Agency Costs of Free Cash Flow and the Effect of Shareholder Rights on the Implied Cost of Equity Capital

Transcription

1 JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS Vol. 46, No. 1, Feb. 2011, pp COPYRIGHT 2011, MICHAEL G. FOSTER SCHOOL OF BUSINESS, UNIVERSITY OF WASHINGTON, SEATTLE, WA doi: /s Agency Costs of Free Cash Flow and the Effect of Shareholder Rights on the Implied Cost of Equity Capital Kevin C. W. Chen, Zhihong Chen, and K. C. John Wei Abstract In this paper, we examine the effect of shareholder rights on reducing the cost of equity and the impact of agency problems from free cash flow (FCF) on this effect. We find that firms with strong shareholder rights have a significantly lower implied cost of equity after controlling for risk factors, price momentum, analysts forecast biases, and industry and year effects than do firms with weak shareholder rights. Further analysis shows that the effect of shareholder rights on reducing the cost of equity is significantly stronger for firms with more severe agency problems from FCFs. I. Introduction Using a sample of 13,140 firm-year observations from 1990 to 2004 in the United States, we examine the association between shareholder rights and the implied cost of equity and explore how this association is affected by agency problems from free cash flows (FCFs). Our paper is motivated by a growing body of analytical and empirical research that investigates the relation between shareholder rights and firm valuation as well as stock returns. On the theory side, several recent papers suggest that strong shareholder rights may lower the cost of capital. Albuquerque and Wang (2008) and Garmaise and Liu (2005) suggest that strong shareholder rights can mitigate overinvestment problems, which, in turn, Chen, acchen@ust.hk, and Wei, johnwei@ust.hk, School of Business and Management, Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong; Chen, chenzhh@cityu.edu.hk, College of Business, City University of Hong Kong, Tat Chee Ave., Kowloon, Hong Kong. We appreciate the helpful comments of Jeffrey Callen, Andrew Chen, Agnes Cheng, Sudipto Dasgupta, Joseph Fan, Vidhan Goyal, Mingyi Hung, Peter Kennedy, Clive Lennox, James Ohlson, Stephen Penman, T. J. Wong, and especially Bin Ke and Mike Lemmon, and seminar participants at the Chinese University of Hong Kong, City University of Hong Kong, Hong Kong University of Science and Technology, National Cheng Kung University, Singapore Management University, the 2004 National Taiwan University Conference on Finance, where the paper was the winner of the best paper award, and the 2005 Asian Institute of Corporate Governance conference held at Korea University. We also thank Paul Malatesta (the editor) and an anonymous referee for insightful comments and suggestions and Dr. Virginia Unkefer for editorial assistance. We acknowledge financial support from the Research Grants Council of the Hong Kong Special Administration Region, China (HKUST6134/02H). 171

2 172 Journal of Financial and Quantitative Analysis lower the cost of capital. Other studies also suggest that strong shareholder rights can reduce the cost of capital by reducing investors out-of-pocket monitoring costs (Lombardo and Pagano (2002a)) or by reducing the idiosyncratic risk (Merton (1987), Giannetti and Simonov (2006), and Himmelberg, Hubbard, and Love (2004)). On the empirical side, however, the evidence is mixed and sometimes puzzling. Gompers, Ishii, and Metrick (GIM) (2003) use the number of antitakeover provisions in corporate charters and bylaws (denoted as the G-index ) as an inverse measure of firm-level shareholder rights or external corporate governance mechanisms and find that firms with strong shareholder rights (i.e., few antitakeover provisions or a low G-index) have significantly higher realized stock returns. 1 This result is surprising because current stock prices should incorporate information on future performance in an efficient market. One possibility is that analysts do not incorporate the benefit of shareholders rights and therefore underestimate their earnings forecasts in firms with strong shareholder rights. However, Core, Guay, and Rusticus (2006) find no such underestimation. This leaves open the alternative interpretation that the negative association between the G-index and the realized returns found by GIM may be attributable to risks or other factors that happen to be correlated with shareholder rights. 2 Nevertheless, the theoretical models mentioned earlier point out that the association between shareholder rights and the cost of capital should be negative. GIM also find that the G-index is negatively associated with firm value. However, this valuation effect may not be driven by different costs of capital (i.e., the discount rate effect). It is also possible that the valuation effect primarily reflects different investment opportunities and different levels of expropriation (i.e., the cash flow effect). 3 Thus, whether and how shareholder rights affect expected returns require further study. The issue is complicated by the debate in current empirical research that ex post returns might not be an appropriate proxy for a firm s cost of equity. As pointed out by Hail and Leuz (2006), (2009), Stulz (1999), and others, this proxy not only captures differences in a firm s cost of equity, but it may also reflect the shocks to a firm s growth opportunities, differences in expected growth rates, and changes in investors risk aversion. As a result, using ex post returns to estimate expected returns requires average realized returns over a long period. However, Elton (1999) argues that realized returns may differ from expected returns, even over a long period. In addition, Fama and French (1997) conclude that expected 1 Bebchuk, Cohen, and Ferrell (2009) find similar results using their proposed entrenchment index (the E-index, to be discussed in Section IV.A) as an alternative measure of shareholder rights. 2 Core et al. (2006) find that high G-index firms outperformed low G-index firms during , which is a reversal of the pattern that GIM (2003) document for Moorman (2004) finds no significant association between shareholder rights and ex post returns when benchmark returns are based on the portfolios of matched firms in the same industry and with a similar momentum and size. Thus, how and why shareholder rights are related to ex post realized returns are still unresolved questions. 3 Core et al. (2006) show that firms with strong shareholder rights have better accounting performance, as measured by operating return on assets (ROA), than do firms with weak shareholder rights. This result confirms that firms with strong shareholder rights have higher expected cash flows, but whether or not these firms have lower discount rates (or lower costs of equity) is still far from obvious.

3 Chen, Chen, and Wei 173 returns estimated by ex post returns and asset pricing models are imprecise because of the uncertainty of the factor premiums and the imprecision in the factor loading estimates. Our study adds to the literature by examining the association between shareholder rights and the ex ante cost of equity implied in stock prices and analysts earnings forecasts. This approach follows the prior literature in finance and accounting. 4 Implied cost of equity models may offer useful insights because they make an explicit attempt to separate the effect of the cost of equity from firm valuations and control for cash flow and growth effects (Hail and Leuz (2006), (2009)). In addition, Pástor et al. (2008) show analytically that under plausible conditions, the implied cost of equity is perfectly correlated with the conditional expected stock return. These advantages have motivated many researchers to test the associations between the implied cost of equity and firm or country attributes, such as institutions and securities regulations (Hail and Leuz (2006)), cross-listing (Hail and Leuz (2009)), default risk (Chava and Purnanandam (2010)), voluntary disclosure (Francis, Khurana, and Pereira (2005), and Botosan (1997)), accounting attributes (Francis, LaFond, Olsson, and Schipper (2004)), and corporate governance in an international setting (Chen, Chen, and Wei (2009)). We find that the G-index is significantly positively associated with the implied cost of equity after controlling for risk factors, price momentum, analysts forecast biases, and industry and year effects. The results imply that, ceteris paribus, a decrease in the G-index by 10 points (the index ranges from 1 to 17 in our sample) on average reduces the cost of equity by 34 basis points (bp). The effect of shareholder rights is not only statistically significant, but also economically significant. Under reasonable assumptions, this reduction in the cost of equity implies a value enhancement of over 6.8% (or $92 million for a firm with the median size of $1.35 billion in our sample). This value enhancement accounts for 23% of the valuation effect implied in the regression results reported by GIM (2003). We perform various robustness tests on the assumptions of the perpetual growth rate in the cost of equity estimates, the endogeneity of shareholder rights, and the effects of other corporate governance mechanisms. We find that the positive association between the G-index and the cost of equity continues to hold. We further investigate whether the association between shareholder rights and the cost of equity varies across firms with different degrees of potential agency problems from FCFs. We find that the marginal effect of the G-index on the cost of equity is stronger for firms with more severe agency problems from FCFs. Specifically, we find that the effect of the G-index on the cost of equity capital is significantly higher in the group of firms with high FCFs and poor investment opportunities than in the group of firms with low FCFs and good investment opportunities. This evidence is consistent with the predictions of Albuquerque and Wang s (2008) and Garmaise and Liu s (2005) models and further supports the 4 See, for example, Friend, Westerfield and Granito (1978), Kaplan and Ruback (1995), Claus and Thomas (CT) (2001), Fama and French (2002), Brav, Lehavy and Michaely (2005), Lee, Ng, and Swaminathan (2009), and Pástor, Sinha, and Swaminathan (2008) in finance and Gebhardt, Lee, and Swaminathan (GLS) (2001), Easton (2004), and Ohlson and Juettner-Nauroth (OJ) (2005) in accounting.

4 174 Journal of Financial and Quantitative Analysis hypothesis that strong shareholder rights reduce the cost of equity by mitigating agency problems from FCFs. Our paper contributes to the literature on corporate governance in general and on shareholder rights in particular. First, we make an important extension to GIM (2003) by identifying a channel through which shareholder rights improve firm value. Core et al. (2006) found that stronger shareholder rights are associated with higher earnings. Masulis, Wang, and Xie (2007) show that acquirers with weaker shareholder rights experience lower announcement-period abnormal returns. Fahlenbrach (2009) reports that firms with weaker shareholder rights have higher chief executive officer (CEO) compensation. While all of these studies show that weaker shareholder rights are associated with lower expected cash flows (the cash flow effect), our study suggests that weaker shareholder rights are also associated with higher discount rates (the discount rate effect). More importantly, we also identify that shareholder rights matter most in reducing the cost of equity when agency problems from FCFs are most severe. The results provide further empirical evidence that supports the presence of a causal link between corporate governance, agency problems, and the cost of equity predicted by Albuquerque and Wang (2008) and Garmaise and Liu (2005). Our paper also contributes to the literature on corporate governance and financing costs. Among others, Ashbaugh-Skaife, Collins, and LaFond (2006b), Klock, Mansi, and Maxwell (2005), and Cremers, Nair, and Wei (2007) study the effect of shareholder rights on the cost of debt. We focus on the association between shareholder rights and the cost of equity. Other studies investigate the association between country-level corporate governance and the country-level cost of equity (Hail and Leuz (2006), Lombardo and Pagano (2002b), Bhattacharya and Daouk (2002), and Daouk, Lee, and Ng (2006)). However, these studies look at legal institutions that affect all firms equally within a single jurisdiction, and they do not address the issue of how firm-specific corporate governance, such as shareholder rights, affects the firm-level cost of equity capital. Understanding this issue is important because there is still significant variation in firm-level corporate governance within each country. We therefore make an important contribution by extending these country-level studies to the firm level. Compared with countrylevel studies, such as Hail and Leuz (2006), our study focuses on voluntary corporate action rather than mandatory regulations. Voluntary corporate action is more susceptible to self-selection issues. To mitigate this concern, we conduct extensive robustness tests to address the endogeneity of shareholder rights and find similar results. We note that two contemporaneous and independent studies are directly related to ours. 5 Ashbaugh-Skaife, Collins, and LaFond (2006a) use dividend forecasts and target prices provided by Value Line to estimate the implied cost of 5 In addition, using firm-level data from 38 countries, Himmelberg et al. (2004) found that the marginal cost of equity, as proxied by the marginal profitability of capital, is positively correlated with insider ownership (a proxy for weak investor protection). Elston and Rondi (2006) extend Himmelberg et al. to firm-level governance and test the relation between the marginal cost of equity and firmlevel governance using German and Italian samples. However, as the proxy for the cost of equity is profitability in these models, it is not clear whether they capture the cash flow effect or the discount rate effect.

5 Chen, Chen, and Wei 175 equity to test the association between the implied cost of equity and a number of corporate governance attributes, including shareholder rights. They find that their implied cost of equity measure is not significantly correlated with shareholder rights, as measured by the E-score consisting of 6 of the 24 antitakeover provisions in GIM (2003). Cheng, Collins, and Huang (2006) use the implied cost of equity estimated from one of the four models that we use (i.e., OJ (2005)) and test the association between the G-index and this implied cost of equity. They find that the implied cost of equity capital is significantly positively associated with the G-index, which is consistent with our findings. Our study differs from these two studies in the following important aspects. First, while Ashbaugh-Skaife et al. s (2006a) approach of using Value Line data to estimate the implied cost of equity has the advantage of not requiring positive earnings forecasts, it is conceptually different from the approach that we use. Their approach estimates the discount rate that is applicable to short-term future cash flows only (i.e., up to the target price date), whereas our approach estimates the long-term geometric average of the term structure of the cost of equity applied to all future cash flows from now to perpetuity (Easton (2006)). Thus, these two approaches may generate different results. Second, both of these studies rely on a single estimate of the implied cost of equity, which can be sensitive to alternative underlying assumptions or implementation approaches. In contrast, our study uses the median of four estimates from four different models, which should reduce the measurement errors. Third, the estimates of the implied cost of equity rely heavily on the assumption of long-term growth rates. Whereas Cheng et al. (2006) do not consider this issue, we explicitly take it into account. Finally, we further support our hypothesis by documenting that the agency problems from FCFs have an important effect on the relation between shareholder rights and the cost of equity. The remainder of this paper is organized as follows. Section II develops the hypotheses. Section III describes the sample selection process, the estimation of the cost of equity capital, and the measurement of the variables. Section IV presents our main empirical results. Section V reports on the sensitivity analysis of the assumptions on the perpetual growth rate. Section VI provides further analysis of the endogeneity of shareholder rights and considers whether shareholder rights are substitutes or complements to other corporate governance mechanisms in reducing the cost of equity. Finally, Section VII concludes the paper. II. Hypothesis Development Several recent analytical studies suggest that strong corporate governance may reduce the cost of equity. In particular, we focus on Albuquerque and Wang (2008) and Garmaise and Liu (2005) to develop our main hypotheses. 6 Both 6 Other papers also suggest that strong corporate governance may reduce the cost of capital. Lombardo and Pagano (2002a) develop a model where outside investors have to pay an out-of-pocket monitoring cost in order to get certain payoff. Monitoring costs may include the cost of collecting information about managers expropriations and the cost of punishing managers when their expropriations are detected. Investors require compensation for the out-of-pocket monitoring costs by requiring higher expected returns. Better corporate governance increases the punishment imposed on managers

6 176 Journal of Financial and Quantitative Analysis studies suggest that weak corporate governance generates excessive investment, which, in turn, leads to a higher cost of capital via overinvestment. Albuquerque and Wang (2008) develop a dynamic stochastic general equilibrium model that considers the effect of investor protection on investment and the equity risk premium, among others. They assume that investment-specific technology shocks are the only source of volatility of investment output, that firms investment decisions are made by controlling shareholders who can extract private benefits from outside investors at a cost, and that controlling shareholders are risk averse. Under this setting, the decision for controlling shareholders to increase investment depends on the tradeoff between the increase in volatility of output and the increase in the amount of private benefit extraction. Strong investor protection imposes higher punishment on the diversion behaviors of controlling shareholders and therefore reduces overinvestment. The model predicts that, in equilibrium, weaker investor protection implies more overinvestment and more volatility in investment output, ceteris paribus. Since, in their model, the equity risk premium is positively associated with the volatility of investment output, Albuquerque and Wang imply that weaker investor protection suggests a higher equity risk premium (or a higher cost of equity). Garmaise and Liu (2005) study how corporate governance (i.e., whether managers or shareholders are in control) and agency problems (i.e., whether managers are honest or dishonest) affect a firm s equilibrium investment level and its systematic risk. In their model, managers always prefer more investment than less because they can obtain more private benefits when there is more investment. In addition, managers can observe private signals about the quality of an investment project and then verify the quality of the private signals in reports. The signals can be general signals (low precision) or specific signals (perfectly revealing). A general signal, on average, may convey either better or worse news about a project s quality than a specific signal can convey. The investment decision is made either by shareholders when corporate governance is strong or by managers when corporate governance is weak according to the reported signal. Managers may be dishonest by hiding a specific signal by claiming that they observed a general signal. Since managers prefer high investment levels, they hide a specific signal only when it indicates that the quality of a project is poor. Shareholders fully realize this possibility. Thus, when they control investment decisions, they adjust the investment level downward when a general signal is reported. In this case, the systematic risk (BETA) is identical to that in the first-best case (i.e., for expropriations, which increases the expected costs of expropriations for managers and therefore reduces their incentive to expropriate. A lower incentive to expropriate reduces the need for monitoring, which in turn decreases out-of-pocket costs and the required rate of returns. Merton s (1987) recognition hypothesis suggests that if investors are unwilling to invest in firms with poor corporate governance because of fear of expropriation, existing shareholders have to bear more idiosyncratic risk, which in turn increases the cost of equity. Existing evidence suggests that investors are reluctant to hold shares in firms with poor corporate governance (Giannetti and Simonov (2006)). In Himmelberg et al. s (2004) model, the equilibrium marginal cost of equity is a weighted average of systematic risk and idiosyncratic risk. Weaker corporate governance forces insiders to hold more shares to commit credibly to less expropriation. Therefore, insiders bear a higher level of idiosyncratic risk, which in turn increases their marginal cost of equity.

7 Chen, Chen, and Wei 177 when shareholders observe the private information of managers). However, when managers control investment decisions, managers can invest more than desired by shareholders by hiding a poor specific signal and making the investment consistent with their reported general signal. In this situation, firm value is further reduced when conditions in the world are poor. Therefore, in the presence of information asymmetry (managers have private information) and agency problems (dishonest managers can hide private information), weak governance increases firms exposure to systematic risk and increases the cost of equity. Hostile takeovers are among the most important governance mechanisms in the United States to monitor and discipline managers (Manne (1965), Jensen (1988)). 7 Consistent with this proposition, Kini, Kracaw, and Mian (2004) show that during the period from 1979 to 1998, the probability of CEO turnover after a takeover is negatively associated with a firm s performance before the takeover and positively associated with the hostility of the takeover. 8 The presence of antitakeover provisions shifts the power from shareholders to managers and decreases the effectiveness of the takeover market in disciplining managers. Therefore, the presence of more antitakeover provisions will induce greater expropriations by managers in various forms. Consistent with this conjecture, GIM (2003) find that firms with more antitakeover provisions tend to invest more and engage in more mergers and acquisitions. Masulis et al. (2007) find that acquirers with more antitakeover provisions have significantly lower abnormal announcement-period stock returns, which suggests that such firms are more likely to indulge in empirebuilding acquisitions that destroy shareholder value. Finally, Fahlenbrach (2009) further finds that firms with weaker shareholder rights offer higher compensation to CEOs. Combining the predictions of the theoretical models and empirical findings discussed previously leads to our first hypothesis, which is as follows: Hypothesis 1. The weaker the shareholders rights, the higher the cost of equity. Since shareholder rights are used as one mechanism to mitigate agency problems, we expect that the benefit of shareholder rights in reducing the cost of equity is stronger when agency problems are more severe. This can be inferred from Garmaise and Liu s (2005) model. More specifically, their model suggests that the difference in systematic risk between having weak and strong corporate governance increases as agency problems become more severe (i.e., the likelihood of dishonest managers increases). 9 In their model, strong corporate governance 7 Stulz (1999) discusses several mechanisms that are useful in monitoring and disciplining managers, including the board of directors and active shareholders such as large shareholders and institutional shareholders. We consider the effects of these alternative monitoring mechanisms in Section VI.B. 8 They also document that such relations are insignificant in the period from 1989 to Their conjecture is that the increased presence of antitakeover statutes may have reduced the effectiveness of takeovers as a disciplinary force. Another reason could be that other corporate governance mechanisms have improved. 9 This argument can be inferred from their Propositions 3, 5, and 6, and Lemmas 4 and 5. The detailed analysis is available from the authors.

8 178 Journal of Financial and Quantitative Analysis means more shareholder control, which, in turn, suggests stronger shareholder rights. Albuquerque and Wang (2008) do not explicitly model the severity of agency problems. However, their results imply that the marginal effect of investor protection on the equity risk premium increases as investor protection becomes weaker (i.e., the cost of diverting a given fraction of output is lower). 10 If investor protection can be interpreted as a combination of the level of shareholder rights and the inverse of the severity of agency problems, the previous result suggests that when agency problems are more severe, the marginal effect of shareholder rights on the cost of equity is stronger. In this paper, we focus on a typical agency conflict, that is, the overinvestment of FCFs as suggested by Jensen (1986) for three reasons. First, the models proposed by Albuquerque and Wang (2008) and Garmaise and Liu (2005) suggest that strong shareholder rights reduce the cost of equity by limiting corporate managers overinvestment. Therefore, providing evidence on how the association between shareholder rights and the cost of equity varies with the severity of agency problems from overinvestment would be a more direct test of the predictions of their models. Second, Jensen (1986), (1988) argues that hostile takeovers are effective in mitigating agency problems from FCFs, and that firms with higher FCFs are more likely to be hostile takeover targets. Third, while it is possible that agency conflicts exist in situations other than the overinvestment of FCFs (e.g., managers may shirk and enjoy the quiet life, which may induce underinvestment (Bertrand and Mullainathan (2003)), it would be less straightforward to find a proxy for the magnitude of the potential damage caused by managerial shirking or underinvestment. We argue that FCFs can be used as a proxy for the potential damage caused by overinvestment (the severity of agency problems). When there is little FCF, managers have fewer economic resources to squander. In this case, investment in additional projects has to be financed by external funds from the capital market, where managers will be subject to extra monitoring (Stulz (1999)). The extra monitoring by the capital market could force managers to reduce their expropriations, since new investors will not buy new shares unless they are compensated for agency costs (Jensen and Meckling (1976)). Thus, the potential damage from overinvestment to existing shareholders would be lower. In contrast, managers of firms with high FCFs can finance investments by internal funds and therefore avoid extra monitoring from the capital market. In this case, the potential damage from overinvestment to existing shareholders would be higher (Lang, Stulz, and Walkling (1991), Lamont (1997)). The previous discussions suggest that when FCFs are high, the marginal effect of shareholder rights on reducing the cost of equity should be more pronounced, which leads to our second hypothesis: 10 We could not analytically derive this conclusion. However, we conducted a simulation analysis and plotted the equity risk premium (λ) on investor protection (η) under reasonable assumptions suggested by Albuquerque and Wang ((2008), pp ). We find that λ is a convex decreasing function of η. The results are available from the authors.

9 Chen, Chen, and Wei 179 Hypothesis 2. The marginal effect of shareholder rights on reducing the cost of equity is more pronounced for firms with high agency costs from FCFs than for firms with low agency costs from FCFs. III. Data and Research Design A. Sample Selection The two key variables used in this study are shareholder rights and the estimate of the cost of equity. Shareholder rights are measured by the Governance Index ( G-index ) constructed by GIM (2003). GIM define the G-index (G INDEX) as the number of antitakeover provisions and restrictions of shareholder rights (with a total of 24) stipulated in a firm s charter and bylaws or by state law. The data are published by the Investor Responsibility Research Center (IRRC). 11 These provisions could delay hostile bidders, balance voting rights toward managers, and effectively protect managers/directors in the event of a takeover. A high G-index value means that the balance of power is more favorable to the management and implies that it is more difficult for outside investors to take over the firm and/or to replace the board of directors and the management. Thus, a higher G-index indicates weaker shareholder rights. GIM include the G-index for 1990, 1993, 1995, and We expand the data set by including the G-index for 2000, 2002, and Following GIM, we assume that the G-index is equal to the previously published value until the next publication. Our initial sample consists of all firms with a G-index (32,103 firm-years in total). We first match these firms to Compustat and the Center for Research in Security Prices (CRSP) to retrieve the Standard Industrial Classification (SIC) codes, book values of equity, and stock prices at month +4 after the fiscal year-end. We delete financial firms (SIC codes ) from the sample. We then match the sample with the Institutional Brokers Estimate System (IBES) to retrieve analysts earnings forecasts to estimate the implied cost of equity. We delete observations with insufficient data to estimate the cost of equity. Because we use the median of four estimates of the cost of equity to mitigate measurement errors (discussed later), we further require that all four estimates of the cost of equity be available. Finally, we eliminate the observations with missing control variables. This selection process results in a final sample of 13,140 firm-year observations (2,161 firms) across the 44 industries as defined by Fama and French (1997) from 1990 to Table 1 gives the sample selection process. B. Estimation of the Cost of Equity We estimate the cost of equity that is implied in current stock prices and analysts earnings forecasts based on four models introduced by CT (2001), GLS (2001), Easton (2004), and OJ (2005) as implemented by Gode and Mohanram 11 For details on the construction of the G-index and an explanation of each governance provision, please see the Appendix in GIM (2003).

10 180 Journal of Financial and Quantitative Analysis TABLE 1 Sample Selection Process Table 1 presents the sample selection process. The initial sample consists of all firms with a G-index. We first match the initial sample from IRRC to Compustat and CRSP to obtain the book value of equity and the stock price at month +4 after the fiscal year-end. We delete financial firms (SIC codes ) from the sample. We then match these firms to IBES to obtain the analysts earnings forecasts. We delete observations with insufficient data to estimate the cost of equity. Because we use the median of four estimates of the cost of equity to mitigate measurement errors, we further require that four estimates of the cost of equity be available. Finally, we eliminate observations with missing control variables. This selection process results in a final sample of 13,140 firm-year observations (2,161 firms) across the 44 industries defined by Fama and French (1997) from 1990 to Sample Selection Process Firm-Year Observations Firm-year observations in IRRC with the G-index from 1990 to 2004, assuming that the G-index equals the previous announced value until the next announcement. 32,103 Match to Compustat to obtain the book value of equity (data item 60). 23,205 Match to CRSP to obtain the stock price at month +4 after the fiscal year-end. 21,744 Delete firms in financial industries (SIC codes ). 18,502 Match to IBES to obtain analysts earnings forecasts at month +4 after the fiscal year-end. 15,933 Delete firm-year observations without all four cost of equity estimates. 14,128 Delete firm-year observations without control variables. 13,140 Final sample 13,140 (2003). The first two models are based on Ohlson s (1995) residual income valuation model, and the latter two models are based on OJ s abnormal earnings growth valuation model. Detailed descriptions of the cost of equity estimations are summarized in the Appendix. Since there is little consensus in the literature on which models perform best or how the models should be evaluated (e.g., Botosan and Plumlee (2005), Gode and Mohanram (2003), Guay, Kothari, and Shu (2003), and Easton and Monahan (2005)), we follow Hail and Leuz (2006), (2009) in using the median of the estimates from the four models as our measure of the cost of equity to mitigate the effect of measurement errors associated with one particular model. Since most listed firms in the U.S. market publish their annual reports within 4 months after the fiscal year-end, we estimate the cost of equity at the end of month +4 after the fiscal year-end. To correct for partial year discounting, we discount the month +4 price to the beginning of the fiscal year, using the corresponding imputed cost of equity to account for the price appreciation from month 0 to month +4. Table 2 presents the descriptive statistics and the correlation matrix for the estimated excess cost of equity (or the implied equity risk premium). The estimated excess cost of equity is calculated as the implied cost of equity minus the risk-free rate, R f, as measured by the yield on 10-year Treasury bonds. The four estimates are denoted as follows: R GLS is the estimate from GLS s (2001) model, R CT is from CT s (2001) model, R MPEG is from Easton s (2004) model using modified price/earnings to growth (PEG) ratios, and R OJ is from OJ s (2005) model as implemented by Gode and Mohanram (2003). The median of these four estimates is our estimate of the cost of equity for each firm in each year and is denoted as R MED. The dependent variable used in this paper is R MED R f. As seen in Panel A of Table 2, R MPEG generates the highest average estimated excess cost of equity, with a mean of 6.425% and a median of 5.326%, whereas R GLS produces the lowest average estimated excess cost of equity, with a mean

11 Chen, Chen, and Wei 181 TABLE 2 Summary Statistics and Correlations for the Cost of Equity Estimates Table 2 presents the summary statistics and simple correlations for the estimates of the cost of equity. The excess cost of equity is the cost of equity minus the risk-free rate (R f). R GLS, R OJ, R CT, and R MPEG are the estimates of the cost of equity by GLS (2001), CT (2001), OJ (2005) as implemented by Gode and Mohanram (2003), and Easton (2004). R MED is the median of these four estimates. All of the cost of equity measures are estimated at +4 months after the fiscal year-end. R f is the risk-free rate, which is measured as the yield on 10-year Treasury bonds. Panel A gives the summary statistics, and Panel B gives the Pearson (below the diagonal) and the Spearman (above the diagonal) correlations between the estimates of the implied excess cost of equity. The p-values are in parentheses. The sample size is 13,140 for all of the variables. Panel A. Summary Statistics of the Excess Cost of Equity Estimates Percentile Variable Mean Std. Dev. 1% 25% 50% 75% 99% R MED R f (%) R GLS R f (%) R CT R f (%) R OJ R f (%) R MPEG R f (%) Panel B. Correlations between the Excess Cost of Equity Estimates Variable R MED R f (%) R GLS R f (%) R CT R f (%) R OJ R f (%) R MPEG R f (%) R MED R f (%) (0.000) (0.000) (0.000) (0.000) R GLS R f (%) (0.000) (0.000) (0.000) (0.000) R CT R f (%) (0.000) (0.000) (0.000) (0.000) R OJ R f (%) (0.000) (0.000) (0.000) (0.000) R MPEG R f (%) (0.000) (0.000) (0.000) (0.000) of 1.562% and a median of 1.421%. 12 The mean and median of R MED R f are 4.944% and 4.490%, respectively. The interquartile range of R MED R f is 2.971% (= 6.162% 3.191%), which indicates a narrow dispersion of the estimates of the cost of equity. Panel B of Table 2 presents the pairwise Pearson (below the diagonal) and the Spearman (above the diagonal) correlations between the four estimates. The four estimates are positively correlated with each other. The Pearson correlations are all above The lowest Pearson correlation is observed between R MPEG R f and R GLS R f, with a value of about (the Spearman correlation is 0.378). The median estimate of the excess cost of equity (R MED R f ) is most correlated with R OJ R f, with Pearson and Spearman correlation coefficients of over Overall, our ex ante cost of equity estimates are comparable to those of previous studies Since the GLS (2001) model generates the cost of equity estimates that appear to have lower means and standard deviations than those estimated by the other three models, we conduct a robustness test using the median value of the cost of equity estimates from the other three models as the dependent variable. The results remain similar. 13 For example, the mean (median) value of R GLS R f in the sample period from 1983 to 1993 in Botosan and Plumlee (2005) is 1.00% (1.00%), and the mean (median) value of R OJ R f is 6.60% (6.10%). Guay et al. (2003) report only statistics for the cost of equity, but not for the risk premium. Based on the risk-free rate in our sample, the implied mean (median) values from 1990 to 2004 in their sample are around 3.63% (3.70%), 6.45% (6.49%), and 6.78% (6.76%) for the CT, OJ, and PEG models, respectively.

12 182 Journal of Financial and Quantitative Analysis C. Model Specification and Control Variables We investigate the effect of shareholder rights on the estimated excess cost of equity by estimating the following regression model: (1) R MEDi,t R f,t = α + βg INDEX i,t 1 + γ 1 BETA i,t + γ 2 IDRISK i,t + γ 3 log(mv i,t 1 )+γ 4 log(bm i,t 1 )+γ 5 LEV i,t 1 + γ 6 MMT i,t + γ 7 FERR i,t + γ 8 FLTG i,t + TIME FIXED EFFECTS + INDUSTRY FIXED EFFECTS + ε i,t, where G INDEX is the G-index from GIM (2003). We match the cost of equity with the G-index value in the most recent data published by the IRRC before the estimation month to ensure that the information about the G-index is available to the market. 14 Our main interest is the regression coefficient on the G-index (i.e., β). Hypothesis 1 predicts β to be positive. 15 We include several control variables that may affect the cost of equity. We first control the market beta (BETA) and idiosyncratic risk (IDRISK) based on historical returns. Specifically, the market beta is estimated by regressing the previous 60 monthly individual stock returns (with at least 24 monthly returns) on the contemporaneous and the lagged market returns (Fama and French (1992)). BETA is the sum of the coefficients on the current and the lagged market returns. Idiosyncratic risk (IDRISK) is measured by the standard deviation of the residuals in the market model regression. We expect that γ 1 > 0 and γ 2 > 0. Fama and French (1992) find that stock returns are negatively correlated with firm size and are positively correlated with book-to-market equity. Firm size (log(mv)) is measured as the logarithm of the market value of common equity. Book-to-market equity (log(bm)) is calculated as the logarithm of the ratio of the 14 The IRRC published the data on corporate charters and provisions in September 1990, July 1993, July 1995, February 1998, December 1999, February 2002, and January For example, the cost of equity of a firm with a fiscal year that ends in June 1995 is estimated using the October 1995 price and earnings forecast and is matched with the G-index based on the July 1995 IRRC data. 15 The out-of-pocket monitoring cost and investor recognition arguments (i.e., Lombardo and Pagano (2002a), Merton (1987)) suggest that shareholder rights directly affect the cost of equity and therefore should be directly included in equation (1). In contrast, the analytical models in Albuquerque and Wang (2008) and Garmaise and Liu (2005) suggest that the effect of shareholder rights on the cost of equity can be captured by a fully specified, forward-looking factor beta (or betas if they are generalized to a model with multiple risk factors) and the marketwide risk premium. To empirically test these latter models, it would be more appropriate to directly test the effect of shareholder rights on the forward-looking beta, if such a beta were available. However, it is difficult, if not impossible, to obtain a perfect measure of such a forward-looking beta. Beta estimates based on historical returns are in general not good measures of forward-looking betas (Fama and French (1997)). Thus, the inclusion of the G-index in equation (1) as a proxy for the shareholder rights effect can be justified on the basis that the G-index may help to capture the imperfect measure of a forward-looking beta when an estimated beta is used as a proxy for the forward-looking beta.

13 Chen, Chen, and Wei 183 book value of equity to the market value of equity. Since Modigliani and Miller (1958) argue that the cost of equity should increase as leverage increases, we include a leverage ratio (LEV), which is long-term debt divided by total assets at the end of the previous fiscal year. We expect that γ 3 < 0, γ 4 > 0, and γ 5 > 0. Following Guay et al. (2003), we include price momentum (MMT) to mitigate biases in the cost of equity estimates driven by analysts sluggishness with respect to information in past stock returns. We define MMT as the logarithm of 1 + the compounded returns over the previous 12 months. We expect that γ 6 < 0. Following Hail and Leuz (2006), we include analysts forecast errors (FERR) to control for the potential effect of analysts biases on the cost of equity estimates. We define analysts forecast errors as the actual earnings minus the consensus forecasts for the forthcoming fiscal year, scaled by stock prices. We expect that γ 7 < 0. We also include analysts long-term earnings growth (FLTG) to control for a potential bias in the cost of equity estimate. This control is necessary because the bias will affect the various approaches we use to estimate the cost of equity. First, GLS (2001) point out that if the fixed reverting period is too short (long) for growth (mature) firms, their GLS estimates for growth (mature) firms will be underestimated (overestimated). Including FLTG helps to control for this estimation bias. Second, the CT (2001) estimates may be biased upward because CT find an optimism bias in analysts forecasts over the next 5 years. The analysts optimism on long-term growth will in general bias the cost of equity estimates. Including FLTG in the regressions helps to control for the potential effect of analysts long-term forecast optimism. Third, the implementation of the OJ (2005) model by Gode and Mohanram (2003) assumes that short-term earnings growth decays asymptotically to a constant rate in the long run and that the rate of decay also depends on this constant rate. As Gode and Mohanram point out, the OJ model overstates future earnings and, in turn, the cost of equity estimates for firms whose decay in the growth rate exceeds the constant rate, and vice versa. Firms with higher short-term growth rates tend to have higher decay rates for growth, which suggests that the OJ model tends to overestimate the cost of equity for these firms. Including FLTG in the regressions helps to control for this bias. Since the effect of growth is uncertain, we make no prediction for the regression coefficient of γ 8. Finally, we include year and industry dummy variables based on the industry classification suggested by Fama and French (1997) to control for time-fixed effects and industry-fixed effects. Equation (1) is estimated using the ordinary least squares (OLS) method and pooling all firm-year observations. Since a firm enters the sample more than once, we use standard errors clustered by firm. Table 3 presents the summary statistics for the G-index and the control variables. The mean of the G-index in the final sample is about 9.18, and the cross-sectional variation is not large, with a standard deviation of The median value of BETA is close to 1, but the mean value is greater than 1, which indicates that our sample firms on average are exposed to an above-market systematic risk. The median market value of equity is US$1.35 billion, which indicates that our sample firms are relatively large. This is not surprising, since IRRC tends to cover large firms.

14 184 Journal of Financial and Quantitative Analysis TABLE 3 Descriptive Statistics for Shareholder Rights and Control Variables Table 3 presents the descriptive statistics for the shareholder rights and the control variables used in this study. G INDEX is the Governance Index of GIM (2003). E INDEX is the entrenchment index from Bebchuk et al. (2009). BETA is calculated by regressing individual stock returns on the current and the lagged market returns in the previous 60 months (at least 24 months), and IDRISK is the standard deviation of the residuals from the regression. log(mv) is the natural logarithm of the equity market value at the end of the previous fiscal year, and log(bm) is the natural logarithm of the book-to-market ratio as measured at the end of the previous fiscal year. Leverage (LEV) is the ratio of long-term debt to total assets. MMT is the natural logarithm of 1 + the compounded return over the previous 12 months. FERR (in percents) is the analysts earnings forecast error for the forthcoming fiscal year, which is measured as actual earnings minus the analysts earnings forecast and scaled by the stock price. If the actual earnings from IBES are missing, then EPS in Compustat is used instead. FLTG is the analysts forecast of the long-term earnings growth rate. If it is missing in IBES, then the growth rate implied in the analysts earnings forecasts is used. The sample size is 13,140 for all of the variables. Percentile Variable Mean Std. Dev 1% 25% 50% 75% 99% G INDEX E INDEX BETA IDRISK log(mv) log(bm) MMT LEV FERR (%) FLTG (%) IV. The Effect of Shareholder Rights on the Cost of Equity A. Regressions of the Implied Cost of Equity on Shareholder Rights Table 4 presents the results of regression (1). The focus of this table is on the G-index regression coefficient. Model 1 is our baseline specification throughout the paper. The point estimate of the G-index coefficient is and is significant at the 1% level (t-stat. = 3.02). 16 Model 2 gives the results of the Fama- MacBeth (1973) regression. The coefficient estimate of the G-index (0.028) is slightly smaller than that in Model 1, but it is still significant at the 1% level (t-stat. = 3.43). Model 3 includes only two extreme portfolios: the democracy portfolio, which includes firms with the strongest shareholder rights (G INDEX 5); and the dictatorship portfolio, which includes firms with the weakest shareholder rights (G INDEX 14). We define the dictatorship dummy variable (denoted as DICTATOR) as 1 if a firm belongs to the dictatorship portfolio, and 0 otherwise. The coefficient on this DICTATOR dummy variable is and is significant at the 10% level (t-stat. = 1.75). Assuming that the G-index is equal to the previously announced value may introduce measurement errors. To address this issue, Model 4 includes only the first observations after each G-index is announced. The coefficient estimate of the 16 If the estimated beta is a good proxy for the forward-looking beta, we should find that the coefficient on BETA increases when the G-index is excluded from the regression. However, we find that this is not case, suggesting that the beta estimated from historical data may be too noisy and may not be a good proxy for the forward-looking beta, which is consistent with the finding by Barry and Brown (1985) and Fama and French (1997).

15 Chen, Chen, and Wei 185 TABLE 4 Regression Analysis of the Association between Shareholder Rights and the Cost of Equity Table 4 presents the results of the regressions of the implied cost of equity on G INDEX: R MEDi,t R f,t = α + βg INDEX i,t 1 + γ 1BETA i,t + γ 2IDRISK i,t + γ 3log(MV i,t 1 )+γ 4log(BM i,t 1 ) + γ 5LEV i,t 1 + γ 6MMT i,t + γ 7FERR i,t + γ 8FLTG i,t + TIME FIXED EFFECTS + INDUSTRY FIXED EFFECTS + ε i,t, where R MED (in percents) is the median of the cost of equity estimates by GLS (2001), OJ (2005) as implemented by Gode and Mohanram (2003), CT (2001), and the modified PEG model of Easton (2004). R f is the risk-free rate, which is measured as the yield on 10-year Treasury bonds. G INDEX is the Governance Index of GIM (2003). DICTATOR is a dummy variable that equals 1 if G INDEX 14 and 0 if G INDEX 5. E INDEX is the entrenchment index from Bebchuk et al. (2009). BETA is calculated by regressing individual stock returns on the current and the lagged market returns in the previous 60 months (at least 24 months), and IDRISK is the standard deviation of the residuals from the regression. log(mv) is the natural logarithm of the equity market value at the end of the previous fiscal year, and log(bm) is the natural logarithm of the book-to-market ratio as measured at the end of the previous fiscal year. MMT is the natural logarithm of 1 + the compounded return over the previous 12 months. Leverage (LEV) is the ratio of long-term debt to total assets. FERR (in percents) is the analysts earnings forecast error for the forthcoming fiscal year, which is measured as actual earnings minus the analysts earnings forecast and scaled by the stock price. If the actual earnings in IBES are missing, then EPS in Compustat is used instead. FLTG is the analysts forecast of the long-term earnings growth rate. If it is missing from IBES, then the growth rate implied in the analysts earnings forecasts is used. The t-statistics are in parentheses and are based on standard errors clustered by firm, and *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels (2-tailed), respectively. Pooled OLS Pooled Regression Using Using the First Regression Fama- Only the Dictator Observation Firm-Fixed Using Pooled OLS MacBeth and Democracy after Each Effect Entrenchment Regression Regression Portfolios Announcement Regression Index Independent Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 G INDEX 0.034*** 0.028*** 0.034*** 0.039* (3.02) (3.43) (2.70) (1.66) DICTATOR 0.253* (1.75) E INDEX 0.037* (1.67) BETA ** 0.232** 0.133** (1.53) (1.92) (2.14) (2.43) (0.19) (1.55) IDRISK 3.888*** 9.127*** *** *** (3.48) (3.84) (1.01) (2.88) ( 0.18) (3.29) log(mv) 0.201*** *** 0.202*** 0.970*** 0.194*** ( 7.29) ( 1.69) ( 3.24) ( 6.71) ( 14.59) ( 7.10) log(bm) 0.702*** 0.647*** 0.689*** 0.670*** *** (11.85) (9.84) (4.92) (10.38) (1.08) (11.89) LEV 2.592*** 2.234*** 1.983*** 2.504*** 2.058*** 2.613*** (11.90) (10.36) (4.26) (10.45) (6.84) (11.97) MMT 1.867*** 1.970*** 1.815*** 1.976*** 2.185*** 1.865*** ( 25.81) ( 10.23) ( 11.12) ( 19.43) ( 32.42) ( 25.72) FERR 0.131*** 0.119*** 0.135*** 0.145*** 0.094*** 0.131*** ( 13.23) ( 9.19) ( 5.11) ( 10.92) ( 8.20) ( 13.21) FLTG 0.055*** 0.047*** 0.057*** 0.057*** 0.127*** 0.054*** (6.97) (4.78) (3.18) (6.52) (17.16) (6.83) Industry-fixed Yes Yes Yes Yes No Yes effects Year-fixed Yes No Yes Yes Yes Yes effects Firm-fixed No No No No Yes No effects Adjusted R N 13,140 13,140 2,011 7,436 13,140 13,140