Dividend policy, dividend initiations, and governance. Micah S. Officer *

Transcription

1 Dividend policy, dividend initiations, and governance Micah S. Officer * Marshall School of Business Department of Finance and Business Economics University of Southern California Los Angeles, CA Phone: officer@marshall.usc.edu This draft: October 5, 2006 Abstract: Dividend policy can either be an outcome of strong governance or a substitute for weak governance. This paper provides evidence that dividend policy is a substitute for weak internal and external governance by focusing on a sample of firms that should pay dividends. Specifically, predicted dividend payers with weak governance are significantly more likely to pay dividends than are predicted dividend payers with strong governance. Firms with weak governance also have significantly higher dividend initiation announcement abnormal returns than other firms, consistent with the notion that dividend policy is a substitute for other governance attributes and that the market prices the decrease in agency costs resulting from the initiation of dividends. * I thank Harry DeAngelo, Linda DeAngelo, Ehud Kamar, Harold Mulherin, Mike Stegemoller, Ralph Walkling, and Mark Weinstein for comments, and Jim Linck for providing some of the data.

2 Introduction While finance academics have long wondered why firms pay dividends when cash distributions in the form of dividends are tax disadvantaged relative to retention or stock repurchases (e.g. Black (1976)), recent theoretical and empirical work significantly expands our understanding of whether, when, and why firms pay dividends (Fama and French (2001), DeAngelo, DeAngelo, and Skinner (2004), DeAngelo and DeAngelo (2006), and DeAngelo, DeAngelo, and Stulz (2006)). The interaction of dividend policy and governance is central to the debate about the agency costs of free-cash-flow (Easterbrook (1984) and Jensen (1986)). In particular, Easterbrook (1984) argues that a policy of paying dividends reduces agency costs by improving the monitoring and risk-taking incentives of managers. While the initiation of a policy of paying dividends should reduce the agency costs of free-cash-flow ex post, the relation between ex ante agency problems and the decision to pay dividends is not as clear. La Porta et al (2000) discuss two models of the relation between ex ante agency problems and dividend policy: the outcome model and the substitute model. 1 In the outcome model, the payment of dividends is the result of effective governance well-governed firms pay dividends because strong governance makes expropriation from shareholders (the worst manifestation of the agency problems of free-cash-flow) more difficult and shareholders successfully pressure managers to distribute excess cash. 2 In the substitute model, the payment of dividends replaces other governance characteristics in the portfolio of policies that firms employ to convince shareholders that they will not be expropriated. 3 The substitute model predicts that poorly- 1 La Porta et al (2000) discuss dividend policy in the context of shareholder protection in various legal regimes around the world. In this paper I take that discussion and apply it to differences in governance characteristics between firms in the same legal regime (United States). 2 Tse (2004) questions the logic of the relation between agency costs and dividend policy in the outcome model if well-governed firms are more likely to pay dividends, then shareholders shouldn t need to rely on the payment of dividends to reduce the agency costs of free-cash-flow because such costs should already be low for well-governed firms. 3 Also see Rozeff (1982). 1

3 governed firms make dividend payments because such firms need an alternate means of establishing a reputation for acting in the interests of shareholders if they intend to raise capital from public markets in the future, and therefore a policy of paying dividends is the most valuable at the margin to firms with agency problems. I identify a sample of firms with fundamentals that suggest that the firms should be dividend payers, and test whether governance characteristics affect the decision to pay dividends in that sample. This sample-selection strategy offers several advantages over the empirical methods employed elsewhere in the literature. First, the outcome model predicts that wellgoverned firms are more likely to pay dividends only when firm fundamentals (size, growth, earned capital, etc.) support the adoption of a policy of paying dividends. In other words, the outcome model doesn t stand a chance against the alternative (the substitute model) unless one conditions first on the appropriateness of adopting a policy of paying dividends. Second, a sample of firms that are predicted dividend payers is likely to contain firms that, all else equal, have higher-than-average agency costs of free cash flow. Therefore, this sample should provide a strong empirical test of whether characteristics associated with strong governance are positively (the outcome model) or negatively (the substitute model) correlated with a policy of paying dividends, because the higher-than-average agency costs of free cash flow give shareholders ample motivation to insist on payouts if the governance structure allows (the outcome model) and managers ample motivation to voluntarily offer payouts to compensate for otherwise weak governance structures (the substitute model). I identify firms that should pay dividends (predicted dividend payers) using the models in Fama and French (2001) and DeAngelo, DeAngelo, and Stulz (2006), and find that predicted dividend payers with characteristics associated with weak governance and managerial entrenchment are significantly more likely to pay dividends than are predicted dividend payers with strong governance. For example, firms with large boards, CEO/Chairman duality, and low ownership by insiders and institutional investors are significantly more likely to pay regular cash 2

4 dividends to shareholders. These results are consistent with the substitute model in La Porta et al (2000), which predicts that firms use dividend policy as a substitute for otherwise weak governance. I offer further support for the substitute model by demonstrating that firms with weak governance have significantly more positive dividend initiation announcement returns than do firms with characteristics consistent with strong governance. This suggests that the market prices the reduction in agency costs associated with the initiation of dividends by firms with weak governance, consistent with the hypothesis that dividend policy is one component of the firm s governance and bonding policies and therefore substitutable with other governance policies. Endogeneity is certainly a concern in interpreting my results. However, the substitute model in La Porta et al (2000) suggests that dividend policy and other governance characteristics are endogenous choices that firms make as part of an equilibrium monitoring/bonding package to reduce agency costs. While my empirical models are structured as tests of whether governance proxies (implicitly exogenous in the regressions) affect dividend policy choices, the substitute model does not suggest that any of these policy choices are truly exogenous. Therefore, the most that can be concluded from my results (and the other results in this literature), is that dividend policy choice and other policy choices that affect governance quality are negatively correlated. Firms that choose governance policies that are associated with entrenchment (such as having many anti-takeover provision insulating managers from hostile takeovers) are empirically more likely to pay dividends, but it is impossible to conclude that one of these policy choices causes the other. As discussed in Rozeff (1982), the data is most consistent with the interpretation that dividend policy decisions are made as part of an endogenous package of governance choices that optimally reduce the agency costs generated by the separation of ownership and control. Recent empirical papers in this literature have offered consistent support for the substitute model in La Porta et al (2000) (although few papers use that language). For example, Fenn and Liang (2001) report that firms with low managerial stock-option holdings (and therefore lowerpowered incentives) have significantly higher dividend and total (including repurchases) payout 3

5 ratios, although this result could be due to the lack of dividend protection afforded by most executive stock option contracts (Lambert, Lanen, and Larker (1989)). Hu and Kumar (2004) find that the likelihood and level of dividend payouts is increasing when factors such as managerial and outside blockholder ownership, CEO compensation policy, and board independence indicate a high likelihood of managerial entrenchment and high agency costs. John and Knyazeva (2006) report that dividend and total payouts (the sum of dividends and repurchases) are significantly more likely when internal and external governance measures indicate weak governance. Pan (2006) shows a similar association between the propensity to pay dividends and measures of managerial entrenchment based on indices of anti-takeover charter provisions. Several earlier studies report complementary evidence. Rozeff (1982) reports that dividend payout ratios are significantly negatively associated with insider ownership and positively associated with the dispersion of outside ownership. Essentially identical results are found in a simultaneous-equations framework by Jensen, Solberg, and Zorn (1992). Gugler (2003) examines Austrian firms and finds that state-controlled firms, for which agency problems are likely to be severe, have higher target dividend payout ratios than other types of firms, and are also the most reluctant to cut dividends when such cuts are warranted. On the other hand, Renneboog and Szilagyi (2006) report that firms with strong shareholders appear to force higher payouts in Dutch firms, and Michaely and Roberts (2006) conclude that strong governance encourages higher and more consistent payouts using data on private firms in the UK. One weakness with the empirical tests in the recent literature (e.g. Hu and Kumar (2004), John and Knyazeva (2006), and Pan (2006)), however, is that none of these authors focus on samples of firms that should pay dividends. As described above, this weakness biases tests in favor of the substitute model. While these studies include firm fundamentals as control variables in regressions of dividend policy choice on governance characteristics, holding fundamentals constant in a regression using all available firm-years is very different from focusing on a sample of firms that should pay dividends. Including firm fundamentals as control variables essentially 4

6 involves asking the following question: if two firms have the same fundamentals, does governance affect the decision to pay dividends? But as a large fraction of the full sample of firmyears used in most studies is comprised of firms that, according to fundamentals (size, growth, retained earnings, and so on) should not be dividend payers, 4 controlling for fundamentals can result in a test of whether governance matters for dividend policy for two firms that are equallyunlikely to be dividend payers. A further omission from this literature is that none of the aforementioned studies examine the relation between governance characteristics and the wealth effects of initiating a policy of paying dividends. The substitute model predicts that dividends are used as a substitute for other governance devices in order to reduce agency costs, and therefore that the announcement that a firm with weak governance will begin paying dividends for the first time should be associated with positive returns as the market prices the reduction in agency costs into the firm s stock price. In other words, initiating dividends offers the greatest reduction in agency costs, and consequently the largest initiation announcement stock price reaction, for firms with weak governance in the substitute model. The contribution of this paper is to address these two weaknesses and omissions. First, by examining firms that should be paying dividends (based on fundamentals) and exploring the relation between whether those firms do pay or not and proxies for internal (board size and composition) and external (takeover defenses, ownership by blockholders and activist investors) governance characteristics, my research design provides a more robust test of the substitute model by, if anything, biasing the test in favor of the outcome hypothesis. Second, I provide complementary evidence by examining the relation between governance characteristics and dividend initiation announcement returns if a policy of paying dividends is chosen as a substitute for other governance characteristics, then the reduction in agency costs should be 4 For example, less than one-third of the firms in the sample in DeAngelo, DeAngelo, and Stulz (2006) are dividend payers, and the empirical models in that paper have considerable success in predicting the propensity to pay. 5

7 priced by the market at the margin. My results strongly support the substitute model and are consistent with the hypothesis that a policy of paying dividends is used as a substitute for other governance devices by firms with weak governance and high agency costs, and that the market prices an anticipated reduction in agency costs around dividend initiation announcement dates. The results in this paper are related to other findings in the literature. Harford, Mansi, and Maxwell (2005) report that firms with weak governance hold substantially lower cash reserves than do firms with strong governance. This is consistent with the notion that firms with weak governance are more likely to pay dividends (thereby reducing cash balances). 5 Gillan, Hartzell, and Starks (2006) find evidence that governance structures are substitutes for one another for example, firms with more independent boards also tend to have more restrictive anti-takeover provisions. Such substitutability between governance structures is consistent with the argument that dividend policy and other governance attributes also serve as substitutes. The remainder of this paper is organized as follows. In Section I, I develop the hypotheses and empirical predictions. Section II describes the relation between the propensity to pay dividends for firms that should be dividend payers and governance characteristics. Section III examines the relation between governance measures and abnormal returns around dividend initiation announcements, and Section IV discusses the implications of the findings and offers concluding remarks. I. Hypothesis development A. Two models for the relation between agency costs and dividend policy As outlined in La Porta et al (2000), there are two potential models for the relation between agency costs and dividend policy. Although each is discussed in detail in La Porta et al, the discussion in that paper centers on a cross-country (and legal system) framework. Therefore, I 5 Harford, Mansi, and Maxwell (2005) conclude, however, that firms with weak governance dissipate cash through acquisition activity rather than by returning it to shareholders. 6

8 adapt each of the La Porta et al models to a setting where firms are in a common legal system (United States) but have attributes that offer shareholders a strong voice in the governance of the corporation ( strong governance ) or a weak voice ( weak governance ): A.1.The outcome model Firms with strong governance are those with, for example, small boards that are more difficult for insiders to manipulate (Jensen (1993) and Yermack (1996)), boards that are dominated by non-executives or outsiders (Weisbach (1988) and Brickley, Coles, and Terry (1994)), blocks of stock controlled by activist outsiders (Barclay and Holderness (1991) and Del Guercio and Hawkins (1999)), and few anti-takeover provisions (Gompers, Iishi, and Metrick (2003) and Bebchuk, Cohen, and Ferrell (2005)). Managers of companies with strong governance systems will find it more costly to expropriate wealth from outside shareholders, raising the relative attractiveness of paying out cash to stockholders. In such firms, shareholder insistence on the distribution of excess cash is less likely to fall on deaf ears than in firms with attributes associated with managerial entrenchment or weak governance. The correct dividend policy is the outcome of the governance regime in this view because managers of firms with good governance are more likely to act in the interests of shareholders and pursue value-maximizing policies, such as the payment of dividends when the firm s fundamentals warrant such a policy, than are managers of firms with weak governance. In this model, strong governance and a policy of paying dividends are complements. A.2.The substitute model 6 This view of the relation between agency costs and dividend policy relies on the assumption that firms and managers need to establish a reputation with public capital markets for 6 This model is derived from the discussion in La Porta et al (2000), but is also consistent with the predictions of the agency model in Allen, Bernado, and Welch (2000) (specifically, prediction #1 on p.2519 of that paper), the discussion in Myers (2000), and the agency model in Nielsen (2005). 7

9 not expropriating wealth from shareholders (whether actual cash transfers or over-consumption of perquisites) because the firm will need to tap such markets to raise new external capital in the future (Easterbrook (1984)) or because the managers reputation is important for their careers. Dividend policy is one part of the optimal portfolio of choices that firms make as bonding commitments to shareholders to reduce agency costs (Rozeff (1982)). A policy of paying dividends reduces the agency costs of free-cash-flow, and such a policy choice is therefore most valuable at the margin to firms with the highest agency costs. Managers of firms with weak governance, such as those with large boards, few outside blockholders, low levels of ownership by managers (Jensen and Meckling (1976)), and charter provisions making control-changes costly, are most likely to prefer a policy of paying dividends, even if firm fundamentals do not warrant such a policy, because dividend policy is a substitute for other forms of governance in the maintenance of a good reputation with external capital markets. B. Hypotheses The outcome and substitute models have opposite predictions about the relation between governance and dividend policy. Assuming for the moment that governance can be measured empirically, 7 the testable hypotheses are: H1a (outcome): Firms with strong (weak) governance are more (less) likely to pay dividends when firm fundamentals suggest that such a policy is warranted. H1b (substitute): Firms with weak (strong) governance are more (less) likely to pay dividends when firm fundamentals suggest that such a policy is warranted. 7 I will return to this issue in Section II, and use a variety of governance proxies that are commonly employed in the empirical governance literature. 8

10 The outcome and substitute models also offer differing predictions about the relation between dividend initiation announcement returns and governance characteristics. In the outcome model, firms with strong governance are predicted to be the most likely to initiate a policy of paying dividends, and yet the value of that policy (reducing the agency costs of free cash flow, for example) for such firms is lower than for firms with weak governance. 8 On the other hand, the substitute model predicts that dividends are used as a substitute for other governance devices in order to reduce agency costs, and therefore that the market will price the anticipated reduction in agency costs into the firm s stock price upon the announcement that a firm with weak governance will begin paying dividends for the first time (initiations). In other words, initiating dividends offers the greatest reduction in agency costs for firms with weak governance in the substitute model, and if the reduction in agency costs is reflected in returns around dividend initiation dates then firms with weak governance will have significantly more positive initiation announcement returns than will firms with strong governance. This suggests the following testable hypotheses: H2a (outcome): There will be no significant difference in abnormal announcement returns around dividend initiation announcements conditional on governance characteristics. H2b (substitute): Firms with weak governance will have significantly more positive dividend initiation announcement returns than will firms with strong governance. 8 See also Tse (2004). 9

11 II. Dividend payers and governance characteristics A. Identifying firms that should be dividend payers The first set of hypotheses (H1a and H1b) predicts differences in the proportion of dividend payers amongst those firms that should pay cash out to stockholders. I identify those firms that should be dividend payers using firm fundamentals that are associated with the propensity to pay. Specifically, I rely on the empirical models employed in Fama and French (2001) and DeAngelo, DeAngelo, and Stulz (2006) to identify firm-years in which the firm should be a dividend payer, and then examine the propensity for those firms to pay dividends conditional on various governance characteristics. Table 1 replicates regressions D2 and D4 from Table 3 in DeAngelo, DeAngelo, and Stulz (2006), which are extensions of the payer-prediction regressions in Table 5 of Fama and French (2001). The logit regressions use the Fama and MacBeth (1973) method to explain the probability that a firm pays dividends with fundamental characteristics such as size, growth, profitability, and the ratio of retained earnings to total equity (the earned/contributed capital mix). The latter explanatory variable is used solely in DeAngelo, DeAngelo, and Stulz (2006) (as a proxy for the firm s stage in its capital infusion/self-financing lifecycle), while the other variables are common to both papers cited above. The sample used for the logit regressions in Table 1 is the same as that described in DeAngelo, DeAngelo, and Stulz (2006). Specifically, the sample contains all firm-year observations for all US-incorporated industrial (non-financial and non-utility) firms that are both in the Compustat database and in the Center for Research in Security Prices (CRSP) database with publicly-traded equity (NYSE, Nasdaq, or Amex) with share code 10 or 11. The sample is from , with the beginning of the sample period dictated by the fact that CRSP expands its coverage to include Nasdaq firms in Each firm-year observation is required to have non-missing values for dividends and income before extraordinary items and positive total common equity. 10

12 The explanatory variables in Table 1 are identical to those described in DeAngelo, DeAngelo, and Stulz (2006) with one exception: the size percentile measures are based on the full sample rather than NYSE break-points. Other than that, I employ identical measures of earned/contributed capital (RE/TE), common equity/total capital (TE/TA), profitability (ROA), and sales or asset growth (Growth) as described in DeAngelo, DeAngelo, and Stulz (2006). The regressions are estimated separately for each sample year, and the coefficients in Table 1 are the average of the 32 annual coefficients. The standard errors used to assess statistical significance are the standard deviations of the annual coefficients scaled by the square root of 32. While my sample period is slightly longer than theirs, the logit regression results reported in Table 1 are practically identical to those in DeAngelo, DeAngelo, and Stulz (2006). The propensity to pay dividends is significantly positively related to the earned/contributed capital mix, profitability, and size (whether measured using equity market capitalization (D2) or total assets (D4)), and negatively related to growth (whether measured as the growth rate of assets (D4) or sales (D2)). The R 2 s reported in Table 1 are the average R 2 for each model, and are 36% for model D2 and 37% for D4. While the regressions are so similar that any choice between them is arbitrary, as I use the regression for predictive purposes I rely on model D4 as it has a slightly higher R 2. Figure 1 shows the dividend-payer lifecycle predictions derived from model D4 in Table 1. The figure plots the average predicted probability of paying dividends (solid line) and the fraction of firms paying dividends (dashed line) for each year in the sample life of firms that have at least five years of data in the sample. 9 The years in sample in the figure is essentially an event year, with year 1 being the first year that each firm appears in the sample. The average predicted probability of paying dividends increases monotonically over the average firm s lifecycle, rising from a 36% probability of being a dividend payer in the first year that a firm 9 While sample life is not the same as firm age, the number of years in the sample (which is taken directly from the Compustat database) should be very close to the number of years as a public company for most firms that live at least five years and do not have a substantial number of missing observations. 11

13 appears in the sample to a 50% probability in the 22 nd year. The actual fraction of payers in the sample rises at a faster rate, but again reflects the lifecycle behavior described in Fama and French (2001), Grullon, Michaely, and Swaminathan (2002), and DeAngelo and DeAngelo (2006) only 31% of firms are dividend payers in their first sample year, and that fraction passes 50% in the 13 th year of sample life. As Figure 1 demonstrates, the propensity to pay dividends is reasonably well captured by models, such as those in Table 1, that take account of the fact that as a firm gets older its size, earned capital, and profitability increase (and sales or asset growth rate decreases) to the point where the payment of dividends appears to be an optimal policy. I use the predicted probabilities from model D4 in Table 1 to identify a sample of firmyears in which firms should be paying dividends. Specifically, dividends should be paid in every firm-year in which the predicted probability of being a dividend payer has exceeded 50% for two consecutive years. Several examples will help highlight the sample derived from this selection procedure. Microsoft Corp enters the sample in 1987 (the year after Microsoft s IPO, allowing one year for the calculation of Microsoft s asset growth rate), at which time model D4 predicts that its probability of being a dividend payer is 79%. The predicted probability of Microsoft being a payer stays above the 50% threshold for all years until 2000, at which point it drops to 49% and stays below 50% for three consecutive years until 2003 (and stays above that threshold until the end of the sample). Therefore, Microsoft is considered to be a firm that should be paying dividends from 1988 to 1999 and again in Microsoft initiates dividend payments in Harley Davidson Inc. enters the sample in 1987 (again, the year after its IPO), at which time its predicted probability of being a dividend payer is 46%. Harley Davidson s predicted probability of making dividend payments rises to 56% in the next year (1988), and stay above 50% for the following 17 years (never dropping below 65%). Therefore, Harley Davidson should pay dividends under the above definition from 1989 through Harley Davidson initiates dividend payments in 1993 and does not omit a dividend for the remainder of the sample years. While the lifecycle for every firm in the sample obviously does not follow this clean pattern, the results 12

14 depicted in Figure 1 illustrate that model D4 does a reasonable job of capturing the average lifecycle pattern of dividend policy. B. Governance characteristics Governance is an extremely difficult concept to capture empirically. However, the literature on governance has blossomed recently, potentially because governance appears to be much more important to investors following the meltdowns of Enron, WorldCom, Adelphia, and so on, but also potentially because several data sources (such as the IRRC data used by Gompers, Iishi, and Metrick (2003)) have recently become available to researchers. In this paper I take as broad and conservative approach as I can, sampling from much of the recent governance literature and choosing variables that capture different aspects of governance. Whilst none of these measures is individually perfect, if the majority of evidence points in one direction or the other then reasonable conclusions can be drawn about the relation between dividend policy and governance. I divide the governance characteristics into internal and external measures. Both internal and external measures of governance capture characteristics that could be expected to entrench managers and isolate them from shareholder discipline ( weak governance) or make managers more responsive to their shareholders ( strong governance). In other words, these measures are empirical proxies for the degree of the agency problem at publicly traded corporations. The measures of internal governance are essentially measures of board structure and ownership by insiders (CEO and executive directors), and are largely taken from the sample in Linck, Netter, and Yang (2006). 10 Linck, Netter, and Yang use a screened sample of all firms in the Compact Disclosure database from 1990 to 2004, and the details of the data selection screens used to generate the sample can be found in that paper. The final sample from Linck, Netter, and 10 I thank Jim Linck for allowing me to use these data. 13

15 Yang (2006) includes over 53,000 firm-year observations representing almost 7,000 firms. One measure of internal governance is not taken from Linck, Netter, and Yang (2006), and that is a measure of board strength. As in Harford and Li (2006), I follow Hermalin and Weisbach (1998) and use CEO tenure as a proxy for the strength or independence of the board. Hermalin and Weisbach (1998) argue that the board becomes less independent vis-à-vis the CEO as the CEO s tenure increases. The measures of board structure and insider ownership used in this paper are: Board size: The number of board members (minimum of 3); Strong board: An indicator variable equal to one if the CEO s tenure (in days) is below the median tenure for CEOs of all firms in the Execucomp database in the given year, and zero otherwise; CEO is Chairman: An indicator variable equal to one if the CEO is also the Chairman of the Board, and zero otherwise; Board is insider dominated: an indicator variable equal to one if the percent of executive directors on board is greater than 50%, and zero otherwise; % ownership by of executive directors (officers): The fraction of the firm s shares owned by executive directors (officers); % ownership by CEO: The fraction of the firm s shares owned by the CEO. The continuous variables (Board size, % ownership by of executive directors (officers), and % ownership by CEO) are expressed in terms of deviations from the median for the given year for all firms in the Linck, Netter, and Yang (2006) sample. Firms are considered to have weak internal governance (and therefore high agency costs) in any given year if board size is greater 14

16 than the median (Jensen (1993) and Yermack (1996)), the board is not strong (Hermalin and Weisbach (1998)), the CEO is also Chairman of the Board (Jensen (1993) and Brickley, Coles, and Jarrell (1997)), the board is insider dominated (Weisbach (1988) and Brickley, Coles, and Terry (1994)), or executive directors or the CEO have low ownership stakes in the firm (Jensen and Meckling (1976)). 11 The measures of external governance are based on external ownership and corporatecharter-provisions data. I consider two measures of external ownership: ownership by institutional investors and by public pension funds. Large institutional investors (including public pension funds) potentially act as external monitors forcing firms in their portfolios to improve decision-making and control, and the incentive to monitor is increasing in the stake held in the firm by these investors. 12 To some extent these two categories overlap, as public pension funds are institutional investors (but not vice versa). However, public pension funds are considered to be some of the most activist institutional monitors (see, for example, Brickley, Lease, and Smith (1988), Smith (1996), Wahal (1996), and Del Guercio and Hawkins (1999)), and so considering these two categories separately may provide insight into whether the type of institutional investor influences monitoring and, in turn, dividend policy. The percent ownership by institutional investors is taken from the sample in Linck, Netter, and Yang (2006), from which data is available at the annual frequency. The percent ownership by public pension funds is derived from CDA/Spectrum (13F) data, and is the percent of the firm s shares owned by the public pension 11 Fama and Jensen (1983) and Morck, Shleifer, and Vishny (1988) recognize that the agency costs of external equity may not be monotonically decreasing in the fraction of the firm that is owned by managers. For example, while low levels of ownership insulate executives from the economic consequences of their decisions (the classic agency problem), high levels of managerial ownership may insulate managers from discipline by the market for corporate control. I take the simple approach of assuming that managers interests are aligned with the interests of their shareholders to a greater extent (i.e. lower agency costs) as ownership by the CEO and executive directors increases, but the results of spline regressions accounting for the effects of different levels of managerial ownership on the propensity to pay dividends are discussed in footnote Brickley, Lease, and Smith (1988) show that institutional investors, especially public pension funds, are more likely to vote against management-sponsored charter-amendment proposals than other types of investors are. Chen, Harford, and Li (2005) argue that institutional blockholders have better monitoring incentives than other investors and present evidence that institutional blockholders respond to those incentives by forcing managers to abandon value-destroying acquisitions. 15

17 funds collectively identified in three recent studies of the impact of public pension fund holdings on stock returns (Cremers and Nair (2005), Dittmar and Mahrt-Smith (2006), and Larker, Richardson, and Tuna (2005)) at the quarterly reporting date closest to the end of the fiscal year. I use data from the Investor Responsibility Research Center (IRRC) database of corporate charter provisions to measure the strength of a firm s anti-takeover provisions. Anti-takeover provisions help shield managers from discipline from the market for corporate control, entrenching managers and potentially weakening governance by reducing managers incentives to act in the best interests of shareholders. Bebchuk, Cohen, and Ferrell (2005) consider an index which equals the sum of six indicator variables that capture whether a firm has a staggered board, limits to amend the bylaws or charter, supermajority-voting provisions, golden parachutes, or a poison pill. Bebchuk, Cohen, and Ferrell define the sum of these six indicator variables as the entrenchment index, which takes values from 0 to 6 and is increasing in the number of antitakeover charter provisions that a firm has. The IRRC data is available for a cross-section of firms in the years 1990, 1993, 1995, 1998, 2000, 2002 and 2004, and the IRRC data contains date ranges for which each observation is valid for each firm. Firm-years are matched to IRRC data by selecting the observation for which the end of the fiscal year falls within the relevant date range. This procedure does generate serial correlation in governance measures for individual firms (as a 1990 observation from the IRRC data, for example, could be matched to the 1990, 1991, and 1992 fiscal years), but as the empirical tests use the Fama and Macbeth (1973) approach this serial correlation should have little effect on my results. Firms are defined as having managerial entrenchment if the Bebchuk, Cohen, and Ferrell (2005) (BCF) entrenchment index is greater than the (pooled time-series and cross-sectional) median of two. The measures of external governance are therefore: % ownership by institutions: The fraction of the firm s shares owned by institutional investors; 16

18 % ownership by public pension funds: The fraction of the firm s shares owned by public pension funds; Managerial entrenchment (BCF definition): An indicator variable equal to one if the BCF entrenchment index is greater than two (the pooled time-series and cross-sectional median), and zero otherwise. The continuous variables (% ownership by institutions and % ownership by public pension funds) are expressed in terms of deviations from the cross-sectional median. The median of % ownership by institutions is computed each year using the firms in the Linck, Netter, and Yang (2006) sample. The median of % ownership by public pension funds is computed for each quarterly reporting date for all firms in the CDA/Spectrum (13F) database. Firms are considered to have weak external governance in any given year if institutions or public pension funds hold low stakes in the firm or if managers are entrenched using the BCF definition. C. Sample and empirical results The sample of predicted dividend payers and matches to the various governance data sources described above is depicted in Table 2. The first column contains the number of predicted dividend payers each year from 1974 to The sample is well distributed over time, with only minor clustering in the late 1970 s and early 1980 s. The second column contains the percentage of the predicted payers that actually do pay dividends in each year. This percentage averages 81.5% over the whole sample, but is markedly higher in the early years of the sample than in the later years. It is possible that this reflects a declining propensity to pay dividends (Fama and French (2001)), but could also be driven by the predictive content of the model 13 The sample in Table 2 starts in 1974 because the regression in Table 1 uses observations from 1973 onwards and it takes two consecutive years of greater than 50% predicted probability of being a dividend payer to enter the sample in Table 2. This implies that the first year that I should observe a predicted dividend payer will be

19 declining over time. On average in the full sample, however, 4 out of 5 firms that are predicted payers do actually pay dividends the main research question in this paper is whether governance characteristics affect the propensity for firms to adopt the predicted financial policy. The remaining four columns in Table 2 show how many firms in each year are matched to the various governance data sources described above. Again, the distribution over time of each series is relatively stable after the point that the first match occurs. Most of the governance measures are available for a substantial fraction of predicted payers from 1990 onwards, although the data from CDA/Spectrum on public pension fund ownership starts considerably earlier (1979) than measures from the other data sources. Table 3 contains the principal empirical results in this section, and demonstrates the empirical relation between the propensity to pay dividends for predicted dividend payers and governance characteristics. The regressions are similar to those in Table 1, with Fama and Macbeth (1973) methods used to analyze annual logit regressions with an indicator variable equal to one if dividends are paid (and zero otherwise) as the dependent variable. The difference between the regressions in Table 1 and those in Table 3 is that in Table 3 the sample is constrained to predicted dividend payers only and the governance variables described above are included as explanatory variables. I include the same explanatory variables as used in Table 1 (RE/TE, TE/TA, ROA, and so on) as control variables because even though all firm-years in the sample for Table 3 have high (>50%) predicted probability of paying dividends, cross-sectional dispersion in firms fundamentals may still affect the decision to adopt the predicted dividend policy. The first three columns introduce the governance variables into the regression in groups: board characteristics in column 1, the anti-takeover-provision-based entrenchment index in column 2, and the ownership variables in column 3. The regressions use all available observations with non-missing data over the longest time span possible. For example, the regressions in column 1 use a total of 5,141 observations for 13 annual logit regressions ( ), with an 18

20 average of 395 firm-specific observations per year. The number of observations used and the time span over which the logits are estimated are similar for the other columns in Table 3, but the loss in sample size (to 3,113 total observations with an annual average of 259) is the most serve in the final column where all governance attributes are included at the same time The regression in column 1 shows that board size and CEO/Chairman duality significantly 14 influence dividend policy in the direction predicted by the substitute model (H1b): firms with larger boards and firms with one individual sharing the title of CEO and Chairman are significantly more likely to pay dividends to shareholders. Jensen (1993) and Yermack (1996) suggest that large boards are easier for the CEO to manipulate, and even more so if the CEO is also Chairman, consistent with the notion that these characteristics are associated with weaker monitoring of insiders and higher agency costs. Consistent with the substitute model, predicted dividend payers with weaker governance or greater agency problems are more likely to actually pay dividends than are predicted payers with strong governance. The same is true in column 2, in which the entrenchment index from Bebchuk, Cohen, and Ferrell (2005) is the only governance variable. The coefficient on the entrenchment index is strongly statistically different from zero and positive (as in Pan (2006)), suggesting that predicted payers with entrenched managers are more likely to pay dividends. The coefficients on the ownership variables in column 3 also support the substitute model (H1b) and offer little support for the outcome model (H1a). Firms in this sample with low ownership (relative to the median) by officers of the firm, the CEO, and institutional investors are significantly more likely to adopt a policy of paying dividends. Executive ownership levels greatly affect the extent to which executives share with stockholders the economic costs and benefits of their actions, thereby influencing the extent of the agency problem (Jensen and Meckling (1976)), and institutional investors potentially provide important external monitoring of 14 The 5% (1%) two-sided cutoff value from the t-distribution with 12 degrees of freedom (there are 13 coefficients averaged to produce each coefficient in Table 3) is approximately 2.18 (3.06). 19

21 executives and the board. Therefore, the ownership evidence is again consistent with the notion that firms with weak governance are more likely to pay dividends as a commitment device. 15 The regression in column 4 includes all the governance proxies at the same time. The sign and significance of the coefficients on board size and ownership by executive directors and institutions do not change. However, the statistical significance of the coefficients on the CEO/Chairman duality, managerial entrenchment, and CEO ownership variables is reduced to the point where those coefficients are not reliably different from zero at conventional levels. However, the coefficient on the indicator variable for an insider dominated board is positive (consistent with H1b) and statistically significant at the 10% level in this specification, and the effect of public pension fund ownership levels on the propensity to pay dividends is statistically significantly different from zero at the 1% level. The latter result suggests that predicted payers with low ownership levels by the most activist institutional investors are more likely to make dividend payments to stockholders. 16 Taken together, these results are the most consistent with the substitute model described above (H1b). There is no evidence that strong governance makes dividends more likely to be paid when firm fundamentals support such a policy, as predicted by the outcome model in La Porta et al (2000). Instead, firms with weak governance appear more likely to pay dividends when they should: predicted payers are more likely to make dividend payments if board size is large, the 15 The negative coefficients on CEO and executive director ownership variables are driven by CEO and executive director ownership below a 25% threshold. Specifically, the results of unreported spline regressions suggest that the propensity for firms in this sample to pay dividends is decreasing as CEO ownership increases between zero and five percent and as executive director ownership increases between five and 25 percent. These results are consistent with the assumption that increased ownership by insiders improves governance only when such ownership begins at a relatively low (less than 25%) base. Interestingly, the coefficient on CEO ownership in unreported spline regressions is positive (but statistically insignificant, t = 1.01) for CEO ownership levels in excess of 25%, consistent with the notion that increases in CEO ownership starting from a high base (over 25%) makes governance weaker because managers are already entrenched at high ownership levels. 16 The control variables in Table 3 are relatively stable in terms of sign and significance across all specifications. The earned/contributed capital mix (RE/TE), return on assets, growth, and size have similar effects on dividend paying behavior as seen in Table 1 and described in DeAngelo, DeAngelo, and Stulz (2006). 20

22 CEO is also Chairman of the Board, anti-takeover provisions protect managers from external discipline, and when ownership levels by executives and important external monitors are low. D. Robustness tests There are several alternative specifications that can be used to test the robustness of the conclusions derived from Table 3. One concern about the results presented in Table 3 is that the effect of governance characteristics on payout policy may be different than the effect on dividend policy. In other words, firms with strong governance may be less likely to pay dividends when fundamentals suggests they should, but that may be because managers at such firms act in the interests of their shareholders and distribute cash in the form of repurchases (which are taxadvantaged) instead of tax-disadvantaged dividends. To that end, the first column of Table 4 replicates the regression in the final column of Table 3 except that the dependent variable is an indicator variable equal to one if the firm makes a payout in the form of dividends or repurchases (and zero otherwise). Stock repurchases are identified as described in Fama and French (2001). Considering total payouts instead of just dividends does weaken the relation between governance characteristics and payout policy, as three of the coefficients that are statistically significantly related to dividend payments in Table 3 (Board is insider dominated, ownership by executive directors, and ownership by public pension funds) are not significantly related to total payouts in the first column of Table 4. However, the coefficients on board size and ownership by institutional investors have signs and significance levels consistent with Table 3 and the predictions of the substitute model. The fact that including repurchases in total payout policy weakens the relation between payout policy and governance measures is consistent with the findings in Knyazeva and John (2006). A further concern with the results in Table 3 is that the definition of predicted payer may not accurately identify those firms that should be paying dividends. The second column of Table 4 replicates the last column in Table 3 for a sample of firm-years for which a firm is only 21

23 considered to be a predicted payer after five consecutive years with the probability of being a payer (from Table 1) above 50%. While this reduces the sample size (the set of firms with five consecutive years of probability above 50% is a subset of the set of firms with two consecutive years of probability above 50%), the results in the second column of Table 4 are qualitatively similar to those in the last column of Table 3. The only significance level affected by the more stringent definition of predicted payers is for the coefficient on the insider-dominated board indicator variable, which is marginally significant in Table 3 but insignificant in the second column of Table 4. It is also true that dividend policy is not a binary decision firms can choose to pay out more or less of (free) cash flow or holdings once the decision to pay dividends has been made. The last column in Table 4 analyzes the effect of governance characteristics on a continuous measure of dividend policy the dividend payout ratio. The dividend payout ratio is defined here as the total amount of dividends paid in the fiscal year divided by the firm s pre-dividend cash balance (sum of the firm s cash balance at the end of the fiscal year plus the total amount of dividends paid in the fiscal year). This continuous variable ranges from zero to 1.27, and measures both whether dividends are paid and how much of the firm s pre-dividend cash balance is paid out to stockholders in the form of dividends. As can be seen in the final column of Table 4, the relation between governance measures and a continuous measure of dividend policy are practically identical to those in the fourth column of Table 3, with the only substantial difference being the loss of significance of the coefficient on public pension fund ownership and the dramatic increase in the statistical significance of the coefficient on the managerial entrenchment index. Overall, the results in Tables 3 and 4 are strongly supportive of the substitute model for the relation between governance and dividend policy (H1b). It appears that firms with characteristics associated with weak governance (large boards, more anti-takeover charter amendments, and low ownership by both insiders (CEO and executive directors) and institutional 22