Compensation Incentives of Credit Rating Agencies and Predictability of Changes in Bond. Ratings and Financial Strength Ratings *

Transcription

1 Compensation Incentives of Credit Rating Agencies and Predictability of Changes in Bond Ratings and Financial Strength Ratings * Andreas Milidonis April 2013 Please address correspondence to: Andreas Milidonis, Ph.D., Department of Accounting & Finance Faculty of Economics & Management University of Cyprus P.O. Box CY-1678 Nicosia, Cyprus Andreas.Milidonis@ucy.ac.cy Tel: Fax: * I would like to thank Massimo Guidolin, Irene Karamanou, Ty Leverty, Gunter Löffler, George Nishiotis, Konstantinos Stathopoulos and participants at the Annual Meeting of the American Risk and Insurance Meeting, the European Group of Risk and Insurance Economists and the workshop for Finance, Risk and Banking organized by the University of Ulm. I would also like to thank Egan-Jones Ratings for providing access to their ratings ( ) William Beaver, Catherine Shakespeare and Mark Soliman for generously sending their data ( ), and Maria Efthymiou for excellent research assistance. This paper has also benefited from a research grant by the University of Cyprus. All errors are mine. Electronic copy available at:

2 Compensation Incentives of Credit Rating Agencies and Predictability of Changes in Bond Ratings and Financial Strength Ratings Abstract Over the past decade there has been mixed evidence on the lead-lag relation between issuerpaid and investor-paid credit rating agencies. We investigate the lead-lag relationship for changes in bond ratings (BRs) and financial strength ratings (FSRs), for the US insurance industry, where FSRs impose market discipline. First, we find that changes in issuer-paid BRs are led by changes in investor-paid BRs, even over a period that issuer-paid agencies have improved their timeliness. Second, information flows in both directions between changes in issuer-paid BRs and FSRs. Third, issuer-paid FSRs are predictable by investorpaid BRs. Fourth, the lead effect of investor-paid downgrades is economically significant as it is associated with an unconditional, post-event, thirty-day cumulative abnormal return of - 4%. This return is a result of investor-paid downgrades in BRs, which predict more downgrades in the following ninety days (same period return of -11%). Keywords: Credit Rating Agencies; Information Dissemination; Timeliness; Predictability; Insurance. JEL Code: G14; G22; G24. 1 Electronic copy available at:

3 1. Introduction Information intermediaries such as credit rating agencies and their incentives play a valuable role in the financial reporting environment given financial markets increased complexity and interconnectedness (Beyer et al., 2010). Their role becomes even more significant in industries that are more opaque, such as the insurance industry (Morgan, 2002), where investors, consumers but also regulators use ratings as proxies for the financial vulnerability of rated entities (e.g., Grace et al., 2004; Weiss and Chung, 2004; Pottier, 2007; Chen et al., 2008). The insurance industry is unique in several aspects and should be examined in isolation from other industries with respect to the types of ratings that characterize it, but also the timeliness of these ratings, for the following reasons: (a) the industry is governed by a market discipline mechanism, which has been shown to be dependent on ratings (Epermanis and Harrington, 2006), (b) in contrast to non-financial industries, it is rated on two major types of risks (ability to repay corporate debt, and ability to repay promises on policyholders contracts), (c) it is rated by the two types of credit rating agencies (issuer-paid and investor-paid), and (d) it has a homogeneous regulatory structure, which is quite different from other non-financial industries. Why are the different types of credit rating agencies important? Critics of issuer-paid rating agencies (they are compensated by the companies they rate) argue that the potential conflict of interest present in the compensation structure of issuer-paid agencies may distort the timely release of changes in their ratings (e.g. White, 2010). In contrast, investor-paid agencies (collect subscription fees from investors to rate third parties) strive to inform their subscribers of potential changes in the credit risk of rated firms and hence have an incentive to supply the market with more timely ratings than their issuerpaid competitors. Over the past decade there has been mixed evidence on the relative timeliness between the two types of rating agencies. Specifically, Johnson (2004) and Beaver et al. (2006) show that changes in bond ratings by investor-paid agencies lead those of issuer-paid agencies. On the other hand, Cheng and Neamtiu (2009) show that issuer-paid agencies have improved their timeliness (relative to themselves) for bond ratings that are close to default, while Berwart et al., (2012) show that issuer-paid agencies have improved their bond ratings timeliness relative to those of investor-paid agencies. 2

4 What are the two major types of ratings published for the insurance industry? Credit rating agencies issue bond ratings (BRs) to rate the credit risk inherent in insurance corporate bond obligations. In addition, rating agencies issue financial strength ratings (FSRs), to rate the overall ability of insurance firms to fulfill policyholder obligations. 1 FSRs have been associated with market discipline in the insurance industry (Epermanis and Harrington, 2006) and also significant market reactions with the announcement of their changes (Halek and Eckles, 2010). Insurance firms typically solicit and pay for FSRs from issuer-paid rating agencies. Most BRs are also produced by issuer-paid agencies, but some BRs are published by investor-paid rating agencies. The relative timeliness of issuer- vs. investor-paid agencies in the market for BRs has implications for FSRs as the insurance firms default and insolvency risks are intimately linked (Pottier and Sommer, 1999; Grace et al., 2003). [Figure 1] In this paper, we examine the unexplored lead-lag relationship between BRs and FSRs for the US insurance industry, across the two types of rating agencies. Then we examine the potential economic value for investors who follow the announcements of changes in these ratings. Our sample comprises issuer-paid and investor-paid rating agencies changes in BRs and FSRs for insurance companies from The largest rating agencies by market share (Fitch, Moody s and Standard & Poor s) are issuer-paid agencies that publish both BRs and FSRs. Investor-paid agencies (e.g., Egan Jones), on the other hand, only produce BRs (Figure 1). We examine the lead-lag relation between types of ratings within the same rating agency, and also across rating agencies using an ordered probit model that takes into account both the level and the timing of upgrades and downgrades, following Güttler and Wahrenburg (2007) and Alsakka and Gwilym (2010). Using event studies, we then examine if there are market reactions in the post-announcement periods of the two types of rating agencies, especially the leader in the lead-lag relations tested. First, we revisit the lead-lag relationship of investor- and issuer-paid BRs but specifically for the 1 In 2008, the total book value of policyholder obligations and corporate debt obligations amounted to about 81% (total policy reserves of 60.6% and separate accounts liabilities of 20%) and 9.9%, respectively, of the total reported insurance liabilities for publicly traded firms ( 3

5 insurance industry, over a period with conflicting evidence on the direction of the lead-lag relationship for all industries. We find that investor-paid rating agencies lead issuer-paid rating agencies in the market for insurance BRs, even after the improvement in the timeliness of issuer-paid rating agencies around the default barrier (Cheng and Neamtiu, 2009; Berwart et al., 2012). Second, we test the lead-lag relationship between changes in BRs and changes in FSRs, within the same rating agency. Our data shows that a significant percentage of changes in BRs and FSRs happens on the same day. Using only the observations that do not happen on the same day, we conduct a lead-lag analysis. We find that there is information spillover between the two types of ratings, that is, information flows in both directions, for those changes in BRs and FSRs. This result, along with the fact that several changes in BRs and FSRs happen on the same day, shows that the probability of default on corporate debt and policyholder obligations are connected. Our third research question follows from the previous two results. Since only issuer-paid rating agencies produce FSRs, then since investor-paid rating agencies lead issuer-paid rating agencies in the market for BRs, and since there is information flow between the two types of ratings, this leads us to ask: do changes in BRs by investor-paid rating agencies lead changes in FSRs by issuer-paid rating agencies? We find evidence that they do. This predictability is important because FSRs impose market discipline (Epermanis and Harrington, 2006), and significant cumulative average abnormal returns (CAARs) are associated with the changes in FSRs of issuer-paid rating agencies (Halek and Eckles, 2010). Our fourth question is: can investors benefit from the documented predictability above? Using short-term event study analysis, we first find that there are significant CAARs surrounding the announcements of changes in BRs by both investor-paid and issuer-paid rating agencies, with the CAARs associated with investor-paid agencies announcements being larger. Second, in line with prior literature (Halek and Eckles, 2010), we confirm the presence of significant CAARs associated with the announcements of changes in FSRs by issuer-paid agencies. In both cases (changes in BRs and changes in FSRs) results are larger in magnitude for downgrades than upgrades. Third, we find that investors can benefit from announcements of downgrades in BRs by the investor-paid rating agency, but not those 4

6 (either BRs or FSRs) of the issuer-paid rating agency. Specifically, we document a negative drift in the thirty trading days following downgrades by the investor-paid agency, which amounts to a cumulative abnormal return of -3.92%. Using identical methodology for the post-announcement periods of issuerpaid rating agencies, we do not find any significant cumulative abnormal returns. Then, we isolate those investor-paid downgrades followed by other downgrades in the subsequent ninety days, to show that the overall cumulative abnormal return of -3.92% is driven by the leading (investor-paid) downgrades (CAAR of %) and not the followers (no significant CAARs). Therefore, evidence suggests that the post-downgrade drift is due to lead-effect of the investor-paid agency and not an under-reaction to bond downgrades (Dichev and Piotroski, 2001). This is the first study to show that changes in ratings published by investor-paid rating agencies are not only predictive in the market for BRs, but they can also predict changes in FSRs, which serve as a mechanism for market discipline in the insurance industry. Because changes in premiums (Epermanis and Harrington, 2006) directly affect the demand for insurance products, markets appear to react with large CAARs surrounding the announcement of BRs by investor-paid rating agencies long before issuer-paid agencies confirm their competitors actions. Our results show that there are inefficiencies not only within a group of information intermediaries (i.e. rating agencies) and the way they publish new information (timeliness), but also in the way they categorize and assess seemingly different but related risks within a firm (i.e. using different types of ratings). The rest of this paper is organized as follows. In Section 2 we provide institutional background on the two types of ratings and the two types of rating agencies, and we develop our hypotheses. In Section 3 we describe the empirical methodology and in Section 4 we describe the data. We present our results and robustness checks in Section 5, and conclude in section Institutional Background, Sample Selection and Hypotheses 2.1. Financial Strength Ratings (FSRs) Insurance firms manage a large portfolio of risks, which they collect mainly in the form of premiums from policyholders and debt and equity from markets. Insurer financial strength ratings (FSRs) 5

7 represent rating agencies opinion about the probability that insurance firms will not be able to service their contractual policyholder obligations. FSRs are commonly used as a marketing tool for consumers since insurance brokers usually advise purchasing insurance only from insurance firms, which are rated three notches above the financially secure threshold (Bradford, 2003). FSRs impose market discipline in the insurance industry. Epermanis and Harrington (2006) document contemporaneous and one-year-ahead declines in premiums paid by risk-sensitive consumers with downgrades in FSRs. Prior research also finds evidence of a positive relation between FSRs and both, insurance prices (Doherty and Tinic, 1981; Berger et al., 1992) and revisions in liability loss reserves (Petroni et al., 2000), and a negative relation between firm risk (i.e., lower FSR), insurance stock prices (Fenn and Cole, 1994) and the price of insurance (Sommer, 1996; Cummins and Danzon, 1997; Phillips et al., 1998). However, while FSRs are used as proxies for the financial solvency of insurers by regulators (Schwartz, 1994) and researchers (Adiel, 1996; Anthony and Petroni, 1997; Cummins and Danzon, 1997; Pottier, 1998), regulation for insurance firms usually is not binding in terms of capital requirements (e.g., Cummins et al., 1995; Grace et al., 1998) and hence FSRs remain an important determinant of financial health for non-distressed insurance firms. Pottier (1997) studies the determinants of FSRs, differences between rating agencies active in the market for FSRs, as well as the decision of insurance firms to be rated, and Pottier (1998) shows that downgrades in FSRs have some explanatory power for insolvencies in the life-insurance industry. Halek and Eckles (2010) further show that investor reactions to announcements of changes in FSRs of publicly traded insurance firms depend in part on the type of rating agency making the announcement, and Halek and Eckles (2011) examine the determinants of CAARs using firm-specific characteristics of insurance firms. To date, however, no study examines the relation between FSRs and BRs for insurance firms Bond Ratings (BRs) BRs represent the opinion of rating agencies about the probability that debt-issuing firms will not 6

8 service their debt obligations (debt default risk is not incorporated in FSRs). 2 The Securities and Exchange Commission (SEC) monitors the investment decisions of institutional investors (i.e., banks, mutual funds and insurance firms) using only BRs from Nationally Recognized Statistical Rating Organizations (NRSROs). Because the SEC considers investments to be speculative if their respective BR falls below the investment grade boundary set by rating agencies, and because institutions are required to invest almost entirely in investment grade bonds (Kisgen and Strahan, 2010; Ellul et al., 2011), debtissuing insurance firms need to obtain a BR from at least one NRSRO to make their debt issues appealing to institutional investors. The NRSRO certification was first introduced in 1975 and until 2001 it was only granted to the largest three issuer-paid rating agencies by market share (Fitch Ratings, Moody s Ratings and Standard & Poor s Ratings). These agencies are compensated based on the issuer-paid model. According to this model, when a company wants to issue debt it will solicit a rating from one (or more) of the NRSROs. 3 Meetings between representatives of the rating agency and the debt-issuing firm before the final publication of a rating are not uncommon, and the fee paid by the debt-issuing firm usually depends on the firm s size and the complexity of the issuance (Kliger and Sarig, 2000). As a result, ratings of these rating agencies have a dual role as they are used not only for contracting (Asquith et al., 2005) and monitoring purposes by the SEC, but also for valuation purposes by investors. Weinstein (1977) and Pinches and Singleton (1978) were the first to study the information value of changes in BRs and find no significant market reaction related to their announcement. The opposite results obtain in more recent studies that consistently provide evidence of investor reactions to changes in BRs (e.g., Holthausen and Leftwich, 1986; Hand et al., 1992; Goh and Ederington, 1993, 1999; Ederington and Goh, 1998). In addition, researchers have studied the determinants of BRs (Kaplan and Urwitz, 1979; Ederington, 1985; Cantor and Packer, 1997), their relation to bond yields (Kliger and Sarig, 2000; Tang, 2009; Kisgen and Strahan, 2010), their relation to accounting quality (e.g. Blume et al., 1998; 2 We use the term BRs to refer to the commonly used term credit ratings. BRs reflect on the creditworthiness of the firm while FSRs on the insolvency probability; a firm can have low creditworthiness but still be solvent. 3 In cases in which debt-issuing firms do not solicit a rating, a rating agency might still publish a rating, albeit based entirely on publicly available information. 7

9 Sengupta, 1998), their differences across agencies (Ederington, 1986), ratings shopping (Skreta and Veldkamp, 2009), post-announcement drift (Dichev and Piotroski, 2001) and conflicts of interest associated with ratings shopping (Bolton et al., 2012). The timeliness of changes in BRs has received a lot of attention in the literature and has been linked to heterogeneity in rating agencies compensation structure (Beaver et al., 2006; Milidonis and Wang, 2007; Strobl and Xia, 2010; Bruno et al., 2012; Berwart et. al, 2012). Popular examples include the delayed downgrades of Enron (December 2, 2001) and Worldcom (July 21, 2002), which were considered to have investment grade debt until a few days before their collapse. The valuation effect of untimely changes in BRs (i.e., the losses or unrealized gains from late adjustments of trading strategies by investors) has provided opportunities for new rating agencies to enter the market. One such new investor-paid rating agency is Egan Jones Ratings (EJR), established in EJR entered the market for BRs quite aggressively by criticizing both the issuer-paid model of its competitors that can create conflicts of interest for a rating agency, and also the small number of NRSROs. Unlike its major issuer-paid competitors, EJR s compensation structure follows the investor-paid model, which means that EJR sells subscriptions to institutional investors and rates (third-party) companies on demand. This compensation model has normally precluded EJR from benefiting from close relations with key managers and chief financial officers at the rated companies, which is customary (but not compulsory) when the ratings are requested and paid for by companies issuing bonds. After several years of arguing in favor of opening the market for BRs (SEC, 2002), EJR was granted the NRSRO designation by the SEC on December 21, Sample Selection In this study we use EJR as the representative investor-paid rating agency, for the following reasons: (a) it bases its ratings only on publicly available information, (b) it rates several industries (financial, industrial, service sectors and more recently sovereign debt) (c) several academic studies have chosen EJR to be the representative investor-paid CRA over the past decade (Johnson, 2004; Beaver et al., 2006; Bruno et al., 2012; Berwart et al., 2012), (d) it was among the most vocal CRAs that 8

10 participated in the SEC debates on the grounds on which the NRSRO certification is granted, (e) it is one of the oldest investor-paid CRAs, i.e. since 1995, (f) it received a lot of attention in the media, (g) it was among the new CRAs that received NRSRO status in December Turning to issuer-paid rating agencies, we focus on the largest rating agencies by market share that are well established in the markets for both BRs and FSRs. Ratings by A.M. Best are not used for three reasons: (a) BRs for A.M. Best are only available since 2005; (b) their ratings went through several phases during the sample period (Doherty and Phillips (2002), find that the properties of A.M. Best rating from when S&P entered the market for FSRs); (c) the regulatory status of A.M. Best changed in 2001, when it became an NRSRO. We focus on changes in BRs and FSRs by Fitch and S&P. S&P entered the market for FSRs in the 1980s to compete with A.M. Best. However, their rating standards until 1994 were capped at a maximum rating of BBB regardless of the firm s financial position and any ratings by competitors (Doherty, et al., 2012). Our sample period begins in (May) 1996, when S&P was rating about 50% of the insurance industry by value of assets. The second issuer-paid rating agency is Fitch, which strengthened its presence in the insurance sector after the acquisition of Duff & Phelps Credit Rating Co. in Our sample for Fitch thus begins in January of Moody s does not have large coverage of the market for insurance FSRs, and it is therefore excluded from the sample Hypotheses Given the unique regulatory structure of the US insurance industry, and the importance that insurance companies place on their creditworthiness to conduct business, it is important to identify if there is a lead effect by one of the two types of CRAs in updating insurance BRs. Moreover, given that insurance companies (and banks) are more opaque than other industries when it comes to the ratings assigned by issuer-paid CRAs (Morgan, 2002), the timeliness of their announcements may also be affected, hence it is important to examine the insurance industry in isolation from the remaining industries. As already mentioned investor-paid agencies have stronger incentives than issuer-paid agencies 9

11 to update their ratings faster, given that their compensation is heavily dependent on institutional investor demand. At the same time however, starting in 2002, issuer-paid rating agencies were subject to heavy criticism and regulatory scrutiny, such as: (a) the Sarbanes-Oxley Act in July 2002 and its reference to CRAs, (b) The SEC Review of the role of CRAs in January 2003 (SEC, 2003), (c) the Congress Hearing on the different types and operations of CRAs in April 2003, and (d) The Credit Rating Duopoly Act in September Cheng and Neamtiu (2009) use a sample of companies that are close to default, before 2002 and after 2003, to find that this pressure on Fitch, Moody s and S&P resulted in improvements in the timeliness of their BRs, in predicting defaults. This finding suggests that since the relative timeliness among issuer-paid rating agencies has improved close to the default boundary, the relative timeliness between investor- and issuer-paid rating agencies may have also changed since 2002 (i.e. when compared to the results of Beaver et al., 2006). In line with this argument, Berwart et al. (2012) find evidence consistent with Cheng and Neamtiu (2009), i.e. that issuer-paid agencies have improved their timeliness, relative to the investor-paid agencies. Therefore, our preliminary hypothesis revisits the lead-lag relationship between issuer- and investor-paid agencies (Beaver et al., 2006), specifically for insurance BRs, over a period that the NRSRO status of the issuer-paid and the investor-paid agencies has not changed, but their relative timeliness seems to have changed: H0: Changes in insurance bond ratings by the investor-paid rating agency will lead (predict) changes in insurance bond ratings by the issuer-paid rating agencies. Our second research question focuses on the relation between changes in BRs and changes in FSRs within the same rated firm. To our knowledge, no prior study investigates the potential link between these two types of ratings. We know that policyholders have priority over debtholders in the case of firm insolvency (Pottier and Sommer, 1999; Grace et al., 2003). However there are three possible scenarios for the relative timing of changes in FSRs and BRs. First, a change in BRs might take place before a change in FSR. This would happen if deterioration in servicing debt obligations justifies a change in BR but is not severe enough to justify a change in FSR, for example, if an increase in the probability of not 10

12 servicing the firm s debt obligations does not affect the probability of not meeting the firm s policyholder obligations. Second, a change in FSR and a change in BR might happen on the same day. This would occur if either deterioration in default risk (change in BR) is severe enough to cause a change in FSR or some other adverse condition causes a change in FSR; in either case, the BR is affected as it is subordinate to the FSR. Because such a scenario would imply simultaneous changes in FSR and BR (regardless of causation), the two would be empirically inseparable, and hence in our models we exclude events taking place on the same day. Third, a change in FSR might happen before a change in BR. This could happen because issuer-paid rating agencies are conservative with changes in BRs given that the NRSRO certification applies to the market for BRs and not FSRs. Therefore, given the three potential relationships between changes in FSRs and changes in BRs, we do make a prediction about the direction of the lead-lag relation between changes in BRs and changes in FSRs within the same rated firm. The third research question we examine builds on the above lead-lag relations. If changes in investor-paid s BRs lead changes in issuer-paid s FSRs (H0), and if changes in issuer-paid s BRs lead changes in issuer-paid s FSRs (or changes in FSRs do not occur before changes in BRs), then we expect that not only will the investor-paid s changes in BRs lead the issuer-paid s changes in BRs, but they will also lead the issuer-paid s changes in FSRs. In the event that H0 is confirmed but either the relation between changes in FSRs and changes in BRs is not conclusive or changes in FSRs leads changes in BRs, then it remains an empirical question as to whether investor-paid changes in BRs lead issuer-paid changes in FSRs. Our remaining hypotheses examine investor reactions to announcements of changes in the ratings used in the previous lead-lag relations. We start by explicitly testing for CAARs to changes in BRs. To our knowledge no prior study explicitly investigates stock return reactions to changes in BRs for insurance firms, mainly because studies on the insurance industry focus on changes in FSRs. In the appendix (Figure A1) we show that the insurance market index behaves differently than market indices typically used in other studies. A more accurate method to test for such abnormal returns would therefore be to use an industry benchmark (industry return) (Halek and Eckles, 2010). The economic foundation for 11

13 our next hypothesis is based on the information value contained in BRs. Given BRs capture a different type of risk than FSRs, in a market of semi-strong or weak form efficiency we expect abnormal stock returns to be associated with changes in BRs for the sample of publicly traded insurance companies. Rating agencies are expected to process public and/or private information better than investors and hence to receive greater value from doing so (Ramakrishnan and Thakor, 1984, Millon and Thakor, 1985, Boot et al., 2006). Splitting our sample of changes in BRs into upgrades and downgrades, we thus hypothesize that: H1a: Downgrades in insurance bond ratings (BRs) are associated with negative cumulative average abnormal returns (CAARs). H1b: Upgrades in insurance bond ratings (BRs) are associated with positive cumulative average abnormal returns (CAARs). There is ample evidence in the finance and accounting literature of an asymmetry in reactions to downgrades and upgrades. Watts (1977) argues that the potential marginal loss from correcting a positive bias is larger than the potential marginal gain from correcting a negative bias. In the case of changes in ratings, a positive (negative) bias would be corrected by a(n) downgrade (upgrade). Consistent with this argument, several empirical studies find larger (absolute) abnormal returns associated with downgrades than upgrades (Holthausen and Leftwich, 1986; Hand et al., 1992; Ederington and Goh, 1998). Since insurers depend on their reputation and creditworthiness to sell insurance, a downgrade could send the company into a downward spiral; the promise to reimburse future policyholder claims is expected to sound less trustworthy after a downgrade, and hence the subsequent non-renewal of insurance business may adversely affect the present value of future premiums and consequently cause a decline in stock prices. In the opposite case, even though an upgrade may bring additional business to the firm, we expect that extra effort will be needed by the firm to capture the same amount of business that the company would lose in the case of a downgrade. These arguments lead to the following hypothesis: H2: Downgrades in insurance bond ratings are associated with larger (absolute) cumulative average abnormal returns (CAARs) than upgrades. 12

14 Following Beaver et al. (2006) we expect that the CAARs to announcements of downgrades in BRs by the investor-paid rating agency will be larger than the respective reactions to downgrades in BRs by its issuer-paid competitors. The rationale is that the investor-paid rating agency strives to publish timelier downgrades in BRs and hence provide greater value to their subscribers, since the investor-paid rating agency s compensation model is more tightly aligned with subscribers interests than the compensation model of issuer-paid rating agencies. Our next hypotheses are thus as follows: H3a: Downgrades in insurance bond ratings by the investor-paid agency are associated with more negative CAARs than those of the issuer-paid agencies. H3b: Upgrades in insurance bond ratings by the investor-paid agency are associated with more positive CAARs than those of the issuer-paid agencies Is there Economic Value in the Predictability of Changes in Ratings? We expect that investor-paid rating agencies will lead issuer-paid agencies (Beaver et al., 2006), and we know that issuer-paid changes in FSRs are associated with statistically significant CAARs (Halek and Eckles, 2010). Furthermore, we expect that changes in bond ratings will be associated with CAARs (H2). Therefore, we expect that the post-announcement period of changes in ratings by investor-paid rating agencies, which are followed by announcements of issuer-paid agencies (both FSRs and BRs), will be associated with significant CAARs. The rationale is that issuer-paid agencies are typically more conservative with changes in ratings (also because of their regulatory role), hence, when they confirm a prior change by the investor-paid agency, the market perceives this announcement as informative. Reaction to follower s changes in ratings is also consistent to Halek and Eckles (2010) reinforcement hypothesis. More formally: H4: The post-announcement period of the leading rating agency s changes in insurance ratings which are subsequently followed by changes in insurance ratings of the follower rating agency, is associated with cumulative average abnormal returns (CAARs). In contrast, we do not expect CAARs in the post-announcement period of changes in ratings that are not followed by other changes in ratings (i.e. either by issuer- or investor-paid agencies). 13

15 3. Empirical Methodology 3.1. Predictive Ordered Probit for Lead-Lag Relations We use methodology from existing literature to test the lead-lag relationship between CRAs. Specifically we use an ordered probit model similar to Güttler and Wahrenburg (2007) and Alsakka and Gwilym (2010). In their paper, Güttler and Wahrenburg (2007) examine the lead-lag relationship of changes in ratings by Moody s and S&P, of a sample of companies that are close to default. Since their analysis focuses on a significantly stressed sample of firms that eventually default, they use a binary variable that captures the time difference between the announcement of dependent variable and the time of default. Alsakka and Gwilym (2010) analyze lead-lag relationships in changes of sovereign credit ratings, which also happen less frequently than changes in US corporate bond ratings. To test for predictability (i.e. lead-lag effects) between rating agencies A and B, we use a predictive ordered probit model with dependent and independent variables comprising changes in ratings ( rating code in Table 1) and lags of those variables, respectively. Lead-lag relations are tested in both directions by running the following two models (Alsakka and Gwilym, 2010):,,, (1),,,, (2) where, (, ) is the change in rating by agency A (B) for insurance firm on month. Dependent variables in equations 1 and 2 take values of -2, -1 for downgrades of one and more than one notch respectively, and +1, +2 for upgrades of one and more than one notch, respectively. Independent variables account for changes in ratings in the previous months 1,6. The range of possible values of the independent variables is -2, -1, 0, 1, 2 and they are coded in a similar way as the dependent variables; the only addition is the value of zero, which represents the case of no change in rating. and are distributed according to a standard normal distribution. In equation 1 we test for the lead effect of over by controlling for the predictability of the dependent variable ( ) by its own lags and then searching for additional predictability by the lags of. 14

16 We run equation 2 separately to test the case of reverse predictability ( leading ). To illustrate, would lead if the coefficients in equation 1 are positive and significant but the coefficients in equation 2 are not Event Studies for Economic Value of Announcement Predictability We define an event as a day on which a change in rating is announced and use event study analysis to test for the economic value associated with such announcements (Campbell et al., 1997). Our testing period starts 45 trading days before the event day. The estimation period starts 300 trading days and ends 46 trading days before the event day. 5 Each firm s industry adjusted return is estimated using:, (3) where is the return of security on day, 300, 46, is the industry return on day, 300, 46, and is the residual, with zero-mean error. For the industry return on day, we use the value-weighted insurance market index, comprised of all actively traded firms with standard industry classification codes 6311, 6321, 6324, 6331, 6351, 6361, 6371, 6399 and 6411, we aggregate the market value of equity (including dividends) and calculate the daily return of the index. We then use the coefficients and to estimate security i s expected returns,, over the testing period, 45, 60 :. (4) Abnormal returns,, are estimated for each security as the difference between the observed return,, and the expected ex-post return over the testing period,, 45, 60 :. (5) We average abnormal returns across all downgrades every day to obtain: 4 To address the concern that some variables may be affecting simultaneously either changes in BRs, or changes in FSRs, or both, we have tried variations of our model with additional explanatory variables such as the combined ratio, short-term debt to total debt, net-to-gross premiums written, among others. We face three problems: (a) the reporting frequency of these variables is lower than monthly, (b) the timing of the announcement of such variables is uncertain, and (c) sample size decreases significantly to make the analysis meaningful. Results remain unchanged. 5 Robustness checks with different estimation periods have also been conducted with similar results. 15

17 , (6) and then calculate cumulative average abnormal returns from day to day as follows:,. (7) 4. Data & Descriptive Statistics 4.1. Issuer-paid Rating Agencies To conduct our analysis we collect in Long-term Issuer Credit Ratings to use as BRs and Insurer Financial Strength Ratings to use as FSRs for the issuer-paid rating agencies. For S&P, data from January 2000 through December of 2007 are available through SNL database; we also hand-collect changes in BRs and FSRs announced from May of 1996 until 2000 from newswires (Dow Jones Global Factiva). In the case of a group company, we identify the subsidiary with the largest Net Total Assets by year and use its FSR as the group rating. We follow the same procedures to collect data for Fitch from January 2000 until December Standard & Poor s Ratings (S&P) [Table 1; Figure 2A; 2B; 2C] The original sample of ratings by S&P comprises initial ratings, affirmations, upgrades and downgrades. The sample of FSRs comprises 3,011 observations, of which 680 are changes in ratings (500 downgrades and 180 upgrades). Similarly, the sample of BRs comprises 1,696 observations, with 220 downgrades and 132 upgrades (Figure 2A). S&P s BRs and FSRs are distributed into 22 rating categories as shown in Table 1. FSR ratings range from the highest quality of AAA (numerical value = 1) to the lowest quality of R (under regulatory supervision; numerical value = 22), where any rating above BB+ (below a numerical code of 10) is considered financially secure. BR ratings range from a high of AAA to a low of D (in default), where any rating over a BB+ rating (below a numerical code of 10) is considered investment grade. While most changes in BRs take place above the investment grade boundary, most changes in 16

18 FSRs are concentrated close to the threshold rating recommended by insurance brokers (Bradford, 2003). This trend is consistent with the fact that there is usually a positive notching effect of FSRs over BRs (Best, 2004) Fitch Ratings Turning to Fitch, the sample of FSRs comprises 1,868 observations, of which 341 are changes in ratings (upgrades and downgrades). More than 75% of changes in FSRs by Fitch are downgrades (261 vs. 80 upgrades), with most of the action taking place above the financially secure grade. The sample of BRs comprises 1,175 observations, with 144 upgrades and 128 downgrades. Most rating actions are concentrated above the investment grade. Fitch s BRs and FSRs are distributed into the same rating categories as those of S&P except that Fitch does not use the CCC+ and CCC- categories for either type of rating (Table 1; Figure 2B) Investor-paid Rating Agency: Egan Jones Ratings (EJR) Our sample begins with the first rating assignment to an insurance firm (August 1997) and ends in December Data for the period and were generously provided by the authors of Beaver et al. (2006) and Egan Jones Ratings, respectively. EJR rates publicly rated firms and thus fewer companies than Fitch and S&P, which also rate private firms. We parse through a large sample of companies and ratings to identify insurance firms or firms with sufficiently large insurance operations. The original sample comprises 601 observations with 271 changes in BRs. The rating scale used by EJR is identical to the scale used by S&P for BRs (Table 1). We observe a larger number of downgrades than upgrades in BRs (161 vs. 110) by EJR, with more rating changes (especially upgrades) taking place above the investment grade (Figure 2C) Matched Samples Table 2 reports descriptive statistics for our sample, which include firms rated by both investorand issuer-paid rating agencies. To maximize coverage we use the intersection of EJR and S&P from August 1997 to December 2007 for the companies that have a BR from EJR, a BR from S&P and an FSR by S&P. The sample comprises a maximum (minimum) of 460 (258) firm-year observations. The average 17

19 (median) firm-year has total asset value of $71,053 ($22,688) million, net premiums written of $7,406 ($3,008) million and net-to-gross premiums written of approximately 87% (88%). Average (median) short-term debt is $1,169 ($34) million, or 20.02% (12.20%) of the company s total debt. [Table 2] 5. Results 5.1. Lead-Lag (Predictability) Relations [Table 3] In our lead-lag analyses, we examine six pairs of ratings: (a) EJR Changes in BRs vs. S&P Changes in BRs; (b) EJR Changes in BRs vs. Fitch Changes in BRs; (c) S&P Changes in BRs vs. S&P Changes in FSRs; (d) Fitch Changes in BRs vs. Fitch Changes in FSRs; (e) EJR Changes in BRs vs. S&P Changes in FSRs; and (f) EJR Changes in BRs vs. Fitch Changes in FSRs. When matching pairs for each of the previous six combinations, we observe that issuer-paid agencies (i.e. combinations (c) and (d)) often seem to change their BRs and FSRs on the same day. Same-day changes in BRs and FSRs are not included in our analysis. 6 The remaining four combinations of ratings across agencies (i.e. (a), (b), (e) and (f)) have negligible changes in ratings happening on the same day. Table 3 shows the distribution of changes in ratings for the six matched samples of the pairs listed above (changes happening on the same day are not included). Looking across all pairs, we find that downgrades of more than one notch constitute a significant number of the downgrades in each sample. We also find that the matching process significantly reduces the number of upgrades in each sample. For each of the six pairs we run equations 1 and 2 and report results in the twelve panels of Table 4. [Table 4, 5, 6] Results of our preliminary hypothesis (H0) are given in Panels A-D of Table 4. Panels A and B show the results of the lead-lag relation between changes in BRs by EJR and changes in BRs by S&P. Positive (negative) coefficients in lagged variables make changes in the same direction as the dependent variable more (less) likely. We find some evidence of predictability from EJR s changes in BRs to S&P s 6 By construction, the first lag of our analysis starts one day before and ends thirty days before the announcement. 18

20 changes in BRs. In particular, we find no positive and significant coefficients on S&P s changes in BRs (panel B) while the second lag of EJR is positive and significant (p<0.07). 7 The evidence is stronger when we examine the lead-lag relation between changes in BRs by EJR and changes in BRs by Fitch (Panels C and D). The 2 nd, 5 th and 6 th lag of EJR s changes in BRs have positive and significant coefficients in Panel C, and no coefficients on Fitch s changes in BRs are positive and significant in Panel D. We therefore find evidence in support of our preliminary hypothesis (H0): changes in BRs by the investor-paid rating agency lead those of the issuer-paid rating agencies. In Table 5, we test the lead-lag relation between changes in BRs and changes in FSRs of the issuer-paid rating agencies. In the case of S&P, we find evidence of predictability from changes in FSRs to changes in BRs: the coefficients on the 1 st, 3 rd and 4 th lags of changes in FSRs are positive and significant in Panel B, while similar results do not obtain in Panel A. In the case of Fitch, in contrast, the 1 st and 3 rd lags of changes in BRs predict changes in FSRs but the 5 th lag makes them less likely (Panel C) and the 4 th lag of changes in FSRs predicts changes in BRs (Panel D). Based on these results we conclude that information flows in both directions between BRs and FSRs of the same issuer-paid rating agency, even though changes in FSRs appear to lead in the case of S&P. Results are similar for a longer time period as shown in the robustness section (section 5.4.1). Given that investor-paid rating agencies lead issuer-paid rating agencies (H0) and information flows both ways between changes in FSRs and changes in BRs, it is natural to ask whether changes in BRs by the investor-paid rating agency lead changes in FSRs by the issuer-paid rating agencies. We find evidence that EJR s changes in BRs lead S&P s changes in FSRs (Table 6). Specifically, the 4 th lag of EJR s changes in BRs in Panel A is positive and significant (p-value<0.03) whereas in Panel B none of the lags of S&P s changes in FSRs is positive and significant. The results are similar when we test the lead-lag relation between EJR s changes in BRs and Fitch s changes in FSRs. In this case, the 2 nd lag of EJR s changes in positive and significant, whereas no predictability is evident in the opposite direction (Panel D). 7 The first lag of EJR is positive with a p-value of

21 5.2. Stock Reactions to Changes in BRs and FSRs and Economic Value of Predictability Having showing evidence of lead effects of investor-paid changes in BRs over issuer-paid changes in BRs and changes in FSRs, we next examine how the market reacts to changes in BRs and also test for any economic value associated with the leading agency s announcements, that could potentially be realized by investors following such announcements. To be included in the sample, firms must have daily stock return data in CRSP database eighteen months before and three months after the change in rating. In Table 7 we show market reactions for several event windows of changes in BRs by EJR, S&P and Fitch, to test hypotheses 1-3. In the appendix (Table A2) we show the same results for issuer-paid FSRs, which are consistent with earlier findings in the literature (Halek and Eckles, 2010). To test hypothesis 4, in Table 8 we split the post-event event windows for the issuer-paid s (leading agency s) downgrades, into those that are, and those that are not followed by other downgrades. [Table 7] In Table 7 (Panel A; first six rows) we show results for downgrades in BRs by EJR, which are matched with S&P s downgrades in BRs to have the largest possible intersection (n=107; results do not change when matched with Fitch). Results reported for the issuer-paid rating agencies are matched with EJR (for S&P: n=54; for Fitch: n=56). We find extensive evidence in support of hypothesis H1a: there are negative CAARs associated with downgrades in bond ratings. Focusing on the windows centered on the event day, CAARs for EJR range from -4.55% to % for the [0,0] to [-30,+30] respectively. CAARs for S&P range from -2.26% to -9.79% for event windows [0,0] to [-30,+30]; Fitch s range from -1.73% to %, respectively. Results for all ratings agencies have p-values less than In Panel B (first six rows), we find evidence in support of hypothesis H1b for EJR and S&P: there are positive CAARs associated with upgrades in bond ratings. CAARs for EJR range from 1.25% (pvalue < 0.01) to 1.55% (p-value<0.05) for the [0,0] to [-10,+10] respectively. For S&P there is no reaction at the event and there are marginal reactions in the windows [-1,+1], [-2,+2], [-3,+3] and [-10,+10]; 1.23% (p-value<0.01), 0.38% (p-value<0.10), 0.51% (p-value<0.10) and 0.85% (p-value<0.10), respectively. The market does not seem to value Fitch s upgrade announcements as informative as there 20

22 are no significant results for Fitch. To test for the existence of asymmetry in CAARs from upgrades and downgrades within the same rating agency (H2), we run t-tests on the absolute values of the estimated CAARs in panels A and B in Table 7 (first six rows). For EJR, we find differences of 3.66%, 5.26%, 5.78%, 7.05%, 10.57% and 14.93% (all p-values<0.01) for windows [0,0], [-1,+1], [-2,+2], [-3,+3], [-10,+10] and [-30,+30], respectively (untabulated results). Similar results obtain for S&P (Fitch), with mean differences of 2.26% (2.49%), 5.09% (4.94%), 5.60% (6.48%), 6.61% (6.58%), 9.91% (9.00%) and (11.33%) and maximum p-value < 0.05 (p-value < 0.01). These results provide evidence in support of H2. Results for upgrades are of smaller statistical and economic significance, consistent with the results of Holthausen and Leftwich (1986), Hand et al. (1992), Ederington and Goh (1998) and Halek and Eckles (2010). Next, we test for differences in market reactions to changes in bond ratings between investor-paid and issuer-paid rating agencies (H3a and H3b). Results are reported in Table 7 and specifically in the columns showing mean differences (MD) of investor-paid CAR minus issuer-paid CAR for each event window reported. We find evidence in support of both hypotheses: market reactions are larger for the investor-paid agency than those of S&P and Fitch. In the case of downgrades (Panel A), we find a negative mean difference between EJR s event day CAARs over the respective CAARs for S&P and Fitch. In the case of S&P the market reaction is 2.29% (p-value < 0.01) and in the case of Fitch it is 2.82% (p-value < 0.01) lower than EJR s. In the case of upgrades (Panel B), we find positive mean differences between the two types of rating agencies, with EJR having 0.99% (p-value < 0.05) and 1.20% (p-value < 0.01) larger (more positive) CAARs than S&P and Fitch, respectively. Differences also exist in broader event windows centered at the event: for S&P s downgrades there exists a difference of -6.98% (p-value < 0.10) at the [-30,+30] and for Fitch -0.59% (pvalue < 0.05) at [-3,+3]. For S&P s upgrades the difference is 1.37% (p-value < 0.10) at the [-2,+2], and for Fitch s upgrades the difference peaks at 2.39% (p-value < 0.01) at [-3,+3]. These results suggest that investor-paid agencies downgrades and upgrades in BRs are associated with stronger market reactions 21

23 than those of issuer-paid agencies, thus providing support for hypotheses H3a and H3b respectively. 8 In un-tabulated results, for each of the three rating agencies, we also test if the actual announcement of a change in BR is associated with larger abnormal returns (in absolute terms) than abnormal returns of each of the three previous days as shown in Table 7. For the investor-paid agency s downgrades, we find that abnormal returns associated with the [0,0] window are larger (more negative) than those of the [-3,-3] window (mean difference = -4.63%; p-value < 0.01), the [-2,-2] window (mean difference = -3.43%; p-value < 0.01) and the [-1,-1] window (mean difference = -2.96%; p-value < 0.01). For the issuer-paid rating agencies the only significant difference is between the [0,0] window and [-1,-1] window for Fitch, with a mean difference of -1.36% (p-value < 0.05). 9 This jump in event day returns suggests that EJR exerts effort (at least more effort than issuer-paid agencies) to communicate their analysis of publicly available information to the market quickly, in which case either their downgrades cause breaks in abnormal returns or their announcements coincide with a publicly available event that causes a sharp decrease in stock returns Economic Value of Predictability (H4) [Table 8] Given the evidence of predictability from investor-paid changes in BRs to issuer-paid changes in FSRs and changes in BRs, we then test if this predictability is linked to any abnormal market returns. Since predictability is stronger for downgrades, and also given the asymmetry between market reactions to upgrades and downgrades, we focus on downgrades for our analysis. In Table 7 we also show event windows before, at and after changes in BRs by both types of agencies. In all cases, we observe that the period before downgrades in BRs is associated with large CAARs (all p-values < 0.01). Specifically, for EJR s downgrades in BRs (panel A) there is a CAAR of -8.31%, -6.36% and -4.31% respectively for the [-30,-1], [-20,-1] and [-10,-1], respectively. For S&P, the respective CAARs are -8.67%, -5.76% and - 8 Similar mean differences exist in stock market reactions between investor-paid changes in BRs and issuer-paid differences in FSRs, as shown in Table A2 in the appendix. 9 Paired t-tests for upgrades show a mean difference of no less than 1.10% (p-value < 0.05) for EJR for all three combinations. For issuer-paid rating agencies, only one test is significant for S&P: the difference between the [0,0] and [-2,-2] windows, with a mean difference of 0.98% (p-value < 0.01). 22