Corporate governance ratings and firm performance.

We examine the corporate governance ratings provided by three premier US rating agencies and find that summary scores are generally poor predictors of primary and secondary measures of future firm performance. However, some component sub-ratings that focus on the eight key dimensions of dynamic governance structures provide more positive and reliable evidence of their information content in predicting the multiple dimensions of firm performance. These results reflect the recent observations by academic researchers and money managers that it is extremely difficult to distill all of the complex governance mechanisms into a single integrated, yet informative overall score.

**********

In the wake of recent corporate scandals and the resulting focus on firms' corporate governance mechanisms, many money managers have been pressured by their clients to integrate corporate governance quality in their portfolio management processes. However, the analysis of how corporate governance can affect firm performance is complicated by the fact that there are very few reliable metrics of corporate governance.

There have been many attempts by rating services to quantify the quality of corporate governance in the form of commercially available ratings. The premier rating agencies in the United States are The Corporate Library (TCL), Institutional Shareholder Services (ISS), Governance Metrics International (GMI), and Standard & Poor's (S&P). These agencies use a proprietary scoring method to assess the soundness of governance practices at public corporations. Large investor groups are the typical customers of these governance rating agencies. Further, since the recent corporate scandals, these governance scores have become increasingly influential and popular among retail investors, client firms, and regulators. But surprisingly, there is little systematic study of the value of these third-party governance ratings in assessing firm performance.

Analyzing these ratings is important for several reasons. First, these rating services claim to use comprehensive governance databases and sophisticated analyses to construct their scores. For instance, TCL asserts, "Our TCL Ratings employ quantitative formulae and methodology that go far above and beyond the industry's typical reliance on "check-the-box" best practices compliance. Our proprietary methodology is based on a variety of governance practices statistically proven to be significant measures of value and risk" (see http://www.thecorporatelibrary.com). If these ratings are indeed useful in assessing the quality of corporate governance, then they might add significant social value by improving corporate governance practices through better measurement and transparency. The fact that these ratings are independent further strengthens the latter argument.

Second, notwithstanding their expert analysis, rating services all use raw data from publicly available documents, such as Securities and Exchange Commission (SEC) filings, tax filings, annual reports, and press releases. This reliance raises questions about the information content of governance ratings supplied by the vendors. Any evidence that they can help construct trading rules to generate systematic profits would seem to question the semi-strong form of the efficient markets hypothesis. Given that large, sophisticated institutions are willing to pay for these ratings, it would be valuable to examine the link between these governance analyst ratings and future firm performance.

Third, the prominence of these ratings has increased considerably over recent years. Frequently, these rating vendors conduct in-house research and offer their expert opinions and recommendations on current governance-related issues. For example, shareholders in prominent firms such as General Electric and Altria (parent company of Kraft Foods and Philip Morris) have used their respective firms' low TCL corporate governance ratings to propose separating the chairmanship of the board from the CEO position. (See proposal 1 in the 2006 Altria proxy statement and proposal 4 in the 2006 GE proxy statement. In both cases, the board of directors has recommended a vote against the proposal.) Fourth, there are instances in which policy makers draw on the research conducted by the rating services to formulate public policy. Finally, many academics and market managers are skeptical about the value of these ratings. The media indicate that governance scores "uncritically staple together every governance reform dimension regardless of the existence of research support" (Wall Street Journal, 2003). Such skeptics also question the link between governance scores and firm performance, giving examples of companies that perform well and yet have low governance scores (e.g., Dell, Southwest, and Wal-Mart). Indeed, there are cases in which even different rating agencies may not agree about the quality of corporate governance in a given firm. For example, Google got the lowest corporate governance rating (compared to all the S&P 500 firms) from ISS in 2004. Its rating was primarily attributed to its dual-share structure. But TCL analysts disagreed, saying "we are not going to give them a D or an F just because it is different" (San Francisco Chronicle, 2004).

Despite their increasing role in the public debate on governance, there is little systematic analysis of the relation between these governance ratings and firm performance. This gap motivates us to test whether TCL, ISS, and GMI ratings really can predict future firm performance. We use several measures of firm performance, which we can broadly classify as primary and secondary performance measures. Our primary performance measures are summary measures that cover operating performance and stock returns. Our secondary performance measures are related to four corporate events: the propensity to delist, the likelihood of bankruptcy, the exposure to class action lawsuits, and the exposure to accounting-related SEC enforcement actions. Moreover, we use two sets of corporate governance ratings, the overall ratings of the three vendors and the eight TCL sub-ratings, which cover board structure, executive compensation, accounting manipulation, etc. Our objective is to find out if the summary and sub-ratings can predict firm performance at both the primary and secondary levels. We examine the links between the overall ratings and future measures of primary and secondary firm performance, and those between the sub-ratings and future primary and secondary performance measures.

Focusing on the overall ratings, we find that the overall TCL and ISS ratings are negatively associated with the future operating performance of rated companies, but GMI ratings are positively correlated. For the link between overall ratings and secondary performance measures, we uncover weak evidence that firms with the lowest aggregate TCL rating (F) have a higher probability of facing accounting-related SEC actions, and firms with lower overall GMI ratings face a higher probability of litigation and delisting. These results provide only weak support for the argument that the aggregate scores can identify firms with governance-related difficulties.

When we evaluate the role of the TCL sub-ratings that cover eight different areas of governance in predicting multiple dimensions of firm performance, our results suggest that the accounting-related TCL sub-ratings, which assess the quality of reported earnings, reliably predict future operating performance. For the other outcome measures, we find that firms rated as having fewer litigation and regulatory problems and positive analyst adjustments have a lower probability of facing class action lawsuits, delisting, and bankruptcy. Firms with higher scores on litigation concerns, which imply lower litigation risk, also have a lower probability of SEC enforcement actions. Firms with higher scores on shareholder responsiveness have lower probability of facing bankruptcy and class action lawsuits, but they are more susceptible to SEC enforcement actions. Further, our results indicate that firms with higher scores on shareholder responsiveness and takeover defenses have a lower probability of delisting in our sample period. However, several sub-ratings have unexpected signs for the outcome variables.

Finally, as Core, Guay, and Rusticus (2006) note, if investors expect that weak governance causes or results in poor operating performance, such expectations would be quickly impounded in market prices and we would not expect any association between weak governance and future stock returns. But if investors are surprised by the poor operating performance of weak-governance firms, then we would expect weak governance to be associated with poor subsequent stock returns.

To investigate the predictive power of the corporate governance ratings, we construct portfolios based on the categories and quartiles of the summary governance ratings and sub-ratings. We also construct a hedge portfolio that is long (short) on stocks with strong (weak) overall governance scores and sub-scores. When we regress returns on the quartile and hedge portfolios against the Fama and French (1995) three factors and the momentum factor of Carhart (1997), we find that the average abnormal returns for portfolios created based on the TCL, ISS, and GMI summary ratings are rarely significantly different from zero. But when we repeat these tests using the sub-ratings, we notice that trading strategies based on several TCL sub-ratings exhibit weakly positive abnormal returns, especially for small firms. Thus, our analysis suggests that in assessing firms' future stock market performance, the separate, focused ratings of the several dimensions are more informative than is a single summary score of the complex governance mechanisms.

Overall, our results suggest that several sub-ratings of governance (i.e., micro measures) perform much better in predicting multiple dimensions of firm performance, but other sub-ratings do not. The overall scores (i.e., macro measures), which cover both the poor-performing and better-performing sub-ratings, typically have poor explanatory power. These findings shed light on the question of how we can measure corporate governance: "Can a complex structure of governance be reduced to a summary statistic? How does one go about identifying which of the myriad governance mechanisms should be included?" (Financial Times, 2005).

The paper is organized as follows. In Section I, we discuss the literature on corporate governance and performance and provide information on the history and methods of corporate governance rating vendors. Section II explains our data and presents the summary statistics. We discuss the empirical results in Section III. Section IV concludes.

I. Literature on Governance and Governance Ratings

Most of the related empirical literature examines the relation between firm performance and some subset of many dimensions of governance, such as insider ownership, board composition, board size, executive compensation, and antitakeover provisions (see, e.g., Morck, Shleifer, and Vishny, 1988; McConnell and Servaes, 1990; Hermalin and Weisbach, 1991; Lee, Rosenstein, Rangan, and Davidson, 1992; Yermack, 1996; Core, Holthausen, and Larcker, 1999; Bhagat and Black, 2002). Several recent papers examine the relation between firm performance and a composite measure of corporate governance. Gompers, Ishii, and Metrick (2003) construct a governance index (G-index) using information on several distinct takeover defense provisions, provided by the Investor Responsibility Research Center (IRRC). These authors find that firms with weaker shareholder rights have lower stock returns and firm valuations. Examining the relation between the G-index and company operating performance over 1991-1999, Core et al. (2006) find some evidence that the G-index is linked to future operating performance. Further, scrutinizing the association between weak governance and weak stock returns, they report the relation between the G-index and stock returns disappears after 1999. Bebchuk, Cohen, and Farrell (2009) use IRRC data to show that a six-factor firm entrenchment index fully drives the correlation between the G-index, firm value, and stock returns during the 1990s. Chi (2005) examines the endogeneity between firm value and shareholder rights and finds that the change in G-index is negatively correlated with the future change in firm's Tobin's Q.

Brown and Caylor (2006a, 2006b) construct a governance score using ISS governance factors and find that firms with lower governance scores have higher return on equity, higher profit margins, and higher firm valuations. Using principal components analysis, Larcker, Richardson, and Tuna (2007) construct 14 governance factors and find that these are related to future operating performance and stock returns. But they report weak results for abnormal accruals and accounting restatements.

We briefly summarize the history of the three governance rating vendors and their methods. Details of these ratings are available on the vendors' websites: www.thecorporatelibrary.com, www.isscgq.com, and www.gmiratings.com. TCL was founded in 1999. Its goal is to provide to its clients information on and analysis of corporate governance, executive, and director compensation. In 2003, it launched its Board Effectiveness Ratings, now known as TCL Ratings. These ratings cover four key components. The corresponding weights in the overall TCL rating are shown in parentheses: board structure (50%), CEO compensation (30%), takeover defenses (10%), and accounting, that is, screens for earnings management (10%). The ratings cover five categories: A, B, C, D, and F. Rating A is given to companies with the most effective boards, and F is assigned to those with the least effective boards. TCL also provides the following eight sub-ratings on various aspects of corporate governance: board composition, CEO compensation, shareholder responsiveness, litigation and regulatory problems, takeover defenses, accounting, strategic decision making, and analyst adjustment. These sub-ratings also range from A to E Brief descriptions of these sub-ratings are provided in Table I. TCL claims that major insurance companies, institutional investors, law firms, investment banks, universities, public corporations, compensation consultants, and executive search firms are among its clients.

Founded in 1985, ISS provides research and proxy advisory services to institutional investors. In 2002, ISS introduced its corporate governance ratings (called the Corporate Governance Quotient, CGQ[R]) that aim to "measure, the strengths, deficiencies and overall quality of a company's corporate governance practices and board of directors." ISS considers the following factors: board of directors (structure, size, attendance, chairman/CEO separation, etc.), audit committees, board, and committee independence, and the presence of a financial expert, charter and bylaw provisions, antitakeover provisions, executive and director compensation, qualitative factors, such as board performance reviews, meetings of outside directors, CEO succession plan, director and executive ownership, and director education. CGQs range from 0 to 100 and measure the quality of governance relative to other firms in the company's primary stock market index.

GMI was founded in 2000 to provide the tools that make it possible for its clients to monitor firms' corporate governance. Similar to TCL and ISS, GMI considers various aspects of governance, such as board accountability, financial disclosure and internal controls, shareholder rights, executive compensation, market for control and ownership base, and corporate behavior. Companies are scored on a scale of 1 (lowest) to 10 (highest). GMI global ratings measure the strength of corporate governance relative to all other companies in the GMI universe. GMI also gives companies a home market rating that reflects the strength of each one's governance mechanisms relative to others in their home country. However, these ratings are not available from Bloomberg (our main data source), which provides data only on overall global rating.

II. Data

To assess the ability of ratings in year t to forecast firm performance in year t + l, we start with 4,820 firm-year observations on the TCL overall ratings available for 2003, 2004, 2005, and the first seven months of 2006. Restricting the data set to firms that have Compustat data for both year t and year t + 1 reduces the number of observations to 4,546. We obtain the ratings for 2003 from TCL and manually collect the ratings for 2004, 2005, and 2006 from Bloomberg. In addition, we use Bloomberg to collect the TCL sub-ratings for the most recent year for which the TCL overall rating is available. For example, if the most recent available TCL overall rating is for 2005, we use the TCL sub-rating data for that year. We also keep the 2006 data for the TCL overall ratings to maximize the number of observations we have for the sub-rating probit regressions.

In addition, we manually collect GMI and ISS ratings from Bloomberg for 2005, the earliest year for which they are available. Our data set has 1,618 firms with ISS and year t Compustat data, and 1,487 firms with GMI and year t Compustat data. Table II presents summary statistics for the TCL and its sub-ratings, ISS, and GMI ratings. Panel A of Table II shows, about 9% of the rated firms receive the highest board effectiveness rating of A from TCL, 38% B, 35% C, and 16% D, and only 2% receive F. Clearly, the ratings show a considerable amount of cross-sectional variation. Although some of the sub-rating distributions are similar to the overall rating distribution, some are quite different. For example, 91% of the firms are rated B for shareholder responsiveness and 94% are rated B for strategic decision making. From Panel B, the ISS ratings range from 0.5 to 100 with a median of 54.45, while the GMI ratings vary from 2.5 to 10 with a median of 7.

Panel C of Table II presents pairwise correlations among ISS, GMI, TCL and its sub-ratings. The correlation between the TCL overall rating and ISS rating is about 19%, but that between the TCL and GMI rating is only 0.3%. ISS and GMI ratings are correlated with each other at 13%. Consistent with the weights assigned to sub-ratings, the TCL overall rating is highly correlated with Board Composition and CEO Compensation sub-ratings. Another sub-rating that is highly correlated with the overall rating is Analyst Adjustment, at 47%. Sizable correlations seem to occur among the following TCL sub-ratings: Accounting, Litigation Problems, Analyst Adjustment, and Strategic Decision Making. The low correlations among the three summary scores imply that these ratings focus on different dimensions of governance. They underscore the considerable disagreement over which governance attributes are most important and what is considered good versus bad governance. Further, they highlight the difficulty of aggregating disparate governance attributes into a single summary score.

To investigate whether the governance analyst ratings can predict litigation and fraud risk, we obtain information on class action lawsuits and accounting-related SEC actions. To identify firms facing class action lawsuits, we utilize the Stanford University Securities Class Action Clearinghouse. This database provides the details of prosecution, defense, and settlement of federal class action securities fraud litigation.

To find data on accounting-related SEC actions, we search the SEC Accounting and Auditing Enforcement Releases (AAERs) from 2003 to July 2006. Some of the reasons for AAERs are fraudulent earnings manipulation schemes, improper revenue recognition, improperly inflated earnings, and false and misleading financial statements. AAERs provide summaries of the SEC's accounting-based enforcement actions and describe the SEC's investigations.

To test whether the governance ratings can predict delisting and bankruptcy probabilities, we obtain the list of firms that filed for bankruptcy between January 2003 and July 2006 from Bloomberg and gather nonmerger related delisting data from the Center for Research in Security Prices (CRSP).

III. Results

In this section, we present and discuss our empirical results on the power of the overall and sub-ratings in predicting operating performance, stock returns, and other secondary measures of firm performance.

A. Operating Performance

To assess how the strength of governance mechanisms affects operating performance, we follow Core et al. (2006) and regress measures of future operating performance on governance ratings and control variables taken from a year earlier:

Operating [Performance.sub.i,t+1] = [a.sub.0] + [a.sub.1] * Governance [Rating.sub.i,t] + [a.sub.2] * ln [(Firm Size).sub.i,t] + [a.sub.3] * Industry [Dummies.sub.i,t] + [a.sub.4] * Year [Dummies.sub.i,t] + [[epsilon].sub.i,t]. (1)

We use future operating performance rather than contemporaneous performance for two reasons. First, endogeneity can be an issue. Not only can corporate governance affect performance, but firm performance can also cause the firm to change its governance structures (Hermalin and Weisbach, 1988; Weisbach, 1988). Lagging the governance variables partially addresses the endogeneity problem. A second reason for using future performance is to avoid the look-ahead bias. Thus, we pay special attention to the fiscal year-ends and the rating announcement dates. We use operating performance measures for the year after the rating is announced. For example, if the rating is announced in August 2004 and the fiscal year-end for the company is May, we use the operating performance data for May 2005 in our regressions.

We measure operating performance as operating income after depreciation scaled by the book value of total assets. As Core et al. (2006) note, depreciation expense is an important component of a firm's performance to the extent that governance affects firm performance through capital expenditure programs. Accordingly, we use operating income after depreciation (Compustat data item 178) to calculate return on assets. We also control for firm size (natural logarithm of book value of assets) and industry differences by using the Fama-French (1997) industry indicator variables in the regressions. Table III presents the results of operating performance regressions for TCL, ISS, and GMI ratings.

We use two specifications for TCL ratings. The first specification includes the ordinal measure, TCL Overall Rating, which takes the value of five, four, three, two, or one if the TCL rating is A, B, C, D, or F, respectively. The second specification includes dummy variables for each rating category (C is the base case). All models include the Fama-French (1997) industry indicator variables, but we do not report their coefficients. Models with TCL Overall Rating and rating dummies also include year dummies. We correct standard errors for heteroskedasticity and clustering of observations at the firm level.

We note that the TCL Overall Rating is negatively related to future operating performance. To check this finding, we use dummy variables for each rating category (to break up the TCL overall Rating). Although the estimates on the four rating dummies are mostly negative, we find that only the B-rated firms have significantly lower future operating performance compared to that of C-rated firms. Moreover, the ISS rating is also negatively related to future operating performance, but the coefficient estimate for the GMI rating is positive and significant. These findings are contrast with the findings of Larcker, Richardson, and Tuna (Financial Times, 2005). These authors report a positive relation between the TCL ratings, as measured by a single indicator variable representing good versus bad governance, and subsequent operating performance over 2002-2004. When we use data over a longer period, along with the ordinal measure of ratings, we find the opposite results. For our sample period, we find that firms that receive higher grades from two of the three premier commercial governance rating agencies exhibit lower future operating performance. Only firms with higher GMI ratings have higher future operating performance.

Table IV presents operating performance regression results for the eight TCL sub-ratings. Of these ordinal measures, only the Accounting sub-rating, which screens for possible earnings management and related questionable accounting practices, has a highly significant positive relation with future operating performance, notwithstanding the unexpected negative relation between the TCL Overall Rating and firm performance. This finding implies that firms with higher Accounting sub-ratings of A, B, and C have superior future operating performance than do their counterparts with D and F ratings. In evaluating the accounting practices used by the rated firms, TCL analysts compare current quarter reports against the prior four quarters for indicators of potential earnings management. TCL warns that firms with low scores may be using accounting methods that favor current period earnings, but that these methods may also result in lower long-term (three- to five-year) market performance. Our finding is consistent with media reports that have given credit to TCL for its ability to predict several of the recent corporate scandals.

We also use indicator variables for the four levels of each sub-rating (C is the base case) in separate specifications to minimize the potential collinearity problem. In untabulated results, we find that firms with the Litigation Problem sub-rating of B and, to a lesser extent, those with the Shareholder Responsiveness (i.e., shareholder-friendly boards) sub-rating of A have higher operating performance. However, firms with Board Composition sub-rating of F (i.e., those with the least effective boards) and Litigation Problem sub-rating of F (i.e., higher incidence of litigation or regulatory investigations) have better operating performance.

B. Probability of Litigation, SEC Enforcement Action, Delisting, and Bankruptcy

TCL claims, "Our unique perspectives on public companies can help identify investments with upside opportunity, "red flag" investments posing downside risks" (www.boardanalyst.com). In promotional materials targeted to the insurance markets, it asserts that their ratings are "a key component for insurers to accurately assess risk in pricing corporate insurance policies and to gain insight into securities class action litigation risk." Moreover, TCL claims that it can "flag companies with practices that present reputational risk and was able to successfully identify Enron, Global Crossing, Worldcom, and Adelphia as problematic at a time when they were still beyond reproach by most analysts." These statements suggest that one of the goals of these governance ratings is to identify firms that might face governance-related difficulties in the future.

To test whether these ratings can predict firms with significant downside risk and potential governance-related difficulties, we identify firms in our sample that experienced securities class action lawsuits, accounting-related SEC actions, delistings, and bankruptcies. We use probit regressions to test the ability of the ratings to predict these events:

[Event.sub.i,t+1] = [a.sub.0] + [a.sub.1] * Governance [Rating.sub.i,t] + [a.sub.2] * [Controls.sub.i,t]

+ [a.sub.3] * Industry [Dummies.sub.i,t] + [a.sub.4] * Year [Dummies.sub.i,t] + [[epsilon].sub.i,t]. (2)

The dependent variable Event denotes one of the four corporate events of litigation, action, delisting, or bankruptcy. Litigation is a dummy variable that equals one if there is securities class action lawsuit filed against the rated firm after the release date of the rating, and zero otherwise. Action is a dummy variable that equals one if there is an accounting-related SEC enforcement action against the firm after the release date of the rating, and zero otherwise. Delisting is a dummy variable that equals one if the company is delisted (nonmerger related delistings) after the release date of the rating, and zero otherwise. Bankruptcy is a dummy variable that equals one if the company files for bankruptcy after the release date of the rating, and zero otherwise. Here again we pay special attention to the dates of the SEC enforcement actions, bankruptcy, etc., and the rating announcement date. For example, if the company goes bankrupt in May 2004 and the TCL rating is announced in August 2004, then we assign the bankruptcy dummy for 2004 a zero value.

For our probit regressions with the litigation and action dummies as the dependent variables, we use operating income scaled by total assets (ROA), market-to-book ratio (Q), and firm size (Size) as control variables. For regressions in which the delisting and bankruptcy dummies are the dependent variables, we also include leverage (book value of debt/book value of assets) as a control variable. We measure all control variables for the fiscal year before the event year. We find that many of their coefficient estimates are significant at or under the 5% level. All regressions include the industry dummies. Models with TCL Overall Rating, rating dummies, and TCL sub-ratings also include year dummies. To conserve space, we have suppressed the regression estimates for the control variables.

Table V presents results of our probit regressions for the TCL, ISS, and GMI ratings. In the first row of Panel A, we note that the TCL Overall Rating has little power to predict the probability of litigation, SEC action, delisting, or bankruptcy of rated firms. The only significant result is that firms with the F rating have a higher probability of SEC enforcement action. This finding supports TCL's claim about identifying companies with practices that represent reputational risks. The corresponding marginal probability is 0.02 (untabulated), indicating that the F-rated firms are 2% more likely to face the accounting-related SEC enforcement actions. When a dummy variable predicts the dependent variable perfectly, we drop it from the regression to increase the numerical stability of the optimization process. For instance, we drop the F rating from regressions with the Delisting and Bankruptcy as the dependent variables. In both cases, the F rating predicts the failure (i.e., not delisted) perfectly.

In Panel B, the ISS ratings are unrelated to the four events, but the GMI ratings show a weak negative association with both the litigation and delisting probabilities. The marginal probability for GMI rating in the litigation probit regressions is 0.58%, but the marginal probability in delisting regressions is very close to zero.

In Table VI we use probit regressions to examine the ability of the eight TCL sub-ratings to predict the probability of the four corporate events. As before, the ordinal measure, Sub-rating, takes the value of five, four, three, two, or one if it is equal to A, B, C, D, or F, respectively. These regressions include the same set of control variables as in the preceding analysis, but for brevity we do not report their estimates. Moreover, we compute marginal probabilities associated with the higher ratings of A or B to facilitate the interpretation of the regression coefficients, but again, we do not report the details for brevity. The results show that Shareholder Responsiveness, Litigation Problem, and Analyst Adjustment are significantly and negatively related to the probability of class action lawsuits. In other words, firms that TCL judges as having more shareholder-friendly boards, lower litigation risk, and favorable analyst adjustment scores face a lower risk of class action lawsuits. As an example, the marginal probability associated with Shareholder Responsiveness dummy is -0.0076 (untabulated), indicating that over our sample period, firms with the ratings of A or B are 0.76% less likely to face litigation. Similarly, the marginal probability associated with Litigation Problem dummy is 0.0086, indicating that firms with the ratings of A or B are 0.86% less likely to face litigation. Further, our untabulated marginal probability results indicate that firms with the analyst adjustment rating of B (about 80% of the sample has a rating of B) are 1% less likely to face litigation than are firms with the analyst adjustment rating of D.

A surprising result is that the Accounting sub-rating, which assesses earnings quality, is positively associated with the litigation probability. The second column of Table VI displays results for the SEC enforcement Action dummy outcome variable. Similar to the class action lawsuit results, firms with higher scores on Litigation Problem face a lower probability of SEC enforcement action. The marginal probability associated with the Litigation Problem dummy (untabulated) is -0.024, indicating that firms with the Litigation Problem rating of A or B are 2.4% less likely to face a SEC action over our sample period. We note especially that the Shareholder Responsiveness and Strategic Decision Making sub-ratings are positively related to the probability of SEC action.

In the Delisting column, Shareholder Responsiveness, Litigation Problem, Takeover Defenses, and Analyst Adjustment sub-ratings are negatively correlated with the probability of delisting, and Accounting and Strategic Decision Making ratings are positively correlated. The marginal probabilities for these variables are low (close to zero). In the Bankruptcy column, we see that sub-ratings on Shareholder Responsiveness, Litigation Problem, and Analyst Adjustment are negatively related to the bankruptcy probability of rated firms. Similar to the delisting probit regression results, we find that Accounting and Strategic Decision Making ratings are positively correlated with the probability of bankruptcy. However, these ratings have low marginal probabilities.

In our results, we find it remarkable that the Litigation Problem sub-rating is significantly and negatively linked not only to Litigation (the securities class action lawsuit dummy) but also to the remaining three corporate events. In constructing this rating, TCL evaluates the amount of disclosure of current and potential liability exposure covering such factors as excessive CEO pay, lack of board independence, firm size, industry-specific risk, institutional ownership, and share trading volume, and estimates the probability of the occurrence of securities class action lawsuits within the next 18 months. Further, TCL considers whether the firm has a history of repeated regulatory infractions and assessed fines. This practice raises the possibility that the observed predictive power of the Litigation Problem rating is due to the fact that this rating includes the start of lawsuits that extend over multiple years to settle and so is mechanically correlated with Litigation. In our analysis, this possibility appears to be small, since we set the Litigation dummy equal to one if there is a class action lawsuit against the firm that has been filed aider the release date of the rating, and zero otherwise. However, without additional data over longer time horizons and analysis, it is difficult to firmly establish the ability of this and the other TCL sub-ratings to predict the four corporate events.

We note that if we allow the coefficient on the sub-ratings to differ in sign and magnitude, the sub-rating analysis can have better explanatory power. However, this result does not imply that the sub-ratings are superior in capturing important aspects of governance. In fact, since several of the sub-ratings' coefficients have the "wrong" sign, the disaggregation of the governance scores only demonstrates the inadequacies of the underlying governance constructs. One explanation could be that governance attributes are not uniformly good or bad, but good for some purposes and bad for others. This finding implies that using a governance construct to predict behavior that it is not designed for is likely to result in poor explanatory power or unexpected signs. Another potential explanation is that these governance watchdogs are not very good at defining good or bad governance. (1)

C. Stock Return Regressions

By using a method similar to that of Gompers et al. (2003) and Core et al. (2006), we first create portfolios based on the A-to-F TCL ratings, the TCL sub-ratings, and the quartiles of ISS and GMI ratings. We calculate equally weighted weekly returns on these portfolios and regress the excess weekly returns on the Fama-French (1995) factors and the Carhart (1997) momentum factor. For the TCL ratings and sub-ratings, we calculate the one-year portfolio returns starting from the 26th week of the year in which the rating is released.

For ISS and GMI, we calculate the 104-week returns for the years 2005 and 2006 for portfolios that we construct using 2005 ratings. Although doing so might create a look-ahead bias, since the release dates for GMI and ISS ratings are August 2005 and October 2005, respectively, we find no evidence of abnormal returns for the portfolios formed based on these rating quartiles. The exceptions are the fourth quartile and the investment strategy that buys the equally weighted portfolio of firms with the highest quartile of ISS ratings and sells the lowest quartile. However, the abnormal returns for this strategy and the fourth quartile portfolio are significantly negative.

The Fama-French (1995) factors are weekly market returns minus weekly returns on Treasury bills (RMF), weekly returns on the small minus big firm size portfolio (SMB), and weekly returns on the high minus low book-to-market value portfolio (HML). Further, we investigate the abnormal returns for an investment strategy that buys firms in the highest quartile of governance ratings (or highest rating) and sells firms in the lowest quartile (or lowest ratings) and report the results in the last two columns of Table VII.

Turning to the entries in the alpha rows of Panels A, B, and C of Table VII, most of the alpha estimates for TCL, ISS, and GMI overall ratings are not significant. The exceptions are the weakly significant negative abnormal returns for firms with high ISS overall rating (quartile 4) and the corresponding hedged portfolio (Q4-Q1). However, after we control for the four asset pricing factors, we find that several of the eight TCL sub-ratings-based portfolios exhibit weakly significant positive or negative abnormal returns. The portfolios include firms with the Accounting rating of A, which signals the high quality of reported earnings, significant at the 5% level and the B rating on Board Composition, which reflects such attributes as more independence, less over-committed members, etc.; and Litigation Problem, which signifies the lowest incidence of litigation, regulatory infractions, and fines; Shareholder Responsiveness; Strategic Decision-making, that is, board approval of mergers and acquisitions; and Analyst Adjustment, that is, discretionary adjustments. All these portfolios have weak but significantly positive abnormal returns. However, none of the portfolios with the lower D and F ratings has significantly negative abnormal returns. Nor do most of the investment strategies that buy firms with an A rating and sell firms with D or F ratings earn significantly positive mean abnormal returns.

As a robustness check, we repeat the stock-return regressions, now using value-weighted portfolios instead of equally weighted portfolios. In untabulated results, with the sole exception of the positive and marginally significant alpha estimate of the A-DF hedged portfolio based on the TCL rating almost all alpha estimates for the three overall ratings are not significant. Further, only the two portfolios with an A rating on Accounting and Strategic Decision Making have marginally significant positive abnormal returns. The rest of the sub-rating alphas are no longer significant. The difference between the equally weighted and value-weighted results implies that when we compare it with large firms, the predictive power of the TCL sub-ratings is marginally superior for small firms.

Consistent with the argument that the market quickly incorporates the effects of governance structures on firm performance into stock prices, our results suggest that it is difficult to profit from trading based on the overall governance scores. However, through several of their sub-ratings, TCL analysts provide new information that potentially could generate (weakly) positive abnormal returns.

One limitation of our stock-return analysis is the limited time-series. This limitation is especially true for ISS and GMI ratings, since our analysis covers only two years of stock-return data for portfolios that use only one year of ratings. Therefore, we should reexamine the links between stock returns and governance ratings when more data becomes available.

D. Endogeneity Concerns

Recent papers underscore the importance of tradeoffs between managerial discretion and agency costs in selecting optimal governance mechanisms and highlight the endogeneity of governance structures to firm and industry characteristics. These papers imply that better governance mechanisms might positively influence corporate performance by reducing agency costs. On the other hand, strong governance rules might dampen managerial discretion and innovation.

The relation between governance and firm performance is much more complex. Firms might change their governance structures over time in response to many endogenous factors. Based on the costs and benefits of each governance mechanism, firms adopt the best form for their characteristics. In a competitive equilibrium, corporate governance would be set at the optimal level, implying no cross-sectional relation between corporate governance and performance. In contrast, the evidence suggests that due to managerial power over the board, board monitoring may not be effective. Thus, many studies assume that the current governance regimes may be suboptimal and investigate the effects of governance on firm performance.

Although corporate governance structures are endogenous, it is difficult to incorporate the endogeneity of governance structures into our empirical analysis. The common approach to addressing the problem of endogeneity is to run an instrumental variable regression. However, this approach can work only if we can find strictly exogenous instruments, which is very difficult to do in practice. Nonetheless, we follow Gillan, Hartzell, and Starks (2006) and use as instruments the lagged values of leverage, capital expenditures over sales, natural log of firm age, and volatility of stock returns. We measure the values for the fiscal year prior to the release date of the rating.

For TCL ratings, the first stage is a probit regression in which the dependent variable is a dummy variable that equals one if the rating is equal to A or B, and zero if it equals C, D, or F. The dependent variable in the second stage is operating income after depreciation scaled by book value of assets. We follow Wooldridge (2002) to estimate this model. For ISS and GMI ratings, both stages use ordinary least squares. In untabulated results, we find the TCL Dummy and ISS (GMI) ratings are negatively (positively) related to future operating performance. This result is consistent with the OLS results presented in Table III. When we repeat our operating performance OLS tests with the TCL dummy that we use in instrumental variable regressions as the independent variable instead of the ordinal TCL overall rating variable, we find that the coefficient of the dummy variable is negative and significant at 5% level.

Further, similar to the OLS results reported in Table IV, the Accounting dummy variable has a significant positive coefficient. However, we note that the instruments we use are open to the criticism that they might actually be the result of governance quality rather than the cause.

As an alternative test of endogeneity, we analyze whether the changes in TCL ratings are associated with the changes in operating performance. The number of observations reduces to 1,914 in these regressions. In untabulated results, we find that the coefficient on the change in TCL rating is not significant. We note that this analysis also has low power, given the stability of the ratings over time and limited time-series we use. Further, it is hard to determine when a change in governance will result in a change in performance.

Given the limitations of our endogeneity tests, we conclude that in this study, it is difficult to establish causality between governance ratings and firm performance.

IV. Conclusion

The recent wave of corporate scandals has put significant pressure on money managers to focus on governance mechanisms that strengthen investor protection. As a result, several companies that rate firms on the strength of their governance mechanisms have emerged.

In this paper, we examine the predictive performance of three commercially available governance ratings: TCL, ISS, and GMI. When we compare the predictive ability of the overall governance ratings, we find that firms with high TCL overall ratings have lower future operating performance. Our results also indicate that the ISS summary ratings are negatively correlated with future operating performance, but the GMI overall ratings are positively correlated. On the other hand, while firms with the TCL rating of F have a higher probability of facing accounting-related SEC actions, firms with lower GMI ratings have a higher probability of litigation and delisting. These results provide weak support for the arguments that these ratings can identify firms with governance-related difficulties.

Furthermore, when we analyze the relation between the TCL sub-ratings and the primary and secondary measures of firm performance, we find that the TCL sub-rating on accounting can predict future operating performance. Litigation Problem sub-rating is negatively linked to the probability of several secondary measures of firm performance. Firms that score high on Shareholder Responsiveness, Takeover Defenses, and Analyst Adjustment exhibit lower probability of delisting. High scores on Shareholder Responsiveness and Analyst Adjustment are also associated with lower probability of class action lawsuits and bankruptcy. However, several sub-ratings have unexpected signs for the outcome variables.

We also investigate whether macro and micro corporate governance ratings can predict stock returns. Our results show that the average abnormal returns for portfolios that are based on the TCL, ISS, and GMI ratings are rarely significantly different from zero. However, trading strategies based on several TCL sub-ratings produce weakly positive abnormal returns, especially for small firms.

The summary measures of the three premier US commercial governance ratings generally perform poorly in predicting operating performance, stock returns, and probability of exposure to class action lawsuits, SEC enforcement action, delisting, and bankruptcy of the rated firms in our sample. In contrast, several of the TCL sub-ratings that focus on the eight components of its summary ratings exhibit significant predictive power with respect to the primary and secondary measures of firm performance, but others have incorrect or unexpected signs. These findings are consistent with recent observations by both academic researchers and practitioners that it is extremely difficult to distill the complex dimensions of numerous corporate governance mechanisms into one integrated summary score. We note one important caveat in evaluating these findings. Governance initiatives and monitoring by rating services may take time to alter the behavior of rated firms, but our study covers a short sample period.

We appreciate the helpful comments of an anonymous referee, the Editor, Joseph Golec, Karthik Krishnan, and Michael Willenborg, seminar participants at the University of Connecticut, University of Toledo, University of Southern Illinois--Edwardsville, University of Wisconsin--Oshkosh, the 2004 Financial Management Association European Meetings Doctoral Student Consortium, 2005 Eastern Finance Association Meetings, and the 2005 Financial Management Association Annual Meetings.

References

Bebchuk, L., A. Cohen, and A. Ferrell, 2009, "What Matters in Corporate Governance?" Review of Financial Studies 22, 783-827.

Bhagat, S. and B. Black, 2002, "The Non-Correlation between Board Independence and Long-Term Firm Performance," Journal of Corporation Law 27, 231-273.

Brown, L. and M. Caylor, 2006a, "Corporate Governance and Firm Operating Performance," Georgia State University, Working paper.

Brown, L. and M. Caylor, 2006b, "Corporate Governance and Firm Valuation," Journal of Accounting and Public Policy 25, 409-434.

Carhart, M., 1997, "On Persistence in Mutual Fund Performance," Journal of Finance 52, 57-82.

Chi, J., 2005, "Understanding the Endogeneity between Firm Value and Shareholder Rights," Financial Management 34, 65-76.

Core, J., W. Guay, and T. Rusticus, 2006, "Does Weak Governance Cause Weak Stock Returns? An Examination of Firm Operating Performance and Investors' Expectations," Journal of Finance 61, 655-687.

Core, J., W. Holthausen, and D. Larcker, 1999, "Corporate Governance, Chief Executive Officer Compensation, and Firm Performance," Journal of Financial Economics 51, 371-406.

Fama, E. and K. French, 1995, "Size and Book-to-Market Factors in Earnings and Returns," Journal of Finance 50, 131-155.

Fama, E. and K. French, 1997, "Industry Costs of Equity," Journal of Financial Economics 43, 153-193.

Gillan, L., J. Hartzell, and L. Starks, 2006, "Explaining Corporate Governance: Boards, Bylaws, Charter and State Provisions," University of Delaware Working Paper.

Gompers, P., J. Ishii, and A. Metrick, 2003, "Corporate Governance and Equity Prices," Quarterly Journal of Economics 118, 107-155.

Hermalin, B. and M. Weisbach, 1988, "The Determinants of Board Composition," The RAND Journal of Economics 19, 589-606.

Hermalin, B. and M. Weisbach, 1991, "The Effects of Board Composition and Direct Incentives on Firm Performance," Financial Management 20, 101-112.

Larcker, D., Richardson, S., and Tuna, I., 2005, "Ratings Add Fire to the Governance Debate," Financial Times, May 26.

Larcker, D., Richardson, S., and Tuna, I., 2007, "Corporate Governance, Accounting Outcomes, and Organizational Performance," The Accounting Review 82, 963-1008.

Lee, C., Rosenstein, S., Rangan, N., and Davidson, W, 1992, "Board Composition and Shareholder Wealth: The Case of Management Buyouts," Financial Management 21, 58-73.

McConnell, J. and Servaes, H., 1990, "Additional Evidence on Equity Ownership and Corporate Value," Journal of Financial Economics 27, 595-613.

Morck, R., Shleifer, A., and Vishny, R., 1988, "Management Ownership and Market Valuation: An Empirical Analysis," Journal of Financial Economics 20, 293-315.

Pender, K., 2004, "Google's Weak Governance Rating," San Francisco Chronicle, August 24.

Sonnenfeld, J. 2003, "Introducing the Watchdogs for Corporate Governance," Wall Street Journal, March 11.

Weisbach, M., 1988, "Outside Directors and CEO Turnover," Journal of Financial Economics 20, 431-460.

White, H., 1980, "A Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroscedasticity," Econometrica 48, 817-838.

Wooldridge, J.M., 2002, Econometric Analysis of Cross Section and Panel Data, Cambridge, MA, MIT Press.

Yermack, D., 1996, "Higher Market Valuation of Companies with a Small Board of Directors," Journal of Financial Economics 40, 185-211.

Mine Ertugrul and Shantaram Hegde *

* Mine Ertugrul is an Assistant Professor of Finance at the University of Toledo, Toledo, OH. Shantaram Hegde is a Professor of Finance at the University of Connecticut, Storrs, CT.

(1) We thank the referee for this insight.

Table I. Brief Descriptions of the TCL Sub-Ratings

TCL Sub-Rating        Description

Board composition     Covers board independence, director age and
                        tenure, overcommitted directors, active and
                        former CEOs on board, and whether the past
                        CEO is the chairman.
CEO compensation      Covers pay-for-performance sensitivity, stock
                        awards and stock ownership, restricted stock,
                        perquisite payments, fixed and variable
                        components of the CEO pay, and several red
                        flags such as base salaries in excess of $1
                        million, bonuses greater than twice the
                        annual salary, high tax or leisure expense
                        payments, a declining number of CEO shares
                        held, and excessive option compensation.
Shareholder           Reflects a board's history in responding to

  responsiveness        shareholder proposals that receive majority
                        of the votes. A board's failure to accept a
                        shareholder proposal that receives majority
                        indicates the degree to which the board acts
                        in the best interest of management rather
                        than shareholders.
Litigation and        Considers the incidence of litigation and
  regulatory            assessed fines. It includes an evaluation of
  problems              the amount of disclosure of current or
                        potential liability exposure and the
                        existence of repeated regulatory infractions
                        or fines.
Takeover defenses     Considers takeover defenses. Firms with more
                        "shareholder-friendly" takeover defense
                        configurations get higher ratings.
Accounting            Compares current quarter reports against
                        prior four quarters for indicators of
                        potential earnings management to identify
                        firms that favor accounting methods that
                        emphasize and enhance current period earnings
                        but may also lead to poor long-term (three-
                        to five-year) market performance. Also
                        accounts for other accounting concerns such
                        as SEC or IRS enforcement actions.
Strategic decision    Focuses on board approval of mergers and
  making                acquisitions with lower ratings assigned to
                        approvals of mergers resulting in significant
                        loss of shareholder value.
Analyst adjustment    Corrections to the regular scoring system by
                        the TCL analysts to account for special
                        factors that do not fall into any of their
                        current screens.

Table II. Summary Statistics

This table presents summary statistics for the three prominent
commercial governance ratings. Panel A presents percentage
distribution of TCL overall ratings and its sub-ratings. There are
4,546 firm-year observations for TCL ratings over 2003 to first 7
months of 2006 and 1,573 observations for TCL sub-ratings over the
same period. Panel B presents summary statistics for both
Institutional Shareholder Services (ISS) Corporate Governance
Quotient [R], and Governance Metrics International overall global
rating (GMI) for 2005. Panel C presents the correlation matrix for
all the ratings for 481 observations.

Panel A. Distribution of TCL Overall Ratings and Sub-Ratings

                     TCL        Board          CEO        Shareholder
                   Overall   Composition   Compensation   Responsive-
                   Rating        (%)           (%)           ness

A                    9.35        6.74          9.54           0.25
B                   37.84       44.37         31.28          90.59
C                   35.31       32.23         35.79           6.81
D                   15.66       14.37         16.27           1.72
F                    1.85        2.29          7.12           0.64
Number of           4,546       1,573         1,573          1,573
  observations

Panel B. ISS and GMI Ratings

Variable              N          Min          Quartl         Mean

ISS                 1,618       0.50          28.50          53.38
GMI                 1,487       2.50           6.50           7.22

Panel C. Correlation Matrix

                     TCL         ISS           GMI           Board
                   Overall                                Composition
                   Rating

TCL overall         1.000
  rating
ISS                 0.186       1.000
GMT                 0.003       0.131          1.000
Board               0.529       0.301          0.020         1.000
  composition
CEO                 0.500      -0.043          0.007         0.018
  compensation
Shareholder         0.148      -0.001         -0.024         0.046
  responsiveness
Litigation          0.215       0.080          0.062         0.015
  problems
Takeover            0.187       0.278          0.070         0.182
  defenses
Accounting          0.180      -0.057          0.050         0.000
Strategic           0.212       0.032          0.030         0.037
  decision-
  making
Analyst             0.473       0.171          0.056         0.175
  adjustment

Panel A. Distribution of TCL Overall Ratings and Sub-Ratings

                    Litigation     Takeover     Accounting   Strategic
                     Problems      Defenses        (%)       Decision
                       (%)            (%)                     Making

A                      0.06           0.32        17.29         1.08
B                     79.59          48.22        45.96        94.28
C                     14.30          27.80        16.15         2.16
D                      5.02          10.05        12.84         1.91
F                      1.02          13.61         7.76         0.57
Number of             1,573          1,573        1,573        1,573
  observations

Panel B. ISS and GMI Ratings

Variable              Median        Quart3         Max       Std. Dev.

ISS                   54.45          79.10        100.00       29.06
GMI                    7.00           8.00         10.00        3.44

Panel C. Correlation Matrix

                       CEO        Shareholder   Litigation   Takeover
                   Compensation   Responsive-    Problems    Defences
                                     ness

TCL overall
  rating
ISS
GMT
Board
  composition
CEO                    1.000
  compensation
Shareholder            0.099         1.000
  responsiveness
Litigation             0.091         0.092        1.000
  problems
Takeover               0.018         0.011        -0.022       1.000
  defenses
Accounting            -0.016         0.080        0.222        0.029
Strategic              0.030         0.085        0.450       -0.024
  decision-
  making
Analyst                0.016         0.147        0.291        0.105
  adjustment

Panel A. Distribution of TCL Overall Ratings and Sub-Ratings

                     Analyst
                    Adjustment
                       (%)

A                      0.06
B                     81.69
C                      7.88
D                      9.09
F                      1.27
Number of             1,573
  observations

Panel B. ISS and GMI Ratings

Variable

ISS
GMI

Panel C. Correlation Matrix

                    Accounting     Strategic     Analyst
                                   Decision     Adjustment
                                    Making

TCL overall
  rating
ISS
GMT
Board
  composition
CEO
  compensation
Shareholder
  responsiveness
Litigation
  problems
Takeover
  defenses
Accounting            1.000
Strategic             0.161          1.000
  decision-
  making
Analyst               0.283          0.249        1.000
  adjustment

Table III. TCL, ISS, and GMI Ratings and Operating Performance

This table presents OLS test results. We use the future operating
income after depreciation divided by total assets as the dependent
variable and the current year governance ratings as test variables.
TCL Overall Rating takes the value of five, four, three, two, or one
if the firm's The Corporate Library Rating is A, B, C, D, or F,
respectively. TCL-A is an indicator variable that equals one if TCL
Rating is A, and zero otherwise. We define TCL-B, TCL-D, and TCL-F
similarly. There are 3,887 firm-year observations for TCL overall
ratings over 2003 to 2005. ISS is Institutional Shareholder Services
Corporate Governance Quotient [R]. GMI is Governance Metrics
International overall global rating. The last two ratings are
available only for 2005. Size is the natural log of total assets. All
models include Fama-French (1997) industry indicator variables and
models with TCL Overall Rating and TCL rating dummies also include
year dummies. For brevity we do not report the coefficient estimates
of industry and year dummies. We correct standard errors for
heteroskedasticity and clustering of observations. p-values appear in
brackets.

                        (1)        (2)        (3)        (4)

TCL Overall Rating    -0.004
                      [0.038]
TCL-A                            -0.01
                                 [0.255]
TCL-B                            -0.006
                                 [0.075]
TCL-D                             0.003
                                 [0.508]
TCL-F                            -0.002
                                 [0.865]
ISS                                         -0.0003
                                            [0.013]
GMI                                                     0.007
                                                       [0.002]
Size                   0.006      0.006      0.008     -0.001
                      [0.047]    [0.053]    [0.003]    [0.795]
Constant               0.03       0.02       0.03       0.027
                      [0.258]    [0.450]    [0.281]    [0.290]
Industry dummies        Yes        Yes        Yes        Yes
Year dummies            Yes        Yes        No         No
Observations           3,887      3,887      1,243      1,148
Adjusted [R.sup.2]     0.09       0.09       0.13       0.13

Table IV. TCL Sub-Ratings and Operating Performance

This table presents OLS test results. We use the future operating
income after depreciation divided by total assets as the dependent
variable and the current year governance ratings as test variables.
Sub-rating takes the value of five, four, three, two, or one if the
firm's TCL sub-rating is A, B, C, D, or F, respectively. There are
952 observations for TCL sub-ratings over 2003 to 2005. Size is the
natural log of total assets. All models include Fama-French (1997)
industry indicator variables and year dummies. For brevity we do not
report the coefficient estimates of industry and year dummies. We
correct standard errors for heteroskedasticity and clustering of
observations. p-values appear in brackets.

TCL Sub-Rating                  Operating Performance

Board Composition                      -0.003
                                       [0.415]
CEO Compensation                       -0.002
                                       [0.548]
Shareholder Responsiveness             -0.002
                                       [0.696]
Litigation Problems                     0.007
                                       [0.494]
Takeover Defenses                       0.002
                                       [0.641]
Accounting                              0.02
                                       [0.000]
Strategic Decision Making              -0.004
                                       [0.721]
Analyst Adjustment                     -0.003
                                       [0.641]
Size                                    0.005
                                       [0.119]
Constant                               -0.014
                                       [0.790]
Industry and year dummies                Yes
Observations                             952
Adjusted [R.sup.2]                      0.14

Table V. Probit Regressions-TCL, ISS, and GMI Ratings

This table presents probit regression results using the dependent
variables Litigation, Action, Delisting, and Bankruptcy. Litigation
is a dummy variable that equals one if there is a securities class
action lawsuit filed against the firm following the release date of
the rating, and zero otherwise. Action is a dummy variable that
equals one if there is an accounting-based SEC enforcement action
against the firm following the release date of the rating, and zero
otherwise. Delisting is a dummy variable that equals one if the
company gets delisted following the release date of the rating, and
zero otherwise. Bankruptcy is a dummy variable that equals one if the
company files for bankruptcy following the release date of the
rating, and zero otherwise. TCL Overall Rating takes the value of
five, four, three, two, or one if the firm's TCL rating is A, B, C,
D, or F, respectively. TCL-A is an indicator variable that equals one
if the TCL rating is A, and zero otherwise. We define TCL-B, TCL-D,
and TCL-F similarly. ISS is Institutional Shareholder Services
Corporate Governance Quotient [R]. GMI is Governance Metrics
International overall global rating. Controls consist of ROA, Q,
Size, Leverage, and industry dummies. Many of the controls'
coefficient estimates are highly significant, but for brevity we do
not report them. Models with TCL Overall Rating and TCL rating
dummies also include year dummies, but we do not report their
coefficient estimates. We correct standard errors for
heteroskedasticity and clustering of observations. p-values appear in
brackets.

Panel A. TCL Overall Rating

                                Litigation        Litigation

TCL Overall Rating                 -0.07
                                   [0.235]
TCL-A                                                 0.073
                                                     [0.687]
TCL-B                                                -0.2
                                                     [0.140]
TCL-D                                                 0.084
                                                     [0.572]
TCL-F                                                 0.211
                                                     [0.404]
Controls                          ROA, Q,           ROA, Q,
                                   Size              Size
Industry and year dummies           Yes               Yes
Observations                       4,541             4,541
Observed probability (%)            1.32              1.32
[Chi.sup.2]                        46.14             52.32
Pseudo [R.sup.2]                    0.11              0.11

Panel B. ISS and GMI Ratings

ISS                                 0.000
                                   [0.870]
GMI                                                  -0.129
                                                     [0.094]
Controls                          ROA, Q,           ROA, Q,
                                   Size              Size
Industry dummies                    Yes               Yes
Observed probability (%)            2.23              2.20
Observations                        1,479             1,359
[Chi.sup.2]                        10.90             13.00
Pseudo [R.sup.2]                    0.03              0.05

Panel A. TCL Overall Rating

                                  Action            Action

TCL Overall Rating                 -0.112
                                   [0.184]
TCL-A                                                 0.157
                                                     [0.531]
TCL-B                                                 0.007
                                                     [0.969]
TCL-D                                                 0.225
                                                     [0.215]
TCL-F                                                 0.706
                                                     [0.020]
Controls                          ROA, Q,           ROA, Q,
                                   Size              Size
Industry and year dummies           Yes               Yes
Observations                        4,377             4,377
Observed probability (%)            0.68              0.68
[Chi.sup.2]                        56.19             76.07
Pseudo [R.sup.2]                    0.08              0.09

Panel B. ISS and GMI Ratings

ISS                                -0.005
                                   [0.416]
GMI                                                  -0.058
                                                     [0.678]
Controls                          ROA, Q,           ROA, Q,
                                   Size              Size
Industry dummies                    Yes               Yes
Observed probability (%)            0.44              0.35
Observations                        1,122              858
[Chi.sup.2]                        48.39             43.35
Pseudo [R.sup.2]                    0.21              0.20

Panel A. TCL Overall Rating

                                 Delisting         Delisting

TCL Overall Rating                  0.104
                                   [0.305]
TCL-A                                                 0.255
                                                     [0.458]
TCL-B                                                 0.213
                                                     [0.360]
TCL-D                                                 0.09
                                                     [0.779]
TCL-F

Controls                       ROA, Q, Size,     ROA, Q, Size,
                                 Leverage          Leverage
Industry and year dummies           Yes               Yes
Observations                       3,414              3,355
Observed probability (%)            0.45              0.45
[Chi.sup.2]                        93.12            104.89
Pseudo [R.sup.2]                    0.17              0.17

Panel B. ISS and GMI Ratings

ISS                                 0.005
                                   [0.459]
GMI                                                  -0.200
                                                     [0.065]
Controls                       ROA, Q, Size,     ROA, Q, Size,
                                 Leverage          Leverage
Industry dummies                    Yes               Yes
Observed probability (%)            0.87              0.68
Observations                        689               730
[Chi.sup.2]                        35.28             67.23
Pseudo [R.sup.2]                    0.39              0.39

Panel A. TCL Overall Rating

                                Bankruptcy        Bankruptcy

TCL Overall Rating                 -0.084
                                   [0.438]
TCL-A                                                 0.065
                                                     [0.876]
TCL-B                                                -0.051
                                                     [0.836]
TCL-D                                                 0.342
                                                     [0.201]
TCL-F

Controls                       ROA, Q, Size,     ROA, Q, Size,
                                 Leverage          Leverage
Industry and year dummies           Yes               Yes
Observations                        3,360             3,303
Observed probability (%)            0.48              0.48
[Chi.sup.2]                        77.75             76.20
Pseudo [R.sup.2]                    0.28              0.29

Panel B. ISS and GMI Ratings

ISS                                 0.007
                                   [0.299]
GMI                                                   0.004
                                                     [0.719]
Controls                       ROA, Q, Size,     ROA, Q, Size,
                                 Leverage          Leverage
Industry dummies                    Yes               Yes
Observed probability (%)            0.9               0.8
Observations                        889               744
[Chi.sup.2]                        48.12             33.99
Pseudo [R.sup.2]                    0.36              0.38

Table VI. Probit Regressions-TCL Sub-Ratings

This table presents probit regressions results for the TCL
sub-ratings using the dependent variables Litigation, Action,
Delisting, and Bankruptcy. Litigation is a dummy variable that equals
one if there is a securities class action lawsuit filed against the
firm following the release date of the rating, and zero otherwise.
Action is a dummy variable that equals one if there is an
accounting-related SEC enforcement action against the firm following
the release date of the rating, and zero otherwise. Delisting is a
dummy variable that equals one if the company gets delisted following
the release date of the rating, and zero otherwise. Bankruptcy is a
dummy variable that equals one if the company files for bankruptcy
following the release date of the rating, and zero otherwise.
Sub-rating takes the value of five, four, three, two, or one if the
firm's TCL sub-rating is A, B, C, D, or F, respectively. Control
variables consist of ROA, Q, Size, Leverage, industry dummies, and
year dummies. Many of the controls have highly significant
coefficient estimates, but for brevity we do not report them. We
correct standard errors for heteroskedasticity and clustering of
observations. p-values appear in brackets.

                                 Litigation          Action

Board Composition                   0.015            -0.077
                                   [0.899]           [0.635]
CEO Compensation                   -0.037            -0.157
                                   [0.743]           [0.353]
Shareholder Responsiveness         -0.464             1.044
                                   [0.002]           [0.001]
Litigation Problems                -0.516            -1.045
                                   [0.000]           [0.000]
Takeover Defenses                  -0.054            -0.101
                                   [0.563]           [0.471]
Accounting                          0.225            -0.058
                                   [0.010]           [0.604]
Strategic Decision Making           0.317             1.088
                                   [0.194]           [0.000]
Analyst Adjustment                 -0.214             0.144
                                   [0.016]           [0.484]
Constant                           -1.664            -7.001
                                   [0.290]           [0.000]
Controls                        ROA, Q, Size      ROA, Q, Size

Industry and year dummies            Yes               Yes
Observations                        1,495              981
Observed probabilities (%)          1.20              0.62
[Chi.sup.2]                       131.88             92.78
Pseudo [R.sup.2]                    0.27              0.30

                                  Delisting        Bankruptcy

Board Composition                  -0.100             0.206
                                   [0.720]           [0.420]
CEO Compensation                   -0.196             0.155
                                   [0.231]           [0.384]
Shareholder Responsiveness         -1.233            -0.55
                                   [0.007]           [0.021]
Litigation Problems                -1.383            -0.515
                                   [0.006]           [0.035]
Takeover Defenses                  -0.546            -0.159
                                   [0.013]           [0.334]
Accounting                          0.728             0.489
                                   [0.003]           [0.003]
Strategic Decision Making           2.298             1.129
                                   [0.008]           [0.008]
Analyst Adjustment                 -0.93             -0.699
                                   [0.010]           [0.002]
Constant                            0.528            -3.453
                                   [0.833]           [0.169]
Controls                        ROA, Q, Size,     ROA, Q, Size,
                                  Leverage          Leverage
Industry and year dummies            Yes               Yes
Observations                         463               748
Observed probabilities (%)          1.30              1.07
[Chi.sup.2]                        31.56             77.36
Pseudo [R.sup.2]                    0.58              0.50

Table VII. Stock Returns Regressions

This table presents OLS test results using weekly equally weighted
portfolio returns based on the categories and quartiles of governance
ratings as dependent variables, Fama and French (1995) asset pricing
factors, and the momentum factor as independent variables. TCL
Overall Rating is The Corporate Library rating. ISS is Institutional
Shareholder Services Corporate Governance Quotient [R]. GMI is
Governance Metrics International overall global rating. Quartile 1
(Ql) covers a portfolio of firms with low governance ratings and
Quartile 4 (Q4) represents a portfolio with high scores. Q4-Q1
represents weekly returns on a hedge portfolio that is long on high
governance rating firms and short on low governance rating firms.
A-DF (A-F) represents weekly returns on a hedge portfolio that is
long on firms with TCL rating of A and short on firms with TCL rating
of D or F (F). RMF is the weekly market return minus weekly return on
Treasury bills. SMB and HML are the small minus big, and high minus
low weekly portfolio returns, respectively, as defined in Fama and
French (1995). Momentum represents weekly returns of a
zero-investment strategy based on past returns as defined in Carhart
(1997). All regressions include the Fama-French plus momentum factors
as control variables, but for brevity we do not report the
coefficients of the factors for regressions with ISS, GMI, and TCL
sub-ratings. We use the White (1980) method to correct for
heteroskedasticity. p-values appear in brackets.

Panel A. TCL Overall Ratings

                          F          D          C          B

RMF                     1.331      1.262      1.349      1.305
                       [0.000]    [0.000]    [0.000]    [0.000]
SMB                     0.209      0.229      0.495      0.756
                       [0.178]    [0.062]    [0.000]    [0.000]
HML                     0.215      0.043      0.035     -0.107
                       [0.320]    [0.733]    [0.840]    [0.476]
Momentum               -0.043      0.101      0.192      0.149
                       [0.697]    [0.307]    [0.113]    [0.154]
Alpha                   0.000      0.001      0.002      0.001
                       [0.708]    [0.183]    [0.144]    [0.125]
Observations             160        160        160        160
Adjusted [R.sup.2]      0.62       0.76       0.77       0.83

Panel B. ISS Ratings (Alphas)

                         Q1         Q2         Q3         Q4

Alpha                   0.0001     0.0002    -0.0001    -0.0005
                       [0.606]    [0.622]    [0.612]    [0.097]
Observations             104        104        104        104
Adjusted [R.sup.2]      0.97       0.97       0.98       0.96

Panel C. GMI Ratings (Alphas)

                         Q1         Q2         Q3         Q4

Alpha                  -0.0002    0.0001     0.0004     -0.0004
                       [0.422]    [0.769]    [0.147]    [0.188]
Observations             104        104        104        104
Adjusted [R.sup.2]      0.98       0.97       0.80       0.97

Panel D. TCL Sub-Ratings (Alphas)

Board composition       0.000      0.001      0.002      0.002
                       [0.952]    [0.301]    [0.121]    [0.074]
CEO compensation        0.001      0.002      0.002      0.002
                       [0.441]    [0.092]    [0.085]    [0.219]
Shareholder            -0.001      0.004      0.000      0.002
  responsiveness       [0.782]    [0.041]    [0.714]    [0.074]
Litigation problems     0.001     -0.001      0.002      0.002
                       [0.800]    [0.827]    [0.201]    [0.092]
Takeover defenses       0.001      0.001      0.002      0.002
                       [0.338]    [0.372]    [0.114]    [0.101]
Accounting             -0.001      0.002      0.002      0.002
                       [0.700]    [0.259]    [0.132]    [0.108]
Strategic               0.004      0.000      0.002      0.002
  decision making      [0.224]    [0.990]    [0.217]    [0.090]
Analyst adjustment     -0.002      0.001      0.006      0.002
                       [0.194]    [0.408]    [0.389]    [0.073]

Panel A. TCL Overall Ratings

                          A        A-DF        A-F

RMF                     1.213     -0.033     -0.118
                       [0.000]    [0.644]    [0.228]
SMB                     0.927      0.568      0.718
                       [0.000]    [0.000]    [0.000]
HML                     0.108     -0.292     -0.107
                       [0.564]    [0.028]    [0.525]
Momentum                0.145      0.181      0.188
                       [0.190]    [0.041]    [0.097]
Alpha                   0.000      0.001     -0.001
                       [0.923]    [0.347]    [0.602]
Observations             160        160        160
Adjusted [R.sup.2]      0.77       0.34       0.31

Panel B. ISS Ratings (Alphas)

                        Q4-Q1

Alpha                  -0.0007
                       [0.053]
Observations             104
Adjusted [R.sup.2]      0.03

Panel C. GMI Ratings (Alphas)

                        Q4-Q1

Alpha                  -0.0002
                       [0.545]
Observations             104
Adjusted [R.sup.2]      0.66

Panel D. TCL Sub-Ratings (Alphas)

Board composition       0.001      0.002      0.001
                       [0.730]    [0.198]    [0.735]
CEO compensation        0.002      0.000      0.001
                       [0.125]    [0.808]    [0.656]
Shareholder            -0.002      0.003     -0.001
  responsiveness       [0.595]    [0.424]    [0.827]
Litigation problems    -0.008      0.001     -0.011
                       [0.537]    [0.735]    [0.385]
Takeover defenses      -0.002      0.001     -0.003
                       [0.782]    [0.653]    [0.643]
Accounting              0.002      0.003      0.003
                       [0.045]    [0.118]    [0.110]
Strategic               0.002     -0.003     -0.002
  decision making      [0.272]    [0.447]    [0.611]
Analyst adjustment     -0.002      0.004     -0.001
                       [0.582]    [0.047]    [0.748]