investor-produced model in his study, where the rating agencies are also the end-users of ratings. Diomande, Heintz and Pollin (2009) and Lynch (2010) suggest US to launch a public CRA, argue that the conflict of interest can be managed by modelling on other successful public independent bodies, such as the Federal Reserve or the Supreme Court. Others like Mathis, Mcandrews and Rochet (2009) and Richardson and White (2009) proposed to introduce a platform as an intermediary between issuers and CRAs, to prevent the issuer from influencing ratings. And it is this platform’s responsibility to assign the rating request to a CRA, either randomly or based on the CRA’s rating quality.

Not only the academic studies, but also many regulatory proposals call for changing the issuer-pay model. For instance, in the United States, the Dodd-Frank Act in 2010 brought in a wide range of regulatory reforms; with the purpose of reducing the conflict of interest and increasing the rating transparency. While regarding issuer-pay vs. investor-pay, it demands an

3

assessment of the conflicts-of-interest by the Office of the Controller General. Another particularly relevant proposal to this paper was the ‘Franken Amendment’, which suggested that for structured products, SEC should introduce a platform to bring all NRSROS and issuers together through an independent organization, which is exactly the platform model as what Mathis, Mcandrews and Rochet (2009) and Richardson and White (2009) have proposed. But after two-and-a-half year's study by SEC, the Franken Amendment was not implemented by the government1. In Europe, the Group of Twenty, the Financial Services Authority, the Financial Stability Board and the OECD have all presented proposals to credit rating industry reform. The most related proposals to my study are about the business model change. For example, the European Commission Consultative Paper (European Commission 2010) proposed to “examine the possibility of creating a public European CRA to compete with the private sector”, which is the Public Utility model, and “to explore ways to mitigate conflicts-of-interest in the issuer-pay business model, especially considering a turn back to investor or subscriber pays, a mixture of these two and alternative approaches involving a third party paying or hiring the services”.

However, there are still debates on which model should be used. Some people do not believe the issue-pay model is at fault. For example, big CRAs, such as Moody’s, S&P and Fitch argue that the conflict of interest is totally manageable (Becker & Opp 2013; Coface 2010; Pagano & Volpin 2010), and CRAs may consider the long-term revenue, to balance the rating fee collected from the issuers who ask for high ratings with the reputation cost (House 1995). Moreover, some arguments arise that other business models can also cause rating inflation and inaccuracy because of investors’ and regulatory overreliance on credit ratings. A various kinds of ‘certification’ role are allocated to ratings, whereby ratings act as a credit-quality threshold or gatekeeper in financial contracts. Here, CRAs' four typical certification roles in the use of contracts and market practice are listed by Deb et al. (2011) in their study. Aside from the use in contracts, ratings are also widely used within the regulatory framework such as supervisory policies (Sec 2002). One most pervasive example is determining net or regulatory capital requirements for banks, securities firms and insurance companies (Gonzalez, Sotelino & Savoia 2011). This overreliance on ratings makes not only the issuers but also the investors prefer the higher ratings to satisfy the benefit target and the regulatory requirement, therefore this reliance deteriorates the efficacy of investor-pay CRAs from reducing the rating inflation. Academically, the model built by Opp, Opp and Harris (2013) incorporates the regulatory purposes of ratings, finding that after considering rating-contingent regulation, CRAs provide more higher ratings, no matter the higher ratings reveal more or less information. This is highly consistent with the empirical finding from Cornaggia and Cornaggia (2013), that Moody’s is willing to give relatively higher ratings for those bonds at marginal investment grade, in order to make those regulated investors convenient to buy. Compared with the issuer-pay business model, Cole and Cooley (2014) suggest that the regulatory reliance is more likely to cause rating inflation and inaccuracy.

Furthermore, debates also focus on the disadvantages of other alternative business models aforementioned. For investor-pay model, the free-rider problem is sought to be the main challenge (Richardson & White 2009). As mentioned before, free-rider problem is one of the reasons why CRAs changed from investor-pay to issuer-pay in the 1970s. Although the lower subscription fee can alleviate this problem somewhat, investor-pay CRAs will still find it difficult to take advantage of economies of scale and compromise their rating quality or leave this industry (Fennell & Medvedev 2011). Other arguments on investor-pay model are

1 See Becker and Opp (2013), SEC (2012), SEC (2014).

4

pressure from investors to require high ratings (Pagano & Volpin 2010), regulatory use of ratings (Goodhart 2008) and how to use ratings for structured products (Fennell & Medvedev 2011). From the recent model of Kashyap and Kovrijnykh (2016), although investor-pay model provide more accurate ratings, the investors ask ratings too often thus damper the social efficiency. For the investor-produced model, investors themself may become the party benefiting most from rating inflation, in terms of lower capital charge (Gonzalez, Sotelino & Savoia 2011; Sy 2009). The model developed by Becker and Opp (2013) shows that if running under the existence of issuer-pay CRAs, the investor-produced ratings are slightly more inflated. This result is in line with the suggestion of Behn, Haselmann and Vig (2014) that the bank internal ratings are inaccurate and over-optimistic. Bongaerts (2014)'s model also shows even the investor-produced CRA can induce more accurate ratings, the traditional issuer-pay CRAs can cater better to issuers, thus make alternative models generate litter demands. The Public Utility model also cannot wipe off the conflict of interest because the government is also an issuer (Public Consultation on Credit Rating agencies. Nov 5, 2010). It is sceptical that the public CRA would bias to the sovereign or municipal bonds, or large companies owned by the government (Cinquegrana 2009). Another worry about the public model is whether the rating quality will decline due to insufficient budget fund. Others also consider how ratings of public CRA would be perceived by the market, and how to keep the rating quality higher given that there is no competitive market pressure (Fennell & Medvedev 2011). Without exception, the platform model is questioned as well. Along with the difficulty to decide which criteria using to choose one CRA for a rating request, the platform model also has the potential to decrease incentives for CRAs to provide high quality ratings as they might be chosen randomly and irrelevant to the rating quality. In addition, if a platform has a monopoly, it would have a role as a systemic regulator apart from a commercial role. This will cause the worry about a new conflict of interest (Fennell & Medvedev 2011). Moreover, Bolton, Freixas and Shapiro (2012) point that the Cuomo plan (issuers pay upfront for ratings) proposed by Pagano and Volpin (2010) can alleviate the conflict of interest while it cannot solve rating shopping. The issuers would like to cover their paid but poor ratings. Bongaerts (2014) suggests the main reason for rating inflation is the private benefit for issuers and investors, with the results from the author's model that the investor-pay model, the investor-produced ratings and the Franken Amendment may have very limited potential to improve the social welfare. To make any of them effective, the issuer-pay CRAs need to be banned.

As discussed before, the alternative models are all born with their own drawbacks, which cannot perfectly eliminate the conflict of interest, leaving no conclusive solutions. Therefore, in practice, the experiment on launching CRAs with alternative models is very limited. For instance, the French credit insurer Coface wanted to sell its investor-produced ratings to other investors, but this even never started (Coface 2010). Another example is that Markus Krall who is the senior managing partner at Roland Berger once tried to set up an European non-for-profit investor-pay CRA, but this plan was finally abandoned because of the lack of interest from investors (Nielsen 2013; Spiegel 2012). Moreover, the SEC did not pass the Franken Amendment, keeping the present issuer-pay CRAs prevalent on the market. Although we have two examples of Black Rock and PIMCO, who offer investor-produced ratings and their ratings are adopted by the association of State insurance regulators in the US, the NAIC for determining capital requirements, they only focus on Commercial Mortgage-Backed Securities and Residential Mortgage-Backed Securities respectively, thus not widespread in the whole industry.

The natures of the credit rating industry, the large economies of scale, the long-time taken to establish reputations, network externalities and the high fixed costs create a significant barrier

5

to entry into the market (Fennell & Medvedev 2011). This is one important reason to explain why there are few trails of the alternative models. So in general, the lack of the widespread practical cases of introducing other business models into the credit rating industry creates a barrier to approach empirical analysis on the effectives of those alternative ratings models.

Contribution of this study

Till now, the issuer-pay CRAs still dominate the rating industry around the world, and after so many regulatory proposals, no individual country or area widely apply the other alternative models. But China launched an independent CRA (China Credit Rating, CCR) in 2010, with a combination of Public Utility model and investor-pay model, providing a quasi-natural experiment chance to analyse if this CRA with the new business model can alleviate the problem of rating inflation. In my study, we apply the difference-in-difference approach to estimate if the rating inflation is relatively controlled for the companies covered by the independent CRA, compared with their counterparties those are not rated by this independent CRA. We find a significantly negative difference between the treatment group (covered by CCR) and the controlled group (not covered by CCR) in terms of their rating changes given by the incumbent issuer-pay CRAs. And this implies that this independent CRA plays a certification role in the credit rating industry, thus the incumbents adjust their ratings to the “benchmark” ratings provide by CCR, leaving the rating inflation relatively less severe.

The study of Xia (2013) is the closest to mine in that it focuses on how the rating initiation of an investor-pay CRA, EJR, affect the rating quality of S&P. The author concludes that S&P is more responsive to credit risk and its rating changes contain higher information content after EJR coverage. In general, S&P improves its rating quality facing the competition from an investor-pay CRA, and this change is because the reputational concerns are elevated by EJR’s coverage. The model of Bongaerts (2014), however, only believes the quality increase of issuer-pay CRAs as a response to the investor-pay CRA entrant is just a temporary outcome, which cannot last long. Analysis of Xia (2013) differs from mine in several important ways. First, we apply the difference-in-difference approach, which is suitable for a quasi-natural experiment. Secondly, EJR is a pure investor-pay CRA, which is different from the independent CRA we analysed in China. Third, we not only focus on reaction from one particular CRA, but try to summarize the general reaction of all the incumbent issuer-pay CRAs in China.

Furthermore, we want to find whether reputation affects the rating behaviours of the incumbents when they are faced with the entry of an independent CRA. Therefore, we apply the difference-in-difference analysis on subgroups. One group contains two reputable incumbent issuer-pay CRAs who have foreign ownership and they together account over 60% of the market share. Plus they are recognized by main regulatory organizations. The other group includes other incumbent issuer-pay CRAs, which are all national rating agencies. And we find the reputable CRAs react more strongly to the CCR entrant than the non-reputable CRAs, as the rating inflation phenomenon of reputable group declines more significantly for those CCR covered firms than the non-reputable group. It to some extent advocates the existence of reputation mechanism in credit rating industry, which is in line with the theory of reputational concern (Bar-Isaac & Shapiro 2013; Bolton, Freixas & Shapiro 2012; Chen, Shi & Zhang 2013; Covitz & Harrison 2003; Li 2013; Wang & Zhang 2013; Xia 2013; Zhang & Zhang 2010).

6

Based on our result, some would argue that China has a poorer legal environment than the developed markets such as those in the US and Europe2 (Li & Filer 2007), and that is maybe why we can find the effectiveness to of an independent CRA alleviate rating inflation. To test this potential sceptical argument on the applicability of our result, we use two ways. First, we divided the sample into two groups according to the marketization index (MI) suggested by (Fan, Wang and Zhu, 2011). The firms belong to the high-MI group come from the provinces with higher-than-median MI. Generally, these provinces have more mature legal framework and higher investor and consumer protection level, as well as more developed financial market and more sophisticated investors. Difference-in-difference method is applied to these two subsamples. Second, we divided our full sample into two groups based on the legal environment index (Law) (Fan, Wang and Zhu, 2011), which represents the maturity level of the legal environment of a province, indication the investor protection level. Similarly, the firms belong to the high-Law group come from the provinces with higher-than-median Law index, otherwise belong to the low-Law group. The results advocate that in higher MI or high Law index provinces, ratings for CCR covered firms relatively significantly decline compared to those not covered by CCR. But the same conclusion cannot be found for the lower MI or lower Law index provinces.

Additionally, this paper takes the sample selection bias into consideration. Propensity score matching method and Heckman two-stage approach are applied to solve the problem of endogeneity.

This paper contributes to a growing body of literature that examines how different business models of CRA affect credit ratings. The financial crisis has boosted a surge of research interest in CRAs, particularly with respect to how to make regulatory reforms in this industry and how to alleviate the conflict of interest thus relieve the problem of severe rating inflation. While it is not realistic to remove the current issuer-pay business model, many researchers and policy makers give proposal to modify the business model, such as the SEC and European financial authorities aforementioned. This paper analyses the influence of the real experiment to launch an independent CRA with new business model in China, making a supplement to the relevant theoretical models, empirical studies and government proposals.

The rest of this paper proceeds as follows: Section 2 analyses the status quo of China’s credit rating industry; Section 3 describes the emergence and business model of CCR; Section 4 represents the data and methodology; Section 5 explains the empirical results and Section 6 is the robustness test; Section 7 concludes.

2. Issuer-pay Credit rating agencies in China

2.1 The history of China’s CRAs

China’s credit rating agencies developed along with the growth of China’s financial market, especially the bond market. In the mid-1980s, China launched a corporate bond market, and CRAs emerged from then on. They can rate stocks, insurance companies, banks, bonds in domestic market. During that decade, China’s corporate bond market experienced a booming. In 1987, People's Bank of China (PBOC) issued “Temporary regulations on the management

2 Li and Filer (2007) introduce an Estimated Governance Environment Index (GEI) which considers five dimensions of governance environment: the political rights, rule of law, free press, the quality of accounting standards and the level of general trust. And from their result, China ranks the last with the lowest GEI.

7

of corporate bonds”. PBOC also encourages its branches to create credit rating departments in their own provinces, and from 1989, some of these departments became the independent CRAs. The number of CRAs surged to over 90 during this period. But back then, POBC set the coupon rate according to the prevailing 1-year bank deposit rate (add 40%), making ratings have no influence on issuing rate. Also, ratings were not disclosed to the public and they were just used by government authority to approve the issue application (Chen 2003).

This booming era lead to a disastrous bond default in the mid-1990. Massive borrowers refused to pay the interest or principal. In Liaoning and Jilin provinces, more than half of the bonds went to default. Thus the local government and PBOC had to pay off all around the world, and the cost was estimated from RMB 3 to 8 billion (Kennedy 2008). The fall of Guangdong Trust & Investment Corporation and the Asian financial crisis make the situation worse. In 1999, the government moved the right of corporate bond approval to the State Planning Commission (SPC, changed to National Development & Reform Commission, NDRC, in 2003). Shortly, the SPC applied a quota system and only the large SOEs with 100% bank guarantee could issue corporate bond. The coupon rate was set about 150-250 bps above the 1-year bank deposit rate (Fleisher 2008). At the same time, PBOC designated 9 of the existing 50 CRAs as the agencies who can rate the publicly issued corporate bonds in December 1997 and corporate bond must have ratings from these CRAs. They are China Chengxin Rating Co., Ltd (CCXR), Dagong Global Credit Rating Co., Ltd (Dagong), Shenzhen Credit Rating (predecessor of Pengyuan), Yunnan Credit Rating, Changcheng Credit Rating, Shanghai Far East Credit Rating (SFE), Shanghai Brilliance Credit Rating & Investors Service Co., Ltd (SBCR), Liaoning Credit Rating, Fujian Credit Rating Committee (predecessor of Lianhe). After that, China’s bond market and credit rating industry experienced low development in the next few years. Until 2003, NDRC for the first time claimed that all corporate bonds must be rated by CRAs who had the rating experiences from 2000, and then the number of qualified CRAs decreased to five from nine (CCXR, Dagong, SFE, SBRC and Lianhe). During this period, CRAs actually played a very small role in the bond issuing process. Because the approval of corporate bond was decided by NDRC, and the price of bond was set by the PBOC Monetary Policy Division’s Interest Rate Section. Moreover, all corporate bonds were guaranteed by a state owned bank or enterprise.

In the next few years, several milestones of China’s bond market appeared. Firstly, in 2005, PBOC created the short-term commercial paper (CP). The maturity of CP is one year or less, and all firms can apply to issue it without the requirement of guarantee, and the price is decided by the market. And PBOC authorized the same five CRAs aforementioned to rate CP. Secondly, in 2008, PBOC again created a new type of bonds, the midterm note (MTN) with same features of CP expect that it has longer maturity. Both CP and MTN are traded on interbank market, which is also regulated by PBOC. The volume of CP and MTN saw a rapid increase from then on (see Fig.1), and meanwhile the business of CRAs surged. More essentially, credit ratings have begun to have some influence on both the initial issuing rate and the price on secondary market (Tu 2006). This in turn proves that the effectiveness of China’s CRAs is highly related to the level of maturity of domestic financial market. That is also why we choose data from 2006. The third breakthrough was that in 2007, the authorizing power for corporate bond was split between NDRC and China Securities Regulatory Commission (CSRC). From then on, CSRC is mainly in charge of the bonds issued by listed companies on the Shanghai and Shenzhen exchange markets, which are called enterprise bonds. Meanwhile, NDRC started to focus on bonds issued with the main purpose of infrastructure construction. A large part of corporate bonds supervised by NDRC are issued by the state owned companies. Then enterprise bonds are issued and traded on exchange

8

market while corporate bonds are issued and traded on both exchange market and interbank market. After this separation, CSRC issued Provisional Administrative Measures on the Credit Rating Business in Chinese Securities Market to regulate the CRAs for enterprise bond. This authorization split added one more supervisors to the credit rating industry, which we will discuss in detail later.

Fig. 1 Volume and number of issues on China’s bond market

Note: Bonds including non-financial institution CP, MTN, Corporate Bond and Enterprise Bond Source: China Bond, Wind

Generally speaking, the power of China’s CRAs is appointed by the financial regulators as well, which is similar to the U.S. The first regulation on CRAs was the revised version of corporate bond regulation in 1992, issued by PBOC. It stated that the issuers “could” seek ratings from CRAs during their application process. Then in 1997, PBOC changed this to “issuance of corporate bond must have ratings”, which enforced the position of nine recognised CRAs. In 2003, China Insurance Regulatory Commission (CIRC) approved insurance companies to invest in corporate bond and only accepted the ratings from China Chengxin International Credit Rating Company Limited (CCXI)3, Dagong, Lianhe and SFE. In the same year, NDRC also gave five CRAs, CCXI, Dagong, Lianhe, SFE and SBCR, the qualification to rate corporate bond. And CSRC required securities companies must hire credit rating when they launch bonds in October 2003. In the next few years, this kind of requirements were applied on different categories of bonds, such as the subordinated debt of banks and insurance companies, ABS, CP, MTN and so force. Nowadays, all public issued bonds are required to be rated by the qualified CRAs.

2.2 The existing CRAs in China

Currently, there are in total nine issuer-pay CRAs in China who can issue ratings on bond market. The big four agencies in China are CCXI, Lianhe, Dagong and SBCR. From 2006 to 2015, for non-financial institutions' CP and MTN, more than 90% of the market is shared by these big four. And for corporate bonds, about 70% issuers are their clients (see Appendix 1).

CCXI was the first credit rating joint venture company in China. It was founded in August 1999 by China Chengxin Securities Credit Rating Company Limited (now China Chengxin

3 A joint venture of China Credit Rating Co., Ltd (CCR) and Moody's.

9

Credit Management Co., Ltd., CCCM) and Fitch (Fitch 1998). CCCM is a private CRA which was established in 1992, one of the first CRAs in China. CCXI then had the permission to rate corporate bonds. In July 2004, Fitch announced the divestment from CCXI as it wanted to focus on domestic market (Fitch 2004). The further reform of China’s bond market in 2005 attracted another investor, Moody’s. In September 2006, Moody’s bought up to the regulatory cap of 49% share of CCXI, and started to provide management and technical support on rating methodologies and training of analysts to the new joint venture company (Peopledaily 2006). Because CCXI acquired the rating qualification early and it has expertise adopted from two internationally famous CRAs, it has a generally higher market share in China’s bond market (see Appendix Table 1), then a higher reputation. CCXI has more than 200 employees at the end of 2015, and it rates issuers around the whole country. Till now, CCXI has the qualifications and licences from PBOC (1997), NDRC(2003) and CIRC(2003, 20134).

The Fujian Credit Rating Committee was formed in 1995, and in July 2000 Lianhe was established after recombination and name change of the former one. So from the beginning, Lianhe had the permission to rate corporate bond. In August 2007, Fitch came back to China’s market, and also bought up to 49% share of Lianhe from its original shareholder, Lianhe Credit Information Service Co., Ltd (LCIS). At the end of 2015, Lianhe has about 200 staff and its business is also throughout the country. It has the same qualifications as CCXI. Because CCXI and Lianhe are all joint venture firms, they do not have the approval from CSRC to rate bonds on Shanghai and Shenzhen exchange markets 5 . But their brethren companies with same domestic shareholders have this qualification. Further, Shanghai Stock Exchange forbade the trading of new corporate bonds on exchange markets, which are rated by CCXI and Lianhe, from August 2014, and this requirement was enforced as an act in May 20156. This is a kind of protection of local CRAs, preventing the financial market from controlling by foreign rating agencies.

Dagong was founded in March 1994, and it is also one of the earliest CRAs in China. In 2010, the SEC rejected the application of Dagong to enter into the NRSRO because SEC thought it did not have the capability to analyse a transnational corporation. Dagong does not have the issue of foreign capital. It has full licences7 to rate all kinds of bonds on different markets, and it is also recognized by CIRC from 2003. Dagong can rate borrowers all around the country and it employed more than 300 workers at the end of 2015.

Other issuer-pay CRAs in China, SBRC, CCXR, United Ratings (UR), Pengyuan Credit Rating Co., Ltd (Pengyuan), Golden Credit Rating International Co., Ltd (Golden) and Shanghai Far East Credit Rating Co., Ltd (SFE) are all national rating agencies.

2.3 The charging models of CRAs

China’s credit rating industry did not experience the change from investor-pay to issuer-pay model. Initially, China’s CRAs collected most of their revenue from the borrowers. One agency, Xinhua Far East successfully sold their reports about the quality of around 100 listed companies on China’s stock market. This business model started from early 2002, but it was

4 CIRC 2013: The recognition of seven credit rating agencies. http://www.circ.gov.cn/web/site0/tab5214/info3887803.htm 5 CSRC 2007 Decree No.50: Provisional Administrative Measures on the Credit Rating Business in Chinese Securities Marke. 6 Shanghai Stock Market 2015：The rules of the Shanghai stock exchange listed corporate bonds (revised in 2015). http://www.sse.com.cn/lawandrules/sserules/listing/bond/c/3932583.shtml 7 PBOC (1997), NDRC (2003), CIRC (2003, 2013), CSRC (2007).

10

not applied to the whole bond market. And it disappeared along with the fall of SFE. Nowadays, the only investor-pay rating agency in China is China Rating, and only the subscribers can read their reports in detail. Investors without subscription just can review the ratings and brief introduction about a particular firm. This situation is very similar to that of the U.S., but China’s CRAs did not accumulate so much reputation from the investor-pay model compared with the international big three who have around 60 years investor-pay history.

About the charging fee, there was fierce price fight among CRAs before, which lead to malignant competition (Chen 2010). From October 2007, urged by POBC, China’s five major interbank CRAs (CCXI, Lianhe, Dagong, SBRC, SFE) started to implement the Rating Charge Self-discipline for Credit Rating Agencies on Interbank Bond Market. In this agreement, five CRAs agree to abide by the Guiding Opinions of the People's Bank of China for the Management of Credit Rating (PBOC 2006) and the People’s Republic of China financial industry standard Specification for credit rating in the credit market and interbank bond market (PBOC 2007), and unify the minimum charging rate, in order to prevent the conflict of interest when rating bonds. In detail, for non-financial institution’s CP, the minimum price for rating the issuers and the bond are 100,000 Yuan and 150,000 Yuan separately. For the long-term bonds, minimum charging rate for corporate bond, convertible bond and MTN is 250,000 Yuan, and price for financial institution’s bonds should be more than 350,000 Yuan. The fee of monitor rating for long-term bonds is charge by year in the duration, and it is 20% of the initial rating fee. Moreover, minimum charging rate for ABS is 600,000 Yuan and for MBS is 1000,000 Yuan specifically. And this self-discipline agreement is also applied to other CRAs afterwards, and also becomes a normal standard for the bonds on exchange market. Further, in 2008, PBOC released the Notice of the People's Bank of China on Strengthening the Management of the Credit Rating Practices in Inter-bank Bond Market, which required CRAs to report their potential clients and business to POBC before starting the rating process. And this regulation requires the minimum time for finishing the rating report for the conglomerate is 45 days. This also prohibits the price competition and assures the report quality to some extent.

Therefore, different from the U.S., the minimum charging rate of China’s credit rating is decided by the government. And in practice, because of their fear to lose clients, CRAs in China basically charge the minimum rate rather than increasing the price. From this perspective, it is obvious that the international big three CRAs have more power on the market as they still can put up the rating price. In other word, China’s credit rating industry is in more intense competition.

2.4 The process of getting rated

The process of China’s bond rating is similar to that of the U.S. There is also a preliminary rating process. During this period, the companies who want to issue bonds will be led by the underwriters to seek appropriate CRAs. Generally, CRAs will give an initial rating based on the public information and materials supplied by the borrowers and underwriters. Issuers will compare the ratings with the consultation of underwriters. And during this process, they need to consider both the ratings of different CRAs and the reputation of them, or say the value of the ratings, and how much the ratings can save their cost. After deciding the agency, the CRA and issuer need sign a contract and report this to the CRA’s supervisors. Then there will be a team made of at least analysts, and they will go to the issuer’s office to start investigation. All the rating process and documents should follow the relevant standards

11

issued by PBOC and CSRC. After voting on the rating committee, the final result will be released to the public. If the issuers disagree with the final rating, CRAs will not release this result. The CRAs will monitor the issues on an ongoing basis.

Difference from the U.S., unsolicited rating is not prevalent in China. Another difference is double rating. Most of China’s bonds are rated by only one agency. For the major categories of bonds, there are no requirement on double rating except for asset-based securities8 and SCP9. And China’s issuers are not willing to seek more than one CRA to rate their bonds. Therefore, for SCP and asset-based securities, it is possible to discuss the split rating problem. While for other types of bonds, it needs another angle to understand same ratings given by different CRAs for different companies.

3. CCR's institutional background

The background is that China's bond market experienced an extraordinary rapid growth from 2008, and the government authorities found the ratings given by issuer-pay CRAs had misleading information and chaos. Thus the government proposed to build an independent CRA. China Credit Rating Co., Ltd (CCR) was founded in August 2010 by National Association of Financial Market Institutional Investors (NAFMII), with 50 million RMB registered capital. NAFMII is a self-regulation organization with the purpose to propel the development of China Over the Counter (OTC) financial market, under the direction of PBOC. The registered capital comes from the membership fee collected by NAFMII. Although CCR announced it is totally independent from PBOC10, PBOC is in charge of its shareholder NAFMII 11 , and both its chairman of the board and its chairman of the supervisory board once worked at PBOC12. Thus from one side, CCR has the feature of Public Utility model. From the other side, CCR accepts the investor-pay model. CCR releases its ratings through ChinaBond, Wind and its own Web site. All investors in the market can see its rating announcement and the rating notches. But only the subscribers can receive the full access to CCR's current and historical rating reports, and customised services from CCR like report for particular companies or industries13. CCR initiated its first rating at the end of 2011, and from 2012 to 2015, CCR has covered 870 companies who have already issued bonds on the market. Additionally, CCR regularly releases relevant industrial credit analysis reports and credit risk perdition models to its customers, and organizes credit risk related trainings as well. According to CCR's internal policy, its coverage is random.

As a combination of investor-pay model and public model, CCR has some unique features. Firstly, it has some connection with the government (PBOC) and declares its purpose is non for benefit, thus it is to some extent immune from pressure from the investors, which require higher ratings as mentioned in Chapter 1. Secondly, CCR applied a relatively sufficient approach to prevent excessive free-riding problem, which is the main issue for investor-pay CRAs. The investors (eg. banks, securities firms and insurance companies) subscribe directly to CCR for each rated asset class in which they conduct business. CCR allocates a unique

8 PBOC, CBRC, CSRC 2012: Notice on further expansion of the securitization of credit assets. (关于进一步扩大信贷资产证

券化试点有关事项的通知) 9 SCP must have double ratings, and one of which comes from investor-pay agency. Issuers who launch SCP must have both ratings above AA+. http://news.cnstock.com/news/sns_jr/201310/2780021.htm 10 http://www.nafmii.org.cn/ztbd/zzzxpgyxzr/201202/t20120227_4048.html 11 http://www.nafmii.org.cn/fzlm/GYWM/201202/t20120227_4278.html 12 http://www.chinaratings.com.cn/AboutUs/Governance/ 13 CCR provides a range of service packages to the investors with different levels of subscription fee.

12

subscriber number (akin to a PIN) to each investor per asset class. Each time a subscriber wants to download the full-version rating report or rating database, it much enter the related user number and validated PIN. The advantages of this approach is described in the study of Deb et al. (2011). Thirdly, as a combined model, subscription fee solves the problem of lacking incentives to provide high quality ratings and the shortage of government fund, which are the main concerns for public utility model. Last, CCR's ratings are not considered in the bond valuation system of ChinaBond, thus till now it does not have a powerful influence on the bonds price. In addition, ratings from CCR are not widely included in regulatory framework yet. Therefore, CCR has less pressure to inflate rating than the incumbents, from the investors' benefit aspect or from the perspective of regulatory purpose. It combines the advantages of investor-pay and public models, at the same time overcomes their drawbacks.

From Table 1, we can see the comparisons between ratings by CCR and issuer-pay CRAs after 2012 for different sub-groups. It compares every firm-year-CRA rating given by issuer-pay CRAs with the rating given to that firm in the same year by CCR. The ratings by CCR are like the benchmarks for all issuer-pay CRAs, as no one particular rating given by CCR is higher than that by the incumbents. And the situation that CCR's rating lower than issuer-pay CRAs' accounts for the majority. For example, in the state-owned-enterprises (SOEs) group, 77.77% observed ratings by issuer-pay CRAs are higher than CCR's ratings, but this percentage is lower than that of the non-SOEs group, which is 98%. Table 2 presents summary statistics for issuer-pay CRAs’ and CCR’s ratings between 2012 and 2015, categorized by different groups, and t-test results of mean differences between groups are reported as well. We can observe that in each group, CCR's ratings are all significantly lower than ratings by issuer-pay CRAs, at 1% level of significance.

4. Data and Methodology

We collect the data from Wind Information Co. Ltd (Wind), China Central Depository & Clearing Co., Ltd (ChinaBond) and CCR. The author worked in a securities company in China for several years. Accepted by the previous employer, the author can keep on using the Wind Info database and CCR database for research purpose.

Our sample consists of 17,521 firm-year-CRA observations from the 1,534 firms who have rating records for both before- and after-2012, and 645 firms are covered by CCR for this sample. We include all the long-term credit ratings of Chinese firms assigned by all the issuer-pay CRAs. The ratings include both initial ratings and monitoring ratings on an ongoing basis. The study period is between 2006 and the end 2015. The data sample excludes ratings for financial institutions, treasury bonds and enterprise set bond and other non-rated or small volume bond categories. Following the prior literature, a numerical value is assigned to each rating on a notch basis as follows: AAA=9, AA+=8, AA=7, AA-=6, A+=5, A=4, A-=3, BBB+=2, others=1.

To identify the effect of CCR entrant on the rating inflation of the incumbent issuer-pay CRAs, we conceive the entry of CCR as quasi-natural experiment. At first, we are interested in whether the new entrant makes the incumbents more conservative when giving ratings. Thus the default treatment group consists of firms covered by CCR. As the incumbents reacted to the CCR entrant only after CCR officially initiated a number of ratings on the market, we consider the start of 2012 as the start of the treatment.

13

Table 1 Statistics of comparison between ratings given by CCR and issuer-pay CRAs Table 1 presents comparison between ratings of issuer-pay CRAs and CCR between 2012 and 2015 The sampled firms have been rated both before and after 2012. It compares every firm-year-CRA rating given by issuer-pay CRA with the rating given to that firm in the same year by CCR. Column 1-6 present the comparison between different sub-groups. Firms in SOEs group are owned by the government, otherwise the Non-SOEs; Reputable group consists observations rated by reputable issuer-pay CRAs; otherwise Non-reputable group; Large group consists of firms whose total assets are greater than the median level, otherwise the Small group; Listed group consists of firms who listed publicly on the Shanghai or Shenzhen exchange markets, otherwise the Unlisted group; High frequency group consists firms who received ratings from the issuer-pay CRAs more than the median level, otherwise the Low frequency group; High MI group consists firms that come from the provinces whose marketization index are higher than the median level, otherwise the Low MI group.

Comparison of ratings between CCR and issuer-pay CRAs Ownership (1) Reputation (2) Size (3) Listed (4) Rating times(5) MI (6) Issuer-pay ratings

SOEs Non-SOEs

Reputable Non-reputable

Large Small Listed Unlisted High frequency

Low frequency

High MI

Low MI

>CCR 2,708 736 2,029 1,415 2,757 687 928 2,516 2,812 632 1,638 1,806 Proportion 77.77% 98.00% 78.67% 85.55% 78.04% 98.14% 81.98% 81.14% 79.32% 91.86% 71.59% 92.85% =CCR 774 15 550 239 776 13 204 585 733 56 650 139 Proportion 22.23% 2.00% 21.33% 14.45% 21.96% 1.86% 18.02% 18.86% 20.68% 8.14% 28.41% 7.15%

<CCR 0 0 0 0 0 0 0 0 0 0 0 0 Proportion Nil Nil Nil Nil Nil Nil Nil Nil Nil Nil Nil Nil

14

Table 2 Mean difference of ratings between different groups. Table 2 presents the mean differences of ratings given by CCR and issuer-pay CRAs between 2012 and 2015. The sample includes firms who have been rated both before and after 2012. Rating is a numerical value based on a notch basis as follows: AAA=9, AA+=8, AA=7, AA-=6, A+=5, A=4, A-=3, BBB+=2, others=1. Firms in SOEs group are owned by the government, otherwise the Non-SOEs; Large group consists of firms whose total assets are greater than the median level, otherwise the Small group; Listed group consists of firms who listed publicly on the Shanghai or Shenzhen exchange markets, otherwise the Unlisted group; High frequency group consists firms who received ratings from the issuer-pay CRAs more than the median level, otherwise the Low frequency group; High MI group consists firms that come from the provinces whose marketization index are higher than the median level, otherwise the Low MI group. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Mean Rating Issuer-pay CRAs CCR (Issuer-pay - CCR) Ownership SOEs 7.718 6.548 1.170***

Non-SOEs 6.755 4.596 2.159*** (SOEs - Non-SOEs) 0.963*** 1.952***

Size Large 7.841 6.654 1.187*** Small 6.627 4.611 2.016*** (Large - Small) 1.214*** 2.043***

Listed Listed 7.426 6.225 1.201*** Non-Listed 7.569 6.175 1.394*** (Listed - Non-Listed) -0.143 0.050***

Rating times High frequency 7.858 6.557 1.301*** Low frequency 6.984 5.254 1.730*** (High frequency - Low frequency) 0.874*** 1.303***

Marketization index

High MI 7.724 6.602 1.122*** Low MI 7.333 5.701 1.632*** (High MI - Low MI) 0.391*** 0.901***

15

We observe that there are significant differences between the treatment and control groups on the main financial characteristics (see Table 3). So there is a possibility that the larger firms are more exposed to the market and it should be more convenient to get access to their information. Thus the CCR coverage is inclined to choose those firms although its own policy is random choice. As a consequence, the assignment into treatment and control groups is not totally random, as a true experiment would require. However, for a quasi-experiment it is sufficient that the CCR entrant (a reform dominated by government) can be thought of as if they were randomly assigned into treatment and control groups.

Fig. 2 plots the trends for mean ratings assigned by nine issuer-pay CRAs from 2006 to 2015. The left describes the mean ratings for firms from control group while the right indicates that from treatment group. We can see both groups experienced rating increasing, to some extent representing the rating inflation phenomenon. For control group, the rating change range is 1.28. However, the rating change range for treatment group is 1.17, which is 0.11 notches lower than the control group. In other words, the rating inflation is less severe for firms who have been covered by CCR. But we need more sophisticated methods to reach more pervasive results.

Fig. 2 Mean rating trends of treatment and control groups.

Therefore, we apply the difference-in-difference (DID) approach to estimate the influence of CCR entrant on rating inflation. The graphical analyses of Fig.2 are formalized in a multivariate regression setting. An industry-fixed-effect OLS regression model following Becker and Milbourn (2011), Jiang, Harris and Xie (2012), Xia (2013) and Baghai, Servaes and Tamayo (2013):

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖,𝑡𝑡 = 𝑐𝑐 + 𝛽𝛽1𝐶𝐶𝐶𝐶𝑅𝑅𝑐𝑐𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 ∗ 𝑃𝑃𝐶𝐶𝑃𝑃𝑅𝑅𝑖𝑖,𝑡𝑡 + 𝛽𝛽2𝑃𝑃𝐶𝐶𝑃𝑃𝑅𝑅𝑖𝑖,𝑡𝑡 + 𝛽𝛽3𝐶𝐶𝐶𝐶𝑅𝑅𝑐𝑐𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 + 𝛽𝛽4𝑋𝑋𝑖𝑖,𝑡𝑡 + 𝜀𝜀𝑖𝑖,𝑡𝑡 (1)

In Eq. (1), Rating denotes the numerical values of ratings from incumbent issuer-pay CRAs; CCRcover is a dummy variable that equals one if the rated firm is also covered by CCR, otherwise zero. Post is also a dummy variable which equals one if the firm-year-CRA observation is from the period after 2012 when CCR officially started its business, zero otherwise; and X is a set of firm characteristic controls. X includes Sales which is the natural

66.

57

7.5

8

2006 2009 2012 2015 2006 2009 2012 2015

control group treatment group

mea

n ra

ting

16

logarithm of sales; Debt rate which is the ratio of total liability from the balance sheet to total assets; ROA which is the return on assets that represents the profitability; Tangibility that is the ratio of net property, plant and equipment to total assets; and Growth which is the year to year increase of operating income; and industry and year that correspond to industry and year dummies. In this equation, if the CCR entrant can alleviate the problem of rating inflation, then we would expect to see a negative coefficient on the interaction term, β1.

Endogeneity concerns

Although CCR announce its coverage is randomly selected, we still observe some features for its covered firms from Table 3. In fact, on average, covered firms are significantly different from uncovered firms. This may cause endogeneity problems in econometric analyses investigations the role of CCR entrant lead to lower rating inflation in the form of omitted variable bias due to self-selection. If firms’ financial characteristics along are sufficient to cause incumbents issuer-pay CRAs to adjust their ratings regardless of CCR’s coverage, the effect of CCR will be overestimated.

To correct this problem, firstly, we apply the propensity score matching (PSM) approach (Xia 2013). We matched CCR-rated firms in the sample (the treated group) to those not rated by CCR as of the end of 2015 (the control group) based on various dimensions that are likely to predict CCR's coverage decision. The idea is that, by putting together firms (firm-year-CRA) that are similar in these dimensions, we obtain matched firm-year-CRA that designates when CCR would have begun to rate the firm had it decided to cover the firm. We match the sample firms based on a set of pre-treated (i.e., one-year prior to CCR coverage) characteristics, and we use one-to-one matching method. These matched firms are the most similar ones to the covered sample. From the control group, we find 625 matching firms for the treatment group. Through the regression results using PSM sample, we can observe a much clearer effect of CCR entrant, supposed the treatment and control groups are very similar to each other in terms of those chosen dimensions.

Secondly, we address the well-recognized issue of endogenous matching of Heckman (1980) two-stage approach in the manner of Ross (2010) and Schenone (2004). We estimate in the first stage of Heckman approach selection equations for firms’ characteristics lead CCR’s coverage. We then construct from this regression inverse Mills ratio that is added as control variables in the second-stage regression. Suggested by Heckman (1980), this procedure solves the omitted variable (or self-selection) bias caused by endogenous matching.

We generally follow our observations from Table 3 to decide the independent variables for the CCR selection equations. We thus use in the first-stage regression a number of variables that differ significantly for CCR covered firms and uncovered firms. Furthermore, as suggested by (Prabhala & Li 2007), the first-stage regressions include variables that are not in the second-stage equations. We use ownership and listed as my main instruments for CCR selection, as the SOEs and listed firms generally have more transparent information to access for CCR. The selection equation is estimated using probit regressions with issuer-clustered standard errors and contains year and industry controls. The regression results are reported in Table 4. Next, we apply our difference-in-difference analysis using Eq. (2) which adds the inverse Mills ratio to Eq. (1) for both the raw sample and PSM matched sample, in order to seed after controlling the sample selection bias, if the entry of CCR can still alleviated the rating inflation.

17

𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖,𝑡𝑡 = 𝑐𝑐 + 𝛽𝛽1𝐶𝐶𝐶𝐶𝑅𝑅𝑐𝑐𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 ∗ 𝑃𝑃𝐶𝐶𝑃𝑃𝑅𝑅𝑖𝑖,𝑡𝑡 + 𝛽𝛽2𝑃𝑃𝐶𝐶𝑃𝑃𝑅𝑅𝑖𝑖,𝑡𝑡 + 𝛽𝛽3𝐶𝐶𝐶𝐶𝑅𝑅𝑐𝑐𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖,𝑡𝑡 + 𝛽𝛽4𝑋𝑋𝑖𝑖,𝑡𝑡 +𝛽𝛽4𝐼𝐼𝑅𝑅𝐶𝐶𝐶𝐶𝐶𝐶𝑃𝑃𝐶𝐶 𝑀𝑀𝑅𝑅𝑀𝑀𝑀𝑀𝑃𝑃 𝐶𝐶𝑅𝑅𝑅𝑅𝑅𝑅𝐶𝐶𝑖𝑖,𝑡𝑡 + 𝜀𝜀𝑖𝑖,𝑡𝑡 (2)

5. Empirical results

5.1 Descriptive statistics

Table 3 presents the characteristics of all firms who received ratings by the incumbent nine issuer-pay CRAs between 2006 and 2015. Among the total 17,521 ratings assigned to these firms by the issuer-pay CRAs, the mean rating is 7.436, which means on average, the long-term rating is approximately AA.

Next, we compare the characteristics of treatment and control groups between 2012 and 2015. Panel B of Table 3 presents summary statistics for control group and treatment group respectively. We can observe that the mean rating and median rating of treatment group are all significantly higher than the control group. And their financial characters are significantly different from each other. For example, mean rating perceived by firms who are covered by CCR is significantly 0.893 notches higher than that of firms in the control group; and he debt rate of firms in the treatment group is significantly 7.4% higher than the control group. According to CCR’s internal policy, it chooses the covered firm on a random basis14. But from Panel B of Table 3, on average, firms from the covered group (treatment group) have larger size, higher leverage level, higher proportion of tangible assets and lower growth rate.

14 Interview from the insiders of CCR.

18

Table 3 Rating sample summary statistics. This table presents descriptive statistics for the rating sample from 2006 to 2015. The firms in the sample have been rated both before and after the official entry of CCR. Panel A presents the statistics of full sample. Rating is a numerical value based on a notch basis as follows: AAA=9, AA+=8, AA=7, AA-=6, A+=5, A=4, A-=3, BBB+=2, others=1. Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise. Total assets and Sales are in 100 Millions of RMB yuan; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Panel B presents the statistics of controlled group and treatment group and the mean differences between these two groups. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Summary statistics of full sample Rating Post CCrcover Total assets Sales Debt

rate ROA Tangibility Growth

Whole sample

N 17,521 17,521 17,521 17,476 17,457 17,470 17,466 17,460 17,406 Mean 7.436 0.652 0.571 999.321 512.253 0.601 5.213 0.259 0.653 Std.dev. 1.213 0.476 0.495 3,660.22 1,970.14 0.151 12.141 0.202 12.365 Min 1 0 0 2.511 0.01 0.003 -65.783 0 -0.989 Max 9 1 1 56,099.30 28,803.11 1.622 861.116 0.969 645.822

Panle B: Summary statistics of controlled group and treatment group Firms not covered by CCR

N 7,522 7,493 7,473 7,487 7,483 7,479 7,436 Mean 6.926 329.192 121.603 0.559 4.459 0.203 1.221 Std.dev. 1.142 580.51 309.702 0.158 4.942 0.181 18.889 Min 1 2.511 0.01 0.003 -65.783 0 -0.989 Max 9 9,346.49 4,614.14 1.622 63 0.961 645.822

Firm covered by CCR

N 9,999 9,983 9,984 9,983 9,983 9,981 9,970 Mean 7.819 1,502.30 804.654 0.633 5.778 0.301 0.229 Std.dev. 1.121 4,755.08 2,552.54 0.137 15.455 0.207 0.662 Min 1 9.306 0.1 0.068 -40.72 0 -0.833 Max 9 56,099.30 28,803.11 1.398 861.116 0.969 23.685

Mean Difference 0.893*** 1173.108*** 683.051*** 0.074*** 1.319*** 0.098*** -0.992*** Median Difference 1*** 211.589*** 124.220*** 0.082*** 1.212*** 0.107*** 0.008***

5.2 Baseline empirical results

Table 4 presents the regression results for Eq. (1) and Eq. (2). In Table 4, we firstly test the raw sample, which includes 17,389 firm-year-CRA observations. Column 1 and 2 presents the regression results with and without industry and year effects respectively. The negative correlation between incumbent issuer-pay CRAs’ ratings and the interaction term suggests that overall, ratings for treated firms are relatively lower than controlled firms after the entry of CCR. In other words, CCR’s entrant has a statistically significant influence on alleviating the rating inflation problem of the issuer-pay CRAs in China. Other control variables show the expected signs. For example, in the post-period, ratings are higher which is already shown on Fig. 2, and this is also in line with the previous study on rating inflation in China’s bond market (Kennedy 2003, 2008; Lee 2006; Song 2013). But after all, those incumbent CRAs are more conservative. Moreover, the covered firms overall have higher ratings than non-covered firms, which is consistent with the data description in Table 3. We also find larger sales, more tangibility and higher growth rate are all associated with a more favourable rating, while a higher leverage ratio leads to a lower rating.

19

Table 4 Difference-in-Difference analysis of ratings

This table presents the difference-in-difference analysing results for raw sample and PSM matched sample between 2006 and 2015 respectively. The raw sample consists of firm-year-CRA 17,389 observations of 1,534 firms between 2006 and 2015. The Propensity Score Matching (PSM) sample consists of firm-year-CRA 12,263 observations of 883 firms and from which 238 firms are one-to-one matched firms from the control group. Column 1, 2, 5 and 6 represent the regression results of Eq. (1). The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Column 3, 4, 7 and 8 present the DID result using Heckman test to correct the sample selection bias. Column 3 and 7 present results of probit regression of CCR coverage choice on firm-specific characteristics for the sample of ratings given by issuer-pay CRAs from 2006 to 2015. The left-hand-side variable is CCRcover. Ownership equals one if the firm is owned by the government, otherwise zero; Listed equals one if the firm is listed on Shanghai or Shenzhen exchange market, otherwise zero. Column 4 and 8 present the regression results of Eq. (2). Standard errors are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Raw sample PSM matched sample

(1) (2) (3) First stage (4) (5) (6) (7) First stage (8) CCRcover*Post -0.626**

(0.031) -0.149*** (0.026)

-0.150*** (0.026)

-0.117*** (0.044)

-0.187*** (0.029)

-0.188*** (0.027)

CCRcover 0.402*** (0.026)

0.484*** (0.024)

0.470*** (0.024)

0.344*** (0.035)

0.396*** (0.028)

0.334*** (0.026)

Post 0.303*** (0.024)

0.361*** (0.040)

Debt rate -1.596*** (0.547)

-1.646*** (0.054)

0.481* (0.264)

-1.772*** (0.057)

-1.505*** (0.066)

-1.612*** (0.065)

-0.323 (0.346)

-0.961*** (0.064)

ROA -0.0003 (0.001)

0.000 (0.001)

0.025*** (0.008)

-0.001 (0.001)

-0.000 (0.001)

-0.000 (0.001)

0.009 (0.009)

-0.002*** (0.001)

Tangibility 0.182*** (0.038)

0.372*** (0.038)

0.924*** (0.195)

0.159*** (0.049)

0.473*** (0.042)

0.604*** (0.041)

0.286 (0.237)

-0.127*** (0.044)

Sales 0.371*** (0.005)

0.364*** (0.005)

0.339*** (0.029)

0.285*** (0.013)

0.404*** (0.006)

0.394*** (0.005)

0.077** (0.031)

0.193*** (0.008)

Growth 0.001** (0.001)

0.001* (0.001)

-0.334*** (0.012)

0.015*** (0.002)

0.014* (0.007)

0.014** (0.007)

-0.027** (0.013)

0.098*** (0.007)

Ownership 0.164* (0.096)

0.233* (0.122)

Listed -0.185** (0.089)

-0.161 (0.107)

Constant 6.302*** (0.035)

-205.513*** (8.849)

46.655** (18.589)

-19.804*** (9.089)

6.040*** (0.053)

-226.510*** (11.312)

-38.495* (20.793)

-157.649*** (10.901)

Inverse Mill ratio -0.418*** (0.062)

-6.141*** (0.167)

Industry fixed effects No Yes Yes Yes No Yes Yes Yes Year dummies No Yes Yes Yes No Yes Yes Yes Observations 17,389 17,389 17,389 17,389 12,263 12,263 12,263 12,263 Prob (Chi-squared) 0.000 0.000 (Pseudo) R-squared 0.368 0.396 0.199 0.397 0.342 0.369 0.020 0.431

20

Column 3 of Table 4 is the probit regression result of Heckman first stage sample selection bias method. We can observe that whether CCR covers a particular firm or not is highly related to the financial characteristics of that firm. And Column 4 reports the second stage regression result of Heckman test with industry and year effect, also the result of Eq. (2). We can still observe a significantly negative value of 𝛽𝛽1, which means after correcting the sample selection bias, the entry of CCR can alleviate the rating inflation as well. For the PSM matched sample, we can receive the same results from Column 5 to 8.

This implies that the rating given by CCR does have some benchmark effect for other issuer-pay incumbent CRAs, elevating some mechanism behind and leading the incumbents to give more fair ratings. Furthermore, ratings of CCR add private information to the market (represented by the Inverse Mills ratio), to some extent alleviate the information asymmetric problem.

5.3 The information environment and information asymmetry

Till now, we have suggested that CCR, a combined CRA of public and investor-pay model, can provide private information to the market and alleviate the rating inflation of issuer-pay CRAs. We want to find the mechanism behind that CCR entrant can reduce information asymmetry.

It is acknowledged that the firms with better information environment are more exposed to the public. Therefore, if CCR release more information on the market, the ratings on firms with better information environment should change more. We apply three variables to represent the information environment. The first one is the marketization index (MI), developed by Fan, Wang and Zhu (2011). As we know, China has more than 30 provinces, and the economic development level among different provinces is seriously unbalanced. Some highly-developed provinces such as Guangdong, Zhejiang, Beijing and Shanghai, have more mature legal system, more sophisticated investors and more friendly investment environment. But other provinces like Gansu, Xinjiang, have relatively incomplete market and legal environment. So we separate the sample to two sub-groups based on the MI. The high-MI group consists of firm who come from provinces with MI above the median level; otherwise the company belongs to the low-MI group.

From Table 5, we can find significant evidence for high-MI group but no clear evidence for low-MI group, that after CCR entrant the incumbent issuer-pay CRAs decrease their rating inflation for the treatment-group firms. And the differences between these two groups are significant in terms of the coefficient of the interaction term for all three models. This result indicates that in the area with advanced legal system and better investor protection environment, the issuer-pay CRAs are more conservative. The reason behind may be that the issuers in those provinces are more sophisticated than other provinces, plus the more sophisticated investors are also concentrated in those high-MI areas. Thus the CRAs need to care more about the loss of reputation capital and the regulatory cost for keeping on inflating the ratings.

Secondly, we use the investor protection index (IP) in 2009 (Fan, Wang & Zhu 2011). It represents the development level of the intermediary institutions and the maturity of law environment. We divided the sample into two groups. The high-IP group consists of firms

21

come from the provinces with higher IP than the median level; the low-law group contains those come from the lower than median provinces. The result does not change (see Table 6).

Next feature we considered is the CRAs following, which represents the rating times each firm received from 2006 to 2015. We divided the sample to high-frequency and low-frequency groups. The high-rated frequency group consists of those whose number of ratings between 2006 and 2015 are above the median rating times; otherwise the company belongs to the low-rated frequency group. The median rating times is fourteen. The frequency of ratings indicates how often the firms are covered and monitored by the issuer-pay CRAs. It is assumed that the more the times CRAs rate, the closer the relationship they have with the issuers, that means issuers have more repeated request for ratings. And as we mentioned before, in the long-run game among rating agencies, issuers and investors, CRAs need to consider more about their reputation to secure their long-term revenue. Therefore, we have the hypothesis that for high-frequency group, issuer-pay CRAs decrease their rating inflation for CCR covered companies compared with uncovered firms, while for low-frequency group, there is not such relationship. From our empirical analysis (see Table 7), we observe statistically significantly negative relationship between ratings and interaction term for high-frequency group while no significant results for low-frequency group, which proves our hypothesis. In addition, three of the four models show there is significant differences between the coefficients of the two groups at 5% level.

In conclusion, firms with better information environment experience a greater reduction of information asymmetry after the entry of CCR. This also emphasizes the importance of the information environment construction in emerging markets.

22

Table 5 Difference-in-Difference analysis of rating: categorized by Marketization Index.

This table presents the difference-in-difference analysing results categorized by marketization index. The high MI group consists of firm come from the provinces with higher-than-median MI. The raw sample and Heckman test sample consist of firm-year-CRA 17,389 observations of 1,534 firms between 2006 and 2015. The Propensity Score Matching (PSM) sample consists of firm-year-CRA 12,263 observations of 883 firms and from which 238 firms are one-to-one matched firms from the control group. The firms in these three samples all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Robust standard errors clustered at the firm level are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. Raw Sample (1) PSM matched Sample (2) Heckman test Sample (3) High MI Low MI High MI Low MI High MI Low MI CCRcover*Post -0.219***

(0.055) -0.056 (0.050)

-0.237*** (0.051)

-0.122** (0.048)

-0.225*** (0.054)

-0.050 (0.050)

Diff of coefficient

-0.163**

(chi2=4.89)

-0.115*

(chi2=2.72)

-0.175**

(chi2=5.68)

CCRcover 0.516*** (0.101)

0.448*** (0.076)

0.341*** (0.114)

0.430*** (0.089)

0.474*** (0.104)

0.437*** (0.076)

Post Debt rate -1.408***

(0.268) -1.648*** (0.250)

-1.046*** (0.294)

-2.074*** (0.336)

-1.704*** (0.277)

-1.798*** (0.309)

ROA 0.000 (0.001)

0.001 (0.011)

-0.000 (0.001)

-0.002 (0.013)

-0.001 (0.002)

-0.004 (0.011)

Tangibility 0.527*** (0.192)

0.371** (0.157)

0.767*** (0.198)

0.699*** (0.194)

0.064 (0.239)

0.152 (0.290)

Sales 0.361*** (0.027)

0.317*** (0.025)

0.353*** (0.031)

0.402*** (0.031)

0.181*** (0.058)

0.233*** (0.089)

Growth 0.001*** (0.000)

0.001** (0.001)

0.112*** (0.032)

0.002 (0.009)

0.034*** (0.007)

0.015 0.013)

Constant -193.168*** (24.361)

-214.152*** (21.804)

-212.173*** (26.913)

-270.374*** (27.417)

-226.094*** (25.486)

-251.977*** (26.653)

Inverse Mill ratio -1.016*** (0.335)

-0.412 (0.417)

Industry fixed effects Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Observations 8,595 8,794 6,333 5,930 8,595 8,794 R-squared (adjusted) 0.431 0.331 0.381 0.336 0.441 0.332

23

Table 6 Difference-in-Difference analysis of rating: categorized by Investor Protection Index.

This table presents the difference-in-difference analysing results categorized by if the law environment index of the province of the rated firms is above the median level. The high-IP group consists of firm who come from provinces with IP above the median IP; otherwise the company belongs to the low-IP group. The raw sample and Heckman test sample consist of firm-year-CRA 17,389 observations of 1,534 firms between 2006 and 2015. The Propensity Score Matching (PSM) sample consists of firm-year-CRA 12,263 observations of 883 firms and from which 238 firms are one-to-one matched firms from the control group. The firms in these three samples all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Robust standard errors clustered at the firm level are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively. Raw Sample (1) PSM matched Sample (2) Heckman test Sample (3) High IP Low IP High IP Low IP High IP Low IP CCRcover*Post -0.213***

(0.054) -0.060 (0.050)

-0.240*** (0.050)

-0.118** (0.048)

-0.218*** (0.054)

-0.053 (0.051)

Diff of coefficient

-0.153** (chi2=4.27) -0.122* (chi2=3.10) -0.165** (chi2=4.98)

CCRcover 0.495*** (0.100)

0.461*** (0.077)

0.343*** (0.114)

0.402*** (0.092)

0.454*** (0.103)

0.448*** (0.077)

Post Debt rate -1.412***

(0.264) -1.605*** (0.255)

-1.002*** (0.296)

-2.131*** (0.334)

-1.692*** (0.277)

-1.765*** (0.309)

ROA 0.000 (0.001)

0.004 (0.012)

-0.000 (0.001)

-0.006 (0.013)

-0.001 (0.002)

-0.002 (0.011)

Tangibility 0.548** (0.191)

0.548*** (0.191)

0.787*** (0.200)

0.643*** (0.190)

0.116 (0.240)

0.096 (0.295)

Sales 0.363*** (0.027)

0.363*** (0.027)

0.352*** (0.031)

0.404*** (0.031)

0.195*** (0.059)

0.219** (0.091)

Growth 0.001*** (0.000)

0.001*** (0.000)

0.115*** (0.045)

0.007 (0.012)

0.032*** (0.011)

0.016 (0.014)

Constant -195.351*** (24.137)

-195.351*** (24.137)

-206.930*** (26.891)

-264.026*** (27.283)

-226.473*** (25.277)

-254.628*** (27.136)

Inverse Mill ratio -0.955*** (0.338)

-0.453 (0.422)

Industry fixed effects Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Observations 8,762 8,627 6,062 6,201 8,762 8,627 R-squared (adjusted) 0.431 0.328 0.381 0.331 0.440 0.329

24

Table 7 Difference-in-Difference analysis of rating: categorized by rated times

This table presents the difference-in-difference analysing results categorized by the frequency of rated times. The high-rated frequency group consists of those whose number of ratings between 2006 and 2015 are above the median rating times; otherwise the company belongs to the low-rated frequency group. The raw sample and Heckman test sample consist of firm-year-CRA 17,389 observations of 1,534 firms between 2006 and 2015. The Propensity Score Matching (PSM) sample consists of firm-year-CRA 12,263 observations of 883 firms and from which 238 firms are one-to-one matched firms from the control group. The firms in thesethree samples all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Robust standard errors clustered at the firm level are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Raw Sample (1) PSM matched Sample (2) Heckman test Sample (3) High frequency Low frequency High frequency Low frequency High frequency Low frequency CCRcover*Post -0.234***

(0.043) -0.067 (0.060)

-0.246*** (0.043)

-0.077 (0.058)

-0.234*** (0.044)

-0.068 (0.061)

Diff of coefficient

-0.167**

(chi2=5.08)

-0.169**

(chi2=5.47)

-0.166**

(chi2=4.96)

CCRcover 0.269** (0.110)

0.344*** (0.082)

0.185 (0.131)

0.270*** (0.091)

0.330*** (0.083)

0.257** (0.114)

Post Debt rate -1.028***

(0.375) -2.055*** (0.197)

-1.104*** (0.395)

-2.629*** (0.272)

-2.446*** (0.240)

-1.138** (0.455)

ROA 0.001 (0.001)

0.002 (0.009)

0.001 (0.001)

-0.014** (0.007)

-0.013 (0.008)

0.000 (0.001)

Tangibility 0.497** (0.197)

0.206 (0.145)

0.510** (0.200)

0.720*** (0.181)

-0.410 (0.258)

0.333 (0.311)

Sales 0.330*** (0.028)

0.342*** (0.022)

0.338*** (0.030)

0.444*** (0.029)

0.117 (0.075)

0.272*** (0.077)

Growth 0.000* (0.000)

0.001* (0.001)

0.072** (0.030)

-0.002 (0.009)

0.037*** (0.011)

0.008 (0.021)

Constant -251.707*** (24.734)

-206.144*** (21.005)

-262.943*** (27.210)

-214.553*** (25.891)

-230.050*** (22.561)

-262.848*** (28.170)

Inverse Mills ratio -1.091*** (0.347)

-0.353 (0.519)

Industry fixed effects Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Observations 8,087 9,302 7,455 4,808 8,084 9,298 R-squared (adjusted) 0.359 0.277 0.341 0.334 0.360 0.284

25

5.4 The reputation and information asymmetry

The second mechanism we considered is the reputation concern.

The alleviation of information asymmetry might reflect a reputation mechanism at work. That is, CCR's independent and more conservative ratings have elevated reputational concerns of those more reputable issuer-pay CRAs, which in turn strengthen their incentives to lower the rating inflation (the "incentive/reputation" channel). Specifically, as CCR covers a firm, its more lower ratings begin to reveal the lower quality of the ratings from issuer-pay CRAs as their misaligned incentives which is discussed in Section 1. This, in turn, can eventually either lower investors' confidence in these incumbents, making them less valuable to issuers, or lead to closer scrutiny and intervention from regulators. This reputation cost provides a mechanism through which CCR entrant can discipline issuer-pay CRAs into providing lower rating inflation (Bar-Isaac & Shapiro 2013; Bolton, Freixas & Shapiro 2012; Mathis, Mcandrews & Rochet 2009; Opp, Opp & Harris 2013; Skreta & Veldkamp 2009).

But in China, the reputation of issuer-pay CRAs is criticized (Kennedy 2008; Lee 2006). In specific, Bottelier (2003) finds that most listed corporate bonds in China have AAA ratings. According to the qualitative research of Xu and Han (2013), the credit rating industry in China is thought to be immature, the reputation of China’s CRAs is low, rating results lack fairness, and that severe competition dominates the market. But other people stand by that since their beginnings, China’s CRAs have earned money from issues rather than the investors, while their American counterparts collected revenue from investors for about 70 years before they changed to an issuer-pays model, during which period they built their reputation. Thus China’s CRAs need a long time to be perfect (Poon & Chan 2008). Zhang and Zhang (2010) and Zhang (2013) all build models to prove the effect of the reputation mechanism under certain contexts in China, and offered policy suggestions such as decreasing the competition on the basis of their findings. We believe there are quality differences among issuer-pay CRAs in China, some of whom have higher credibility while others might be notorious. Therefore, we aim to estimate if the influence of CCR entrant stems from the differences of reputational concerns between reputable issuer-pay CRAs and the non-reputable ones.

In order to test the above assumption, we again divided all CRAs into two groups, the reputable one and the non-reputable one. The reputable group consists of CCXI and Lianhe while the non-reputable group consists of others. There are three criteria to define if a CRA belongs to the reputable group. First, according to previous studies on the reputation of other financial intermediates, such as underwriters and banks, market share is one widely used criterion (Andres, Betzer & Limbach 2014; Megginson & Weiss 1991; Schenone 2004). Appendix 1 shows the market share of each issuer-pay CRA, based on the issue volume and number of bonds they rated between 2006 and 2015. Apparently, in total, CCXI and Lianhe account for over 70% of the market. They are the top two issuer-pay CRAs in China in terms of market share.

Second, from the previous studies, the international CRAs own higher reputation than the national CRAs, thus the second criterion is whether the issuer-pay CRA has foreign ownership from the well-known big three (Moody's, S&P and Fitch). The big three are all long established global rating agencies (GRA). GRAs are generally at a great advantages in their competition with national rating agencies (NRA). GRAs enjoy their market power

26

mainly due to their reputation capital (Hill 2004; Partnoy 1999). As the NRAs lack a well-established reputation, investors are more likely to trust the ratings issued by GRAs that have a reputation at stake. Additionally, GRAs are thought to hold a higher level of independence. Empirical studies comparing GRAs and NRAs mainly focus on Japanese market which has the second largest bond market and the most relatively well-known NRAs. Some of these studies provide evidence in support of superiority of GRAs. For example, Beattie and Searle (1992) present that CRAs are inclined to be more lenient when judging issuers from their own countries, raising doubts about their integrity and credibility. Cantor and Packer (1997) suggest that Moody's and S&P assign lower ratings than smaller CRAs, even after modifying their rating symbols to the same meaning. Packer (2002) from a sample of Japanese nonfinancial companies, also find the evidence that NRAs' ratings are systematically 3.5 notches higher on average, with respect to ratings by GRAs. Shin and Moore (2003) show inflated natures of NRAs in Japan as well. Furthermore, Nickell, Perraudin and Varotto (2000) suggest that NRAs are slower in downgrading compared with GRAs. Additionally, Carow (1999) and Li, Shin and Moore (2006) advocate the superiority of GRAs by finding that GRAs have more influence on Japanese capital market than NRAs. As we introduced in Section 2, there are two joint ventures CRAs in China, CCXI (Moody's own its 49% share) and Lianhe (Fitch own its 49% share). Moody's and Fitch not only provide advanced rating techniques to their affiliates, but also assign representatives to participate the rating decision process of CCXI and Lianhe respectively. Therefore, according the second criteria, CCXI and Lianhe should be more independent with higher reputation capital and concern more about their long-term reputation.

Third, CRAs in China need to acquire recognitions from different supervisors, among which the qualification from PBOC and CIRC are the most difficult. Because POBC is in charge of the inter-bank market which is the largest trading market in China, the recognition from POBC means the bonds issued by these CRAs can be traded by almost all participants on the market. In addition, CIRC is the regulatory organization for insurance companies, which have the most conservative investment strategy among all kinds of financial institutions. Therefore, the recognitions from POBC and CIRC are essential to CRAs. Both CCXI and Lianhe have this qualification, satisfying the third criterion.

Combine the above three criteria, only CCXI and Lianhe are recognized as the reputable issuer-pay CRAs in China. Table 8 shows the comparison between reputable group and non-reputable group, using raw sample, PSM matched sample and Heckman test sample respectively. Then from Table 8, for reputable group, all three models show a significantly negative coefficient for the interaction term at 1% level while this is not found for non-reputable group. It means the CCR entrant disciplines the rating inflation behaviour of reputable incumbents, however it does not have significant influence on non-reputable issuer-pay CRAs. This is consistent with the above literatures in that reputation mechanism can be elevated through competition from a more credible CRA, especially for those who care reputation more. Moreover, the difference between coefficients of these two groups are statistically significant. This is also in line with the literatures aforementioned that GRAs concern more about their rating quality and long-term reputation than NRAs. In other words, when CCR initiate rating for an issuer which is also covered by a reputable issuer-pay CRA, after controlling the issuer's financial characteristics, this CRA will give lower ratings to this issuer with respect to the non-CCR covered firms, in the post period. And this behaviour is significantly different from that of non-reputable CRAs, who are not sensitive to CCR entrant.

27

Table 815 Difference-in-Difference analysis of rating: categorized by the reputation of issuer-pay Credit Rating Agencies

This table presents the difference-in-difference analysing results categorized by reputable and non-reputable Credit Rating Agencies. The reputable group consists of two CRAs, CCXI and Lianhe. The non-reputable group consists of other seven issuer-pay CRAs. The raw sample and Heckman test sample consist of firm-year-CRA 17,389 observations of 1,534 firms between 2006 and 2015. The Propensity Score Matching (PSM) matched sample consists of firm-year-CRA 12,263 observations of 883 firms and from which 238 firms are one-to-one matched firms from the control group. The firms in these three samples all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Robust standard errors clustered at the firm level are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

15 From this table on, we will focus on the model with industry and year fixed effects.

Raw Sample (1) PSM matched Sample (2) Heckman test Sample (3) Reputable Non-reputable Reputable Non-reputable Reputable Non-reputable CCRcover*Post -0.244***

(0.048) -0.023 (0.058)

-0.246*** (0.042)

-0.100 (0.062)

-0.246*** (0.048)

-0.018 (0.059)

Diff of coefficient

-0.221***

(chi2=9.07)

-0.146*

(chi2=3.77)

-0.228***

(chi2=9.67)

CCRcover 0.533*** (0.082)

0.391*** (0.094)

0.404*** (0.094)

0.341*** (0.112)

0.519*** (0.084)

0.372*** (0.095)

Post Debt rate -1.913***

(0.295) -1.282*** (0.222)

-1.745*** (0.345)

-1.586*** (0.325)

-2.054*** (0.338)

-1.549*** (0.267)

ROA 0.001 (0.001)

-0.002 (0.011)

0.001 (0.001)

-0.022** (0.009)

-0.000 (0.001)

-0.011 (0.010)

Tangibility 0.378** (0.158)

0.382** (0.166)

0.618*** (0.173)

0.637*** (0.196)

0.147 (0.234)

0.025 (0.239)

Sales 0.391*** (0.024)

0.328*** (0.023)

0.403*** (0.027)

0.386*** (0.030)

0.303*** (0.061)

0.198*** (0.060)

Growth 0.002** (0.001)

0.001** (0.000)

0.099** (0.042)

0.001 (0.010)

0.018 (0.012)

0.028** (0.006)

Constant -241.978*** (21.571)

-164.349*** (22.884)

-251.693*** (22.919)

-187.853*** (33.146)

-258.873*** (22.931)

-178.682*** (24.396)

Inverse Mill ratio -0.490 (0.354)

-0.684** (0.315)

Industry fixed effects Yes Yes Yes Yes Yes Yes Year dummies Yes Yes Yes Yes Yes Yes Observations 9,121 8,268 7,264 4,999 9,121 8,268 R-squared (adjusted) 0.422 0.347 0.389 0.337 0.424 0.350

28

6. Robustness check

It may be possible that we have a higher proportion of observations with AAA rating in the treatment group, thus there is less space to inflate the ratings anymore. To control for a possible sample bias of this sort in examining the effect of CCR entrant on rating inflation, we delete the AAA observations from the sample. The regression results illustrate that the results still hold (see Appendix 2). Furthermore, it is also possible that AAA rating accounts more proportion in reputable group than non-reputable group, therefore, we also delete AAA observations when compare the reputable and non-reputable issuer-pay CRAs (see Appendix 3). We find our results are not qualitatively different.

Moreover, as discussed before, the reputable issuer-pay CRAs pay more attention on their credibility, thus they should be more cautious than non-reputable CRAs when deal with the ratings for firms which hold the high-frequency ratings. Thus we assume for high-frequency rated firms, reputable issuer-pay CRAs should lower their rating inflation, and this reaction should be more significant than non-reputable CRAs. The empirical result confirms my guess (see Appendix 4).

Another issuer is about the time period. That is, we also consider that the influence of CCR entrant may have some time lag. So we delete the observations in 2012, to assume the market needs one year to react to CCR coverage. And to keep the pre- and post-period balance, we also delete observations between 2006 and 2008. The result does not change qualitatively (see Appendix 5).

29

7. Conclusion

In this paper, we study how the entry of CCR, a new independent credit rating agency utilizing a combination of public utility model and investor-pay model, affects the behaviour of incumbent issuer-pay CRAs in China. As far as we know, this combined business model is unique in China. We find a significant decline of the information asymmetry for those firms covered by CCR, with respect to those not covered by CCR. That means, CCR acts a certification role to discipline the incumbents and creates a benchmark for them. In particular, this discipline effect is more significant on the firms with better information environment and the more reputable issuer-pay CRAs, for which the reputation capital is more important. Thus we further find the importance of information environment construction and reputation of the issuer-pay CRAs. .

The findings in this paper complement the existing literature that documents a negative link between the entry of a new issuer-pay CRA and the incumbents' rating inflation. These findings also shed light on the debate concerning whether CRAs with alternative business models can alleviate the rating inflation problem. They also generate policy implications regarding the distinct effects of different types of CRAs on the existing providers' rating strategies.

In the future study, we will estimate why investors would like to choose those reputable issuer-pay CRAs even they assign relatively lower ratings after CCR entrant. Moreover, this paper just reveals the change of rating inflation without mentioning the change of information content of ratings, thus we will estimate how the quality of incumbent issuer-pay CRAs change after CCR's rating initiation on a firm.

30

Appendix

Appendix 1 Market share of different issuer-pay Credit rating agencies

This table presents the market share of issuer-pay CRAs among different bond markets between 2006 and 2015. SCP is super commercial paper; CP is commercial paper; MTN is mid-term notes; Corporate bond is supervised by NDRC; Enterprise bond is issued by listed firms and supervised by CSRC. Panel A reports the market share based on the bond issue number. Panel B reports the market share based on the bond issue volume. Panel A: Market share based on issue number SCP CP MTN Corporate

bond Enterprise bond

Total

CCXI 43.98% 34.86% 35.17% 18.52% 0.00% 31.69% Lianhe 24.95% 24.56% 26.46% 15.89% 0.00% 22.36% Dagong 21.67% 22.75% 21.48% 20.34% 12.20% 21.35% SBRC 8.70% 17.03% 16.46% 10.45% 9.16% 14.23% Pengyuan 0.00% 0.00% 0.00% 30.79% 18.94% 6.10% CCXR 0.00% 0.00% 0.00% 0.08% 32.14% 1.67% UR 0.05% 0.00% 0.00% 0.23% 25.40% 1.35% Golden 0.64% 0.36% 0.43% 3.55% 2.16% 1.04% SFE 0.00% 0.45% 0.00% 0.15% 0.00% 0.20% Panel B: Market share based on issue volume SCP CP MTN Corporate

bond Enterprise bond

Total

CCXI 44.51% 38.92% 36.07% 20.38% 0.00% 34.32% Lianhe 27.14% 28.69% 36.84% 17.65% 0.00% 27.25% Dagong 23.30% 21.29% 17.93% 26.72% 15.89% 21.36% SBRC 4.53% 10.59% 8.91% 9.37% 8.00% 8.50% Pengyuan 0.00% 0.00% 0.00% 22.78% 8.64% 4.02% CCXR 0.00% 0.00% 0.00% 0.09% 39.94% 2.22% UR 0.33% 0.00% 0.00% 0.24% 26.58% 1.58% Golden 0.19% 0.17% 0.26% 2.64% 0.95% 0.63% SFE 0.00% 0.33% 0.00% 0.14% 0.00% 0.12%

31

Appendix 2 Difference-in-Difference analysis of rating without AAA rating.

This table presents the difference-in-difference analysing results. The normal sample consists of firm-year-CRA observations between 2006 and 2015. The Propensity Score Matching (PSM) sample consists of firm-year-CRA observations of treatment firms and firms are one-to-one matched firms from the control group. The firms in these two samples all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 8 (AA+). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Standard errors are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Raw Sample PSM matched Sample CCRcover*Post -0.095***

(0.028) -0.120*** (0.031)

CCRcover 0.396*** (0.025)

0.326*** (0.029)

Post Debt rate -1.152***

(0.063) -1.234*** (0.084)

ROA 0.009*** (0.002)

0.002 (0.002)

Tangibility 0.127*** (0.041)

0.411*** (0.047)

Sales 0.199*** (0.006)

0.229*** (0.008)

Growth 0.001* (0.001)

0.007 (0.007)

Constant -220.552*** (9.000)

-232.806*** (11.992)

Industry fixed effects Yes Yes Year dummies Yes Yes Observations 13,239 8,271 R-squared (adjusted) 0.201 0.181

32

Appendix 3 Difference-in-Difference analysis of rating: categorized by the reputation of issuer-pay Credit Rating Agencies without AAA rating.

This table presents the difference-in-difference analysing results categorized by reputable and non-reputable Credit Rating Agencies. The reputable group consists of two CRAs, CCXI and Lianhe. The non-reputable group consists of other seven issuer-pay CRAs. The sample consists of firm-year-CRA observations between 2006 and 2015. The firms in the sample all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 8 (AA+). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Robust standard errors clustered at the firm level are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Rating (1) (2) Reputable Non-reputable Reputable Non-reputable CCRcover*Post -0.127

(0.083) 0.065 (0.070)

-0.162*** (0.061)

-0.019 (0.059)

Diff of coefficient

-0.192*

(chi2=3.24)

-0.143*

(chi2=3.07)

CCRcover 0.363*** (0.081)

0.289*** (0.082)

0.402*** (0.073)

0.381*** (0.077)

Post 0.415*** (0.061)

0.276*** (0.041)

Debt rate -1.380*** (0.262)

-0.976*** (0.201)

-1.398*** (0.260)

-0.944*** (0.196)

ROA 0.000 (0.013)

0.007 (0.012)

0.005 (0.013)

-0.013 (0.011)

Tangibility -0.032 (0.145)

-0.121 (0.156)

0.172 (0.145)

0.097 (0.155)

Sales 0.231*** (0.025)

0.180*** (0.022)

0.231*** (0.025)

0.172*** (0.022)

Growth 0.002** (0.001)

0.001*** (0.000)

0.001* (0.001)

0.001** (0.001)

Constant 6.461*** (0.128)

6.514*** (0.105)

-232.854*** (23.989)

-210.487*** (22.414)

Industry fixed effects No No Yes Yes Year dummies No No Yes Yes Observations 6,394 6,845 6,394 6,845 R-squared (adjusted) 0.165 0.150 0.209 0.195

33

Appendix 4 Difference-in-Difference analysis of rating: categorized by the reputation of issuer-pay Credit Rating Agencies for high-frequency firms

This table presents the difference-in-difference analysing results categorized by reputable and non-reputable Credit Rating Agencies for high-frequency firms. The reputable group consists of two CRAs, CCXI and Lianhe. The non-reputable group consists of other seven issuer-pay CRAs. The firms all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Robust standard errors clustered at the firm level are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Rating (1) (2) Reputable Non-reputable Reputable Non-reputable CCRcover*Post -0.466***

(0.109) 0.096 (0.144)

-0.316*** (0.050)

-0.082 (0.080)

Diff of coefficient

-0.562***

(chi2=10.46)

-0.234**

(chi2=6.37)

CCRcover 0.411*** (0.140)

-0.097 (0.179)

0.366*** (0.125)

0.063 (0.164)

Post 0.714*** (0.097)

0.156 (0.127)

Debt rate -1.284*** (0.476)

-0.348 (0.438)

-1.409*** (0.479)

-0.369 (0.441)

ROA 0.001 (0.001)

-0.033*** (0.012)

0.001 (0.001)

-0.027** (0.012)

Tangibility 0.290 (0.224)

0.415 (0.279)

0.473** (0.226)

0.605** (0.271)

Sales 0.367*** (0.035)

0.319*** (0.031)

0.348*** (0.035)

0.300*** (0.033)

Growth 0.000 (0.000)

0.001** (0.000)

0.000 (0.008)

0.000 (0.000)

Constant 6.179*** (0.275)

6.456*** (0.276)

-277.307*** (29.073)

-189.669*** (43.101)

Industry fixed effects No No Yes Yes Year dummies No No Yes Yes Observations 5,073 3,014 5,073 3,014 R-squared (adjusted) 0.343 0.325 0.376 0.349

34

Appendix 5 Difference-in-Difference analysis of rating for particular time period.

This table presents the difference-in-difference analysing results. The sample consists of firm-year-CRA observations from 2009 to 2011 and from 2013 to 2015. The firms in the sample all have ratings both before and after the entry of CCR. The left-hand-side variable is the cardinal value of credit ratings from all issuer-pay CRAs from 1 (below BBB) to 9 (AAA). Post is a dummy variable that equals one if a firm-year-CRA observation is from the period after 2012, and zero otherwise; CCRcover is a dummy variable that equals one if a firm was rated by CCR at any time, and zero otherwise; there are 870 firms covered by CCR. Sales is the natural logarithm of sales; Debt rate is the ratio of total liability from the balance sheet to total assets; ROA is the return on assets which represents the profitability; Tangibility is the ratio of net property, plant and equipment to total assets; Growth is the year to year increase of operating income; all above variables are measured at the time t-1. Industry fixed effects are indicator variables for firms’ industry; year dummies are indicator variables for fiscal years. Standard errors are in the parentheses. R-squared are reported. ***, ** and * indicate statistical significance at the 1%, 5% and 10% levels, respectively.

Raw sample PSM matched sample

(1) (2) (3) (4) CCRcover*Post -0.094***

(0.034) -0.105*** (0.030)

-0.067 (0.048)

-0.085** (0.036)

CCRcover 0.432*** (0.028)

0.460*** (0.027)

0.296*** (0.038)

0.322*** (0.032)

Post 0.243*** (0.026)

0.221*** (0.044)

Debt rate -1.646*** (0.060)

-1.674*** (0.588)

-1.603*** (0.072)

-1.662*** (0.072)

ROA 0.001 (0.001)

0.001* (0.001)

0.001 (0.001)

0.001 (0.001)

Tangibility 0.204*** (0.041)

0.357*** (0.041)

0.491*** (0.045)

0.592*** (0.045)

Sales 0.364*** (0.005)

0.356*** (0.005)

0.390*** (0.006)

0.381*** (0.006)

Growth 0.001** (0.001)

0.001* (0.001)

0.013* (0.007)

0.011 (0.007)

Constant 6.451*** (0.038)

-135.190*** 11.338

6.334*** (0.058)

-125.879*** (16.078)

Industry fixed effects No Yes No Yes Year dummies No Yes No Yes Observations 13,541 13,541 9,440 9,440 R-squared (adjusted) 0.378 0.395 0.339 0.352

35

Reference Andres, C, Betzer, A & Limbach, P 2014, 'Underwriter reputation and the quality of certification: Evidence from

high-yield bonds', Journal of Banking & Finance, vol. 40, pp. 97-115. Baghai, R, Servaes, H & Tamayo, A 2013, 'Have rating agencies become more conservative? Implications for

capital structure and debt pricing', Journal of Finance, forthcoming. Bar-Isaac, H & Shapiro, J 2013, 'Ratings quality over the business cycle', Journal of Financial Economics, vol. 108,

no. 1, pp. 62-78. Beattie, VA & Searle, SH 1992, 'Bond ratings and inter-rater agreement', Journal of International Securities

Markets (Summer), pp. 167-72. Beaver, W, Shakespeare, C & Soliman, M 2006, 'Differential properties in the ratings of certified versus non-

certified bond-rating agencies', J. Account. Econ., vol. 42, no. 3, pp. 303-34. Becker, B & Milbourn, T 2011, 'How did increased competition affect credit ratings?', Journal of Financial

Economics, vol. 101, no. 3, pp. 493-514. Becker, B & Opp, M 2013, Regulatory reform and risk-taking: Replacing ratings, National Bureau of Economic

Research. Behn, M, Haselmann, RF & Vig, V 2014, 'The limits of model-based regulation', SAFE Working Paper No. 75. Bolton, P, Freixas, X & Shapiro, J 2012, 'The Credit Ratings Game', J. Financ., vol. 67, no. 1, pp. 85-111. Bongaerts, D 2013, 'Can alternative business models discipline credit rating agencies', Unpublished working paper,

Erasmus University. Bongaerts, D 2014, 'Alternatives for issuer-paid credit rating agencies', European Central Bank. Bongaerts, D, Cremers, KJM & Goetzmann, WN 2012, 'Tie breaker:certification and multiple credit ratings',

Journal of Finance, vol. 67, no. 1, pp. 113-52. Bottelier, P 2003, ' China's emerging domestic debt markets: facts and issues.', Discussion paper for Conference on

Chinese Economic Policy Reform organized by the Center for Research on Economic Development and Policy Reform.

Brockman, P & Chung, DY 2003, 'Investor protection and firm liquidity', The Journal of Finance, vol. 58, no. 2, pp. 921-38.

Cantor, R & Packer, F 1997, 'Differences of opinion and selection bias in the credit rating industry', Journal of Banking & Finance, vol. 21, no. 10, pp. 1395-417.

Carow, KA 1999, 'Underwriting spreads and reputational capital: An analysis of new corporate securities', Journal of Financial Research, vol. 22, no. 1, pp. 15-28.

Chen, H 2003, 'China's credit rating industry: properity and worries.', The Banker, vol. 9, pp. 70-3. Chen, X, Shi, N & Zhang, X 2013, 'A study on the interest conflict of crdit rating based on the theory of reputation',

Economic Survey, vol. 3, pp. 150-5. Chen, Y 2010, 'The problems about China's credit rating industry and resolutions', Financial Teaching and

Research, vol. 2, pp. 59-62. Cinquegrana, P 2009, 'The reform of the credit rating agencies: A comparative perspective', ECMI Policy Brief, no.

12. Coface 2010, 'Press release',

<http://www.coface.com/CofacePortal/ShowBinary/BEA%20Repository/HK/en_EN/documents/wwa_news_events/20100729CofaceCESR-HK_en>.

Cole, HL & Cooley, TF 2014, Rating agencies, National Bureau of Economic Research. Cornaggia, J & Cornaggia, KJ 2013, 'Estimating the Costs of Issuer-Paid Credit Ratings', The Review of Financial

Studies, vol. 26, no. 9, pp. 2229-69. Covitz, D & Harrison, P 2003, 'Testing Conflicts of Interest at Bond Rating Agencies with Market Anticipation:

Evidence that Reputation Incentives Dominate', SSRN Working Paper Series. Deb, P, Manning, MJ, Murphy, G, Penalver, A & Toth, A 2011, 'Whither the credit ratings industry?', Bank of

England Financial Stability Paper, no. 9. Diomande, MA, Heintz, JS & Pollin, RN 2009, 'Why US financial markets need a public credit rating agency', The

Economists' Voice, vol. 6, no. 6. Fan, G, Wang, X & Zhu, H 2011, NERI INDEX of Marketization of China's Provinces 2011 Report., Economics

Science Press, Beijing. Fennell, D & Medvedev, A 2011, 'An economic analysis of credit rating agency business models and ratings

accuracy', FSA Occasional paper, no. 41.

36

Fitch 1998, Fitch IBCA Announces Rating Joint Venture in Beijing, <http://www.thefreelibrary.com/Fitch+IBCA+Announces+Rating+Joint+Venture+in+Beijing+-+Fitch+IBCA+--a021117106>.

Fitch 2004, Fitch Ratings Announces Divestment of China Chengxin International Credit Rating Co., <http://www.prnewswire.com/news-releases/fitch-ratings-announces-divestment-of-china-chengxin-international-credit-rating-co-71322817.html>.

Fleisher, BM 2008, 'China's credit rating industry: properity and worries.', in BM Fleisher, NC Hope, AA Pena & DT Yang (eds), Policy Reform and Chinese Markets: Progress and Challenges, Edward Elgar, Cheltenham, pp. 159-78.

Gonzalez, R, Sotelino, F & Savoia, J 2011, 'Simplified Standard Approach: an alternative model to mitigate incentives for rating centered regulatory arbitrage', Available at SSRN 1836223.

Goodhart, CA 2008, 'The regulatory response to the financial crisis', Journal of Financial Stability, vol. 4, no. 4, pp. 351-8.

He, J, Qian, J & Strahan, PE 2011, 'Credit Ratings and the Evolution of the Mortgage-Backed Securities Market', American Economic Review, vol. 101, no. 3, pp. 131-5.

Heckman, JJ 1980, 'Sample selection bias as a specification error', Female labor supply: Theory and estimation, pp. 206-48.

Hill, CA 2004, 'Regulating the rating agencies', Wash. ULQ, vol. 82, p. 43. House, R 1995, 'Ratings trouble. (credit rating agencies)(includes interview with Standard & Poor's Pres. Leo

O'Neill)', Institutional Investor, vol. 29, no. 10, p. 245. IMF 2009, 'World economic outlook: Crisis and recovery', <http://www.imf.org/external/pubs/ft/weo/2009/01/>. Jiang, J, Harris, S, Mary & Xie, Y 2012, 'Does it matter who pays for bond ratings? Historical evidence', Journal of

Financial Economics, vol. 105, no. 3, pp. 607-21. Jones, WC 2003, 'Trying to understand the current Chinese legal system', Understanding China’s Legal System:

Essays in Honor of Jerome A. Cohen. New York University Press, New York and London. Kashyap, AK & Kovrijnykh, N 2016, 'Who Should Pay for Credit Ratings and How?', Review of Financial Studies,

vol. 29, no. 2, pp. 420-56. Kennedy, S 2003, 'China's credit rating agencies struggle for relevance', China Business Review, vol. 6, no. 30, pp.

36-40. Kennedy, S 2008, 'China's emerging credit rating industry: the official foundations of private authority', The China

Quarterly, no. 193, p. 65. Lee, JL 2006, 'Credit raters in China take generous view.', Wall Street J. Li, J, Shin, YS & Moore, WT 2006, 'Reactions of Japanese markets to changes in credit ratings by global and local

agencies', Journal of Banking and Finance, vol. 30, no. 3, pp. 1007-21. Li, S & Filer, L 2007, 'The effects of the governance environment on the choice of investment mode and the

strategic implications', Journal of World Business, vol. 42, no. 1, pp. 80-98. Li, Z 2013, 'An Empirical Analysis of the Relationship between China's Bond Credit Rating and Issuance Rates',

Securities & Futures of China, vol. 9, pp. 279-81. Lynch, TE 2010, 'Deeply and Persistently Conflicted: Credit Rating Agencies in the Current Regulatory

Environment', Indiana Legal Studies Research Paper, no. 133. Mathis, J, McAndrews, J & Rochet, J-C 2009, 'Rating the raters: Are reputation concerns powerful enough to

discipline rating agencies?', Journal of Monetary Economics, vol. 56, no. 5, pp. 657-74. Megginson, WL & Weiss, KA 1991, 'Venture capitalist certification in initial public offerings', The Journal of

Finance, vol. 46, no. 3, pp. 879-903. Nickell, P, Perraudin, W & Varotto, S 2000, 'Stability of rating transitions', Journal of Banking & Finance, vol. 24,

no. 1, pp. 203-27. Nielsen, N 2013, 'EU-based credit rating agency buried', <http://euobserver.com/foreign/120005>. Opp, CC, Opp, MM & Harris, M 2013, 'Rating agencies in the face of regulation', Journal of Financial Economics,

vol. 108, no. 1, pp. 46-61. Packer, F 2002, Credit ratings and the Japanese corporate bond market, Springer. Pagano, M & Volpin, P 2010, 'Credit ratings failures and policy options', Economic Policy, vol. 25, no. 62, pp. 401-

31. Partnoy, F 1999, 'The Siskel and Ebert of financial markets: two thumbs down for the credit rating agencies',

Washington University Law Quarterly, vol. 77, pp. 619-712. PeopleDaily 2006, Moody's acquires 49 pct stake in Chinese rating agency,

<http://en.people.cn/200604/15/eng20060415_258515.html>.

37

Poon, WPH & Chan, KC 2008, 'An empirical examination of the informational content of credit ratings in China', Journal of Business Research, vol. 61, no. 7, pp. 790-7.

Prabhala, N & Li, K 2007, 'Self-selection models in corporate finance', Handbook of Corporate Finance, Espen Eckbo ed.

E Commission 2010, Public Consultation on Credit Rating agencies. Nov 5, 2010. Richardson, M & White, LJ 2009, 'The Rating Agencies: Is Regulation the Answer?', Financial Market,

Institutions & Instruments, vol. 18, no. 2, pp. 146-8. Ross, DG 2010, 'The “dominant bank effect:” How high lender reputation affects the information content and terms

of bank loans', Review of Financial Studies, vol. 23, no. 7, pp. 2730-56. Schenone, C 2004, 'The effect of banking relationships on the firm's IPO underpricing', The Journal of Finance, vol.

59, no. 6, pp. 2903-58. SEC 2002, Financial oversight of Enron: the SEC and private-sector watchdogs, SEC. SEC 2012, Report to Congress on Assigned Credit Ratings, ----. SEC 2014, SEC Adopts Credit Rating Agency Reform Rules, ----. Shin, YS & Moore, WT 2003, 'Explaining credit rating differences between Japanese and US agencies', Review of

Financial Economics, vol. 12, no. 4, pp. 327-44. Skreta, V & Veldkamp, L 2009, 'Ratings shopping and asset complexity: A theory of ratings inflation', Journal of

Monetary Economics, vol. 56, no. 5, pp. 678-95. Song, Z 2013, 'Credit Rating in China’s Bond Market: Evidence from Short-Term Financing Bonds', Modern

Economy, vol. 04, no. 02, p. 119. Spiegel, D 2012, 'Dearth of investors: Plan to set up european rating agency under threat',

<http://www.spiegel.de/international/europe/plan-to-set-up-european-rating-agency-is-failing-according-to-newspaper-a-827876.html.>.

Sy, ANR 2009, 'The systemic regulation of credit rating agencies and rated markets', IMF Working Paper. Tu, Y 2006, 'Dispersion of domestic 108 corporate commercial paper credit spread levels more marked day by day.',

China Securities News, 25 April. Wang, X & Zhang, C 2013, 'Reputation mechanism, credit rating and Mid-term Note financing cost', Joural of

Financial Research, vol. 8, pp. 58-68. White, LJ 2002, 'The credit rating industry: An industrial organization analysis.', in RM Levich, C Reinhart & G

Majnoni (eds), Ratings, rating agencies, and the global financial system, Kluwer, Boston, pp. 41-63. White, LJ 2010, 'Markets: The Credit Rating Agencies', Journal of Economic Perspectives, vol. 24, no. 2, pp. 211-

26. Xia, H 2013, 'Can investor-paid credit rating agencies improve the information quality of issuer-paid rating

agencies?', Journal of Financial Economics, vol. 111, no. 2, pp. 450-68. Xu, W & Han, D 2013, 'Factors affect the credibility of China's rating agencies', Seeker, vol. 12, pp. 221-3. Zhang, Q & Zhang, B 2010, 'A study on the reputation mechanism of credit rating agencies of our country in the

background of financial crisis', Economic Survey, vol. 1, pp. 150-3. Zhang, S 2013, 'The explanation function on credit rating to bond spreads', Modern Finance and Economics-

Journal of Tianjin University of Finance and Economics, vol. 1, pp. 71-9.

38