Delivering on the Dividend Promise: Dynamic Dividend Behavior and Managerial Incentives *

Transcription

1 Delivering on the Dividend Promise: Dynamic Dividend Behavior and Managerial Incentives Anzhela Knyazeva New York University This version: August 2006 Previous version: May 2006 Abstract This paper examines the effect of governance and managerial alignment on dynamic dividend behavior of managers. I use a novel empirical approach to provide evidence on the role of governance and managerial alignment in explaining variation in dividend smoothing, incremental dividend decisions, and dynamic dividend behavior. Managers subject to weak monitoring and misaligned incentives are under greater pressure from shareholders to honor the implicit dividend contract as a trade-off between current period s private benefits and future job security. The main finding is that weak board quality and institutional investor monitoring, CEO characteristics associated with misalignment, and greater separation of ownership and control rights in dual class firms contribute to stronger adherence to the implicit dividend contract, avoidance of dividend cuts and omissions, and non-decreasing persistent dividend path. Misaligned managers are also more likely to raise dividends and use dividend increases as the primary form of additional payouts. Methodologically, this paper contributes to the dividends and governance literature by identifying and addressing endogeneity of managerial alignment and dividend behavior in the instrumental variables framework and adjusting for selection bias. JEL classification: G30, G34, G35 Keywords: dynamic dividend behavior, dividend smoothing, incremental dividend decisions, corporate governance, endogeneity, instrumental variables, selection model. I am thankful to my advisers Kose John, Joseph Stiglitz, Daniel Wolfenzon, David Yermack, and Bernard Yeung for valuable comments and discussions and William Greene, Eli Ofek, Anthony Saunders, and Jeffrey Wurgler for helpful suggestions. The title of the earlier version of the paper was Delivering on the Dividend Promise: Managerial Incentives, Dividend Smoothing, and Incremental Dividend Decisions. I am grateful to Paul Gompers, Joy Ishii, and Andrew Metrick for providing the data on dual class shares and to the Institutional Shareholder Services for providing the Corporate Governance Quotient data. I thank Stern School of Business for financial support. All errors and omissions are mine. Stern School of Business, New York University. 44 West Fourth St., Ste , New York, NY aknyazev@stern.nyu.edu.

2 1. Introduction Dividends remain a puzzle for the corporate finance literature. A significant number of firms continue to pay dividends despite their cost. Further, dividend cuts are infrequent and only used as a last resort. Various approaches seek to explain differences in firm dividend policies, including agency, information asymmetry and signaling, taxes. However, with much of the focus on cross-sectional variation in dividend levels and incidence, there is a need for a systematic examination of differences in dynamic dividend behavior. This paper bridges this gap and contributes in the following ways to the existing literature. First, it offers evidence on the effects of managerial alignment and governance on dynamic dividend behavior of managers. The focus of this paper is on managerial decision making, whereby dynamic dividend behavior arises as the outcome of fulfillment of the implicit dividend contract between the manager and shareholders. The testable implications address several understudied aspects of dividend policy: differences in firm propensity for dividend smoothing, explanation of incremental dividend decisions and dynamic dividend behavior, role of dividend changes in incremental payout policy decisions. The main result of the paper is that misaligned managers face greater pressure to uphold an implicit dividend contract with shareholders and choose to deliver on their dividend promise by maintaining a smooth non-decreasing dividend path. Misaligned managers attain this through more persistent dividends, avoidance of cuts and omissions in incremental dividend decisions, smaller magnitude of dividend decreases and use of dividend increases, reliance on repurchase decreases in downward payout policy revisions and preference for dividends during payout increases. Second, the paper employs a novel empirical approach to examine the relation between alignment and dividend behavior. The analyses identify and address endogeneity of

3 alignment and governance to dividend behavior. To the best of the author s knowledge, this is the first paper on dividends and governance to empirically evaluate the issue of causality and implement the instrumental variables framework. New instruments, reliant on the emergence of governance as the outcome of shareholder demand and managerial resistance to shareholder pressure, are constructed and tested for validity. The paper contributes two other empirical improvements: correction for selection bias and new measurement of managerial alignment. The analyses correct for selection bias in coefficient estimates that stems from non-random assignment of firms to the subsample of existing dividend payers using a two-stage selection model. The paper also constructs new measures of internal monitoring and the Alignment / Governance Index (AGI) to test the hypotheses. Empirical analyses use measures of external governance, board monitoring quality, institutional investor monitoring, CEO alignment characteristics, and separation of cash flow and voting rights from managerial ownership in dual class firms on a large panel of US firms. The main predictions of the paper regarding dividend behavior rely on decision making of rational self-interested managers faced with a dynamic tradeoff. The manager s decision to honor the implicit dividend contract with shareholders depends on the immediate benefit of forgoing a dividend payment and the expected future loss due to shareholder disciplinary response. The immediate benefit consists of greater flexibility and/or more private benefits. The future shareholder disciplinary response can include an increase in scrutiny and disciplinary pressure, stock sales resulting in reduction in the value of compensation and increase in the cost of capital, or firing (through a successful hostile takeover or forced turnover). The extent of the shareholder disciplinary response to failure to deliver on the dividend promise depends on the misalignment of managerial and shareholder incentives. 2

4 On the one hand, aligned managers deviating from the promised level of dividends are believed to make value-increasing decisions that increase the expected return on investment. Therefore, as long as the manager can provide a high return, managerial deviations from the implicit dividend contract do not cause an increase in shareholder disciplinary pressure. Shareholders accept a less persistent dividend path with potential for cuts and omissions when managers are aligned and governance is strong. On the other hand, misaligned managers (in the presence of weak board quality, a lack of institutional investor scrutiny, CEO characteristics consistent with misalignment, and dual classes of shares) that deviate from the implicit dividend promise are believed to invest suboptimally and destroy firm value. Shareholders subject misaligned managers that fail to honor the dividend promise to greater disciplinary pressure. As a result, misaligned managers find it optimal to be more committed to the implicit dividend contract in setting the dynamic path of dividend. Conditional on having made a dividend promise, misaligned managers show more persistence in dividends, avoid cuts and omissions, exhibit a smaller magnitude of decreases, and are overall reluctant to compromising the implicit dividend contract in payout policy decisions. The proposed explanation is consistent with observed variation in dividend smoothing, direction and magnitude of dividend changes, and choice of the form of cash distribution increase. The rest of the paper is organized as follows. The second section formulates the main hypotheses. The third section discusses data, variables, empirical methodology and instrument construction. The fourth section presents empirical results. The fifth section concludes and outlines questions for future research. 3

5 2. Hypothesis development Managerial objective and dynamic tradeoff Agency costs of inefficient investment in projects that increase firm size, diversify risk or otherwise enhance private utility of the manager lower firm value, as was suggested by Jensen and Meckling (1976) and in subsequent work. The level of private benefits and inefficient investment in a given period is higher for poorly monitored managers and managers with misaligned incentives. The formulation of the problem is practically unchanged if inefficient investment is instead associated with suboptimally low level of privately costly managerial effort. From the shareholder perspective, a significant loss of firm value due to inefficient investment can be prevented through costly firing and replacement of the current manager. Alternatively, from the perspective of a potential raider, a drop in market value of the firm due to inefficient investment will render a takeover more attractive (the cost of takeover weighed against gain from turning around the firm). Both in the case of forced turnover and a successful takeover, the manager s job security, i.e. the likelihood of remaining in the firm next period, is decreased. The manager has a negative expected utility from being fired. It is plausible that expected future income of a fired manager is lower than that of a manager who kept the job. Performance-based firing can weaken managerial reputation, lower expected future wages and impose costs of job search. In the worst case scenario, the fired manager is not re-hired and loses the entire stream of expected compensation and private benefits from future periods. Therefore, the manager s suboptimal behavior in a given period affects two components of expected lifetime utility: first, it directly increases current period s private benefits; second, it decreases expected future compensation and private benefits through 4

6 increasing the probability of firing. In other words, the incumbent manager faces a dynamic tradeoff between current period s private benefits and future job security. Managerial incentive to uphold a dividend promise The described dynamic tradeoff explains managerial actions that may further shareholder interests in order to alleviate monitoring pressure and increase job security. In related work, Fluck (1999) models managerial choice of the distribution of equity ownership that maximizes private benefits against the risk of potential control threats (see also Zwiebel, 1996). John and Knyazeva (2005) argue that design and composition of payout policy (specifically, the presence of dividends versus repurchases) can serve as a form of precommitment in firms with bad corporate governance. The interest of this paper is managerial choice of characteristics of dynamic dividend path - extent of dividend smoothing and incremental dividend decisions and the effect of misaligned incentives on the likelihood of delivering on the dividend promise to shareholders. An imperfectly aligned manager will set firm dividend policy to achieve an optimal tradeoff between current period s private benefits and probability of keeping the job next period. Poorly monitored managers are able to derivate greater private benefits from suboptimal investment out of free cash flow, resulting in a more severe agency problem and a larger loss in firm value. In order to mitigate the loss of firm value (and the corresponding increase in the probability of firing), a misaligned manager can limit the amount of private benefits by agreeing to disburse cash to shareholders and making a dividend promise. John and Knyazeva (2005) document higher levels of dividend payout among poorly governed firms, so this paper does not focus on the initially promised dividend amount. Instead it examines period-by-period behavior of misaligned managers related to continuing or modifying dividend policy and dividend smoothing, conditional on a past dividend promise. 5

7 Dividend policy credibly restricts suboptimal behavior of a misaligned manager only if it implies regular payouts that are not arbitrarily decreased by the manager. Shareholders rationally expect a misaligned manager to invest inefficiently. Controlling for other factors (change in the cost of dividends to the firm and extent of the free cash flow), deviations from the dividend promise will be attributed to increases in managerial private benefits. When a misaligned manager cuts or omits a dividend payment, shareholders will respond with a stronger negative market reaction on announcement, which imposes additional monitoring and reputation costs and can potentially result in firing. Therefore, misaligned managers face a greater need to uphold the dividend promise they made. Similarly, shareholders expect misaligned managers to increase the promised dividend in response to an increase in internally generated firm cash flow or a decrease in investment opportunities. On the other hand, when a promise of dividends is made by a better aligned manager, the shareholders are more flexible with respect to deviations from the dividend contract. Incremental dividend decisions that lower the promised dividend level are optimal for the firm if the efficiency of allocation of internal cash flow is improved and expected return on investment is increased as a result. A better aligned manager s decision to deviate from the dividend contract is more likely to be due to an increase in expected future investment opportunities or a decrease in expected internal cash flow rather than to an increase in managerial private benefits. Shareholders rationally anticipate this and accept less persistent dividends and a larger likelihood of downward changes. In exchange, they require a higher expected return on investment. Therefore, better incentive alignment allows the manager to have more flexibility with dividend policy without sacrificing firm value and, respectively, job security. 6

8 The proposed argument yields several new empirical predictions about the effect of incentives on dividend policy implementation in the context of individually optimal managerial behavior. In particular, the paper examines the issues of dividend smoothing, incremental dividend changes (dividend cuts, omissions and increases), use of dividends to distribute additional cash to shareholders. Managerial alignment and dynamic dividend behavior: hypotheses Dividend smoothing As was argued above, misaligned managers face greater pressure from shareholders and the stock market to uphold the dividend promise. Incentive misalignment implies that shareholders rationally anticipate less efficient investment and lower expected future returns. In order to mitigate the ensuing shareholder pressure and threat of firing, a misaligned manager can make a credible dividend promise to indicate commitment not to destroy shareholder value. The market s demand for a credible dividend promise by the manager is increasing in the degree of incentive misalignment. The manager can make a more credible dividend promise by undertaking regular and persistent dividend payments. Greater dividend persistence (dividend smoothing) implies that a restriction on managerial behavior is consistently upheld over time. For better aligned managers, high expected return on investment is more important for addressing shareholder pressure and the threat of firing while a credible dividend promise contributes relatively less value since the cost of suboptimal managerial behavior is mild. Existing empirical evidence (see, e.g., Benartzi, Michaely, and Thaler, 1997; Grullon, Michaely, Benartzi, and Thaler, 2005) indicates that dividends do not carry additional signaling power about unexpected changes in future firm earnings. Therefore, better aligned managers will be able to mitigate the threat of firing more effectively by improving earnings 7

9 and firm value directly rather than adhering to a restrictive dividend promise. In addition, better aligned managers can conserve firm resources and improve efficiency of allocation of firm cash flow by opting for a more flexible (less persistent) dividend policy. More frequent deviations from the established dividend level, including downward adjustments, are optimal for better aligned managers and their shareholders. The presented argument is summarized in the Hypothesis H1 below. H1. Misalignment of managers increases dividend smoothing (persistence). Managerial misalignment reduces dividend variability. In examining dividend smoothing, this paper relates to existing literature that documents a high degree of dividend persistence for the full sample of firms (see, e.g. Lintner, 1956; Fama and Babiak, 1968). In subsequent theoretical work, Kumar (1988) develops a model of coarse dividend signaling of firm productivity type that is consistent with smoothing. Warther (1994) derives dividend smoothing predictions in an asymmetric information game between shareholders and management, in which managers must pay a dividend sufficient to demonstrate adequate profitability. Fudenberg and Tirole (1995) provide a different theory of dividend and earnings smoothing based on managerial job security. In more recent work, Dewenter and Warther (1998) perform a comparison of dividend decisions of US and Japanese companies and find higher dividend persistence and lower probability of cuts and omissions in the US. Gugler (2003) uses Austrian data and finds that state-controlled firms are most reluctant and family-controlled firms are least reluctant to cut dividends when cuts are expected. Contributing to the existing literature, this paper explains differences in dynamic dividend behavior through misalignment of managerial incentives, controlling for information environment of the firm, risk and other factors. It also employs instrumental variables framework to endogenize managerial incentives (will be discussed in Section 3.). 8

10 Since consistency of dividend payments over time is essential to restraining managerial opportunism, the empirical findings on this issue provide new insights about the endogenously determined effectiveness of dividends as an agency cost reduction mechanism for poorly monitored managers. Incremental dividend decisions: dividend cuts and omissions The previous hypothesis suggests that a misaligned manager can succeed at making a more credible dividend promise if dividend payments are persistent over time. The second question is how dividend smoothing is implemented by misaligned managers, provided Hypothesis H1 holds. Analysis of incremental dividend decisions focuses on changes (increases and decreases, including omissions) of dividend payments. By exposing the manager to monitoring from the external financing market, a persistent dividend policy mitigates the problem of inefficient investment and private benefit consumption. Therefore, dividend decreases by a misaligned manager represent departures from the promised level of additional monitoring assumed by the manager, controlling for other effects. Shareholders perceive downward deviations from the promised dividend level as a misaligned manager s attempt to increase private benefits at a cost to firm value. Therefore, dividend smoothing is implemented through avoidance of dividend cuts and omissions, i.e. protection against downward changes in the promised dividend level. H2. Misaligned managers avoid dividend cuts and omissions. Hypothesis H2 suggests that incentive alignment is negatively related to the likelihood of dividend decreases and negatively related to dividend changes (more negative changes undertaken by better aligned managers). Combined, hypotheses H1 and H2 predict a persistent, non-decreasing dividend path for misaligned managers. 9

11 The other implication of this argument is that dividend cuts by misaligned managers would encounter a more negative market reaction (discussed in Section 4.). Incremental dividend decisions: dividend increases The next issue is likelihood of dividend increases by misaligned managers. The earlier argument implies that better aligned managers are as likely or less likely to increase the promised dividend level since they do not realize significant gains in job security from committing a higher amount to dividend payments. For misaligned managers, two possibilities arise. We will use empirical tests of the likelihood of changes to distinguish between them. On the one hand, the promised dividend can be maintained at a high constant level that imposes a significant constraint on managerial behavior and additional monitoring from the external financing market. The credibility of the dividend promise is attained through avoidance of decreases. In this instance, the manager may not want to increase dividends since a one-time increase will have to be reflected in higher subsequent dividend payments. If later on a manager is not able to continue these higher dividend payments, a dividend cut announcement would send a negative signal to the shareholders, which could increase the threat of firing, contradicting the initial objective of dividend use to mitigate the shareholder pressure and firing threat. Therefore, misaligned managers are as likely as better aligned managers to increase dividends. On the other hand, a stream of dividends that is strictly increasing over time could be consistent with the proposed argument. Initially, misaligned managers set dividends at the moderate initial level sufficient to maintain a credible constraint on managerial behavior. Since some uncertainty over future free cash flow is present and decreases in dividends are costly to the manager, the manager realizes expected gains from waiting until next period, i.e. 10

12 retaining the option not to increase the dividend in the future. If the firm s cash flow continues to be high or is expected to increase, the misaligned manager can signal a renewed commitment to uphold shareholder interests by raising the dividend. However, if the firm s cash flow does not increase or decreases, the manager can avoid the cost of announcing the cut of a dividend that had been set at a high level earlier. Therefore, after setting the initial dividend level, a misaligned manager will either sustain it or increase it, depending on information about expected future cash flow. Retaining this control over dividend path enables the manager to supply only positive signals to the market in order to improve job security. A further consideration is that the optimal dividend change following an increase in free cash flow (and not only the initial dividend level) depends on the extent of managerial misalignment. Managerial misalignment raises the magnitude of the dividend increase that is necessary to maintain the previously established level of constraint following an increase in free cash flow. Therefore, in this context misaligned managers are overall more likely to announce a dividend increase. H3. Dividend increases are non-decreasing in managerial misalignment The exact sign of the relation between weak incentives and dividend increases (insignificant or positive) depends on which of the two explanations above holds and will be identified empirically. Dynamic dividend behavior: changes in dividend level The previous discussion of hypotheses about incremental dividend decisions allows the following characterization of dynamic dividend behavior conditional on managerial alignment: H4. Aligned managers undertake lower dividend changes. Managerial misalignment/weak governance causes higher dividend changes. 11

13 Managerial misalignment, including in instances of weak governance, is associated with upward pressure on managers in the determination of dynamic dividend path. Aligned managers face less upward dividend pressure and enjoy greater flexibility over downward deviations from the previously committed dividend level. Managerial behavior described above is consistent with individually optimal decision-making of the rational manager in the framework of implicit contracting. In the present context, a previously announced level of dividends is interpreted as an implicit dividend contract with shareholders. Whether managers continue to deliver on the promise of dividend payments (and are subjected to the pressure to expand dividend payments over time) is determined by the amount of discretion that the manager can retain and shareholders are willing to concede with respect to the dividend policy given the firm s investment opportunities, cash flow, and, importantly, managerial alignment and governance quality. Up to this point, the discussion has focused on the amount of pressure exerted by shareholders in the presence of managerial misalignment or weak governance mechanism since misaligned managers are believed not to be guaranteed job security. Besides threatening managerial job security, which is the most direct implication of the previous argument, shareholders can resort to other channels of exercising disciplinary pressure over a deviating manager. The ensuing negative effect of sale of shares ( voting with their feet ) on stock price can have other consequences than facilitating an unfriendly takeover offer or board-initiated turnover. Even if allowed to remain in the firm, the deviating manager can experience lower expected earnings from stock compensation. Further, a drop in stock price would have a detrimental effect on the cost of capital, should the firm require external financing, which would lower profitability and/or scope of accessible projects and lower potential for extraction of private benefits in the future. Use of dividend changes in payout policy revisions 12

14 Previous predictions characterize dynamic dividend behavior, including persistence, variability, magnitude and direction of changes. However, incentive misalignment can also affect the bigger picture of dynamic payout policy. Supplementing the previous argument, the next hypothesis focus on the following question: when would dividend change be the preferred choice in payout policy revision? Both increases and decreases will be considered. Adhering to the baseline dynamic dividend behavior argument, a misaligned manager would benefit from choosing dividends as the form of payout increase due to higher shareholder confidence and job security realized as a result of the added disciplining effect of having to sustain higher future dividends. The cost of the dividend increase (that makes a misaligned manager s action more credible) is the need to sustain payout at a higher level in the future. Alternatively, the manager can make a one-time repurchase instead of distributing the increase over time in the form of a higher dividend. A large temporary cash distribution will have a weaker disciplining effect on the manager since no continued effort to follow a new payout path is necessary but can distribute the same amount of additional cash to shareholders. In addition to a weaker disciplinary effect, information effects associated with repurchases can also interfere: for instance, a misaligned manager may be able to time the repurchase to inaccurately supplied information about firm performance in an attempt to exploit the advantage over uninformed investors, which would lower the positive effects of a repurchase. The author believes that the relatively stronger disciplining effect of a dividend increase is the main reason for the preference of misaligned managers towards dividend increases. H5a. Misaligned managers choose to use dividend increases as the form of distributing additional cash payouts. When downward payout policy decreases become necessary, dividend decreases are expected to be the least preferred choice of misaligned managers. Provided that alignment 13

15 and governance quality remain weak, a downward payout policy revision that does not cause a deviation from the implicit dividend contract will prevent a strong disciplinary response on the part of shareholders. Therefore, a misaligned manager is more likely to scale down the volume of repurchases rather than announce a dividend cut or omission. H5a. Misaligned managers prefer to use repurchase decreases (not accompanied by dividend decreases) as the form of payout decrease. The next section describes data, construction of variables, and empirical methodology. 3. Data and empirical methodology Empirical analyses in this paper focus on the sample of U.S. firms. Data on firm characteristics is obtained from Compustat Industrial Annual for The sample excludes financial firms (SIC codes ) and regulated utilities (SIC codes ). We also omit observations with missing data on book value of total assets (item #6), firms with asset size below 20 mln, firms incorporated abroad (incorporation code 99), LBOs (Compustat stock code 4). CRSP monthly file contains information on cash distributions. We include only ordinary common shares. Certificates, ADRs, shares of beneficial interest, and units are excluded. We also omit data points corresponding to Americus trust components, closed-end funds, and REITs. Dividends are defined as the sum of distributions with distribution code 1232 (ordinary quarterly cash dividends) over four fiscal quarters, adjusted for splits and stock dividends, to form annual dividend level. Managerial Alignment In empirical tests misalignment of managerial incentives is captured through firm governance characteristics that weaken the scope and intensity of scrutiny of managerial actions or lower the likelihood of punishment for inefficient behavior and through firm and CEO characteristics that broaden the gap between managerial and shareholder utilities. Existing governance literature emphasizes the role of monitoring mechanisms for firm 14

16 performance and decision making, including charter and bylaw provisions that increase exposure to the market for corporate control (Gompers, Ishii, Metrick, 2003; Bebchuk and Cohen, 2004), monitoring by a large blockholder (Cremers and Nair, 2005), independence of the board of directors (Agrawal and Knoeber, 1996), small board size (Yermack, 1996). John and Knyazeva (2005) empirically identify the link between external and internal governance mechanisms, payout policy design and dividend behavior. External governance measure is based on the Gompers, Ishii, and Metrick (2003) index of takeover defenses. Managers in firms with fewer takeover defenses are exposed to the market for corporate control and are therefore subject to sufficient discipline without upholding the implicit dividend contract. More generally, the provisions that comprise the G index can be said to influence the decision making process in the firm that takes the bargaining power away from insiders and facilitates independent oversight of the CEO. The remaining alignment measures involve intensity of institutional investor monitoring, quality of board monitoring, CEO-level alignment characteristics. These four characteristics are used together with the Gompers, Ishii, Metrick (2003) external governance measure to form the Alignment/Governance Index (AGI) that captures various aspects of managerial misalignment (due to weak external governance, weak internal board or institutional blockholder monitoring, CEO characteristics consistent with shorter horizon or stronger entrenchment in the firm, and agency conflict due to separation of ownership and control rights) and therefore serves as an inclusive determinant of the extent of shareholder pressure on the manager with respect to the implicit dividend contract. The construction of the AGI variable and its components is described in Appendix A; for the purposes of the index, rankings of firms by characteristics believed to be associated with better alignment are 15

17 rescaled to [0,1] and used instead of the absolute values. All key results are checked against both AGI and five component measures of alignment. Institutional investor monitoring is expected to be stronger in the presence of a larger blockholding in the firm 3. Further, the higher concentration of institutional ownership and the lower number of institutional owners would decrease the costs of coordination and improve the intensity of scrutiny of managerial actions. Finally, some categories of institutional owners are expected to be more active in monitoring the manager; for this purpose, proxies are constructed for public pension fund participation in the firm. In sum, managers in firms with an institutional blockholder and public pension fund investor, larger and more concentrated institutional ownership, and fewer institutional owners are expected to have less severe incentive alignment problems. The quality of board monitoring is expected to increase with independence of directors (and decrease with presence of employee directors), decrease with board size (see Yermack, 1996), decrease with the introduction of the CEO on key committees (nominating, compensation, audit, governance; see Shivdasani and Yermack, 1999) and increase with the frequency of board meetings. Certain CEO characteristics are expected to increase misalignment. First, length of CEO tenure in the firm could indicate the level of control (bargaining ability) the CEO has over the board and ability to resist oversight. Second, CEO age could be associated with a shorter horizon in the firm, indicating potentially higher discrepancy with the profit maximization objective of shareholders. Presence of dual class shares is expected to be associated with greater separation of control and cash flow rights of the manager, as in Gompers, Ishii, Metrick (2005), and weaker alignment of manager and shareholder interests. Control variables 3 Amihud and Li (2002) argue that a lower degree of information asymmetry between management and institutional owners is the underlying reason for the negative relation between institutional ownership and dividend payout. 16

18 The measures described above characterize various dimensions of monitoring that restrict a manager s potential to derive private benefits. The other important part of the manager s objective is the strength of ownership incentives. Existing literature uses CEO ownership stake. Fenn and Liang (2001) find that managerial ownership is positively associated with payout in firms with low management stock ownership and few investment opportunities. Further, stock option compensation is an increasingly important component of executive pay and presence of stock options has been found to limit use of dividends (Fenn and Liang, 2001). Data on the ownership stake of the CEO in the firm and CEO stock option grants is obtained from Execucomp (where the series starts from 1992). Other characteristics that could be correlated with dividend decisions and managerial incentives are included. One should note that all managerial dividend choices are conditional on the availability of cash flow and investment opportunities, therefore the level of cash flow and firm investment opportunities is used to capture the extent of the free cash flow problem. Firm size is expected to be positively related to dividends (Fama and French, 2001). Hoberg and Prabhala (2005) examine the negative effect of risk on dividends, and Jagannathan, Stephens, and Weisbach (2000) emphasize the financial flexibility of payout policies. Measures of firm risk based on cash flow uncertainty and of industry-level volatility are expected to be associated with lower dividends and more dividend decreases. Regressions also control for measures of information asymmetry of the firm - number of analysts with one-year-ahead forecasts of firm earnings. Bid-ask spread proxies for liquidity of the firm s shares; firms with more liquid shares may pay lower and more variable dividends as investors value dividends less. The other control is change in median dividend level, which is used to capture industry conditions and, potentially, propensity of firms to follow a generally accepted dividend level. Variable definitions are described in more detail in Appendix A. 17

19 Econometric analysis Empirical tests address the hypotheses formulated in the previous section. First, the coefficient on the lag of dividends for better aligned and misaligned managers in the dividend regressions for firms that register positive dividends is estimated to test Hypothesis H1. The sign on the interaction term is hypothesized to be negative, i.e. governance quality lowers dividend persistence. Coefficients of persistence are also estimated for each firm using the standard model of dividend level. These coefficients (available only in the cross-section) are subsequently regressed on averages of firm characteristics and governance quality (expected to enter negatively). In addition, several measures of dividend variability are used. Both in the full sample and in the cross-section, dividends are expected to vary less when alignment/governance is weak. To filter out the effect of potential interfering firm-level factors, standard deviation of residuals from the expanded dividend regression is used as a proxy for variability. The other hypotheses examine dividend changes from the point of view of magnitude, frequency, and direction. Hypotheses H2 and H3 that deal with dividend changes of different type are tested in the Probit framework. In addition, the model of the likelihood of any change in dividends (regardless of direction) is estimated. Alignment and governance quality variables are expected to enter positively in the dividend decreases equation and positively (or insignificantly) in the dividend increase equation. Further, the magnitude as well as direction of dividend changes is examined in the regression framework using change in the level of dividend relative to lagged dividend as the dependent variable. Governance is expected to enter negatively in the change in dividend level equation. Dividend data for the previous fiscal year must be available for all of these analyses. 18

20 Finally, managerial preference towards method of payout policy changes is examined in the Multinomial Logit framework. To examine the likelihood of dividend increase (decrease) relative to an upward (downward) payout revision not involving dividends, we construct categorical variables for the type of payout policy change (no change, change in repurchases but not in dividends, change in dividends). Unlike in the case of ordered logit, outcomes are not subject to a specific ordering. From the methodological standpoint, this paper differs from existing literature on dividends in recognizing and empirically addressing the potential issues of endogeneity of governance and sample selection bias. Therefore, regressions correcting for selection bias and endogeneity are presented following regular empirical analyses. Endogeneity of governance When measures of managerial incentive alignment used on the right hand side are not exogenous to incremental dividend decisions, explanatory variables are correlated with the error term in dividend equation. Endogeneity of governance will bias OLS estimates even if all relevant control variables are included in the dividend equation (no omitted variable bias) and incentive alignment is measured without error (no error-in-variables problem). Palia (1999) examines endogeneity in the link between CEO ownership and performance. Brick, Palia, and Wang (2005) and Chidambaran, Palia, and Zheng (2006) examine the causal link between governance and performance. Hermalin and Weisbach (1998) provide an analytical argument regarding endogeneity of boards. To the best of the author s knowledge, this is the first paper to identify and correct for the endogeneity of governance and alignment to dividend behavior in the instrumental 19

21 variables framework 4. One could offer an argument that illustrates potential for simultaneity or reverse causality specifically in the case of dividends. First, endogeneity of governance could be due to simultaneous choice of incentive structure and characteristics of the implicit dividend contract that binds the manager. For instance, instead of increasing the intensity of monitoring and degree of incentive alignment, shareholders can agree to a non-decreasing persistent dividend path that provides an alternative way of constraining the manager. Depending on the anticipated future cost of dividends to the firm relative to the cost of setting up a strong monitoring and incentive structure, an optimal tradeoff of governance and dividends is attained. Although governance would be negatively related to dividends, it will not cause dividend decisions in this case. Second, instead of being simultaneously determined, governance and dividends could exhibit reverse causality. From the manager s perspective, consistent compliance with the implicit dividend contract could earn a more favorable shareholder attitude and potentially a reduction in monitoring intensity. In both cases, the empirical relationship between governance and features of dividend policy will have the hypothesized sign but will not be informative about causality. This paper examines the hypothesized dividends governance effects with explicit recognition of potential endogeneity in the instrumental variables framework. At the first stage, governance is predicted using a set of instruments (described below). At the second stage, dividend behavior examined using instrumented governance variables. The right-hand-side economic controls are believed to be exogenous variables (X1) whereas the AGI variable and component alignment and governance measures are suspected for endogeneity (X2). A set of exogenous instruments (Z), with at least one instrument per each variable in the set X2, is chosen. Instruments affect governance but do not affect 4 Grinstein and Michaely (2005) employ VAR methodology to analyze institutional ownership and payout policy, but the scope of the present paper is governance and managerial alignment (and instrumental variables methodology is used to account for potential endogeneity). 20

22 dividends through channels other than governance quality (alignment). In the instrumental variables estimation framework, two conditions on the validity of instruments should be met. First, instrument relevance, i.e. significant predictive power of instruments in explaining endogenous governance variables in the first-stage (governance) equation: Corr(X2,Z)0. Violation of this condition results in nonnormal sample distribution of the IV statistics, resulting in unreliable hypothesis tests (for more on the weak instrument problem, see, e.g. Stock and Yogo, 2002); in the extreme case of instrument irrelevance, estimates are inconsistent. Second, instrument excludability, i.e. insignificant correlation of instruments with the errors of the second-stage (dividend) equation: Corr(,Z)=0; violation of this condition results in inconsistent estimation (as with original endogenous regressors). In the course of the empirical analysis the proposed instruments are tested against the above criteria. First-stage F statistics (the rule of thumb in Staiger and Stock (1997) for a single endogenous regressor is that F statistic should exceed 10), Shea partial R 2 values (that address correlation between instruments), and Anderson canonical correlations underidentification test statistics are examined. In addition, rejection of the null hypothesis of weak instruments using the Cragg and Donald (1993) test is used to validate instrument strength, as suggested in Stock and Yogo (2002). The null hypotheses of instrument irrelevance and instrument weakness are rejected (see discussion in the next section). Another econometric issue concerns the need to use instrumental variables. If ordinary least squares produces consistent estimates, i.e. there is no evidence of significant correlation between governance variables and errors in the dividend equation, OLS estimation is the preferred choice since it is more efficient. Tests of overidentifying restrictions using the Hansen J statistic are performed to assess exogeneity of governance variables and their respective instruments. This test is robust to the presence of 21

23 heteroscedasticity, including within-cluster heteroscedasticity used to adjust standard error estimates in the performed regressions. Rejection of the null hypothesis suggests that endogeneity is present. The null hypothesis is expected to be rejected in the equation without instrumental variables if governance is endogenous; it is expected not to be rejected in the equation with instrumental variables if instruments are exogenous. The results of the tests of overidentifying restrictions and instrument relevance satisfy these requirements, as discussed in the next section. The construction of instruments is described below. The search for instruments is focused on determinants that affect corporate governance and managerial incentives but not dividend policy changes. Governance quality is believed to arises as the outcome of shareholder demand for a sound monitoring structure (level of concern about strong governance) and managerial ability to thwart shareholder efforts aimed at maintaining good governance (managerial resistance to monitoring). First, shareholder demand for strong governance is captured by market premium for good governance quality. Market governance premium is computed at the industry-year level (see Appendix A for details of its construction). The premium is defined separately for the aggregate AGI measure and component governance measures and is related to the interpretation of the dividend premium in Baker and Wurgler (2004). Market-wide governance premium is exogenous to firm-level dividend decisions and serves as a measure of intensity of shareholder demand for good corporate governance and therefore can be argued to predict choice of governance quality. Institutional investor involvement with other firms in the portfolio can serve as a proxy for the attention span and level of dedication of the investor to an individual firm. Three measures are used to proxy focus (concentration of holdings in the portfolio), involvement (average ownership stake among firms in the 22

24 institutional blockholder s portfolio), and attention span (number of firms in the portfolio). More focused and involved blockholders build up more significant exposures to each individual firm and therefore have a stronger demand for governance quality of an individual firm. Firms that are owned by distracted institutional investors with dispersed portfolios will be under relatively less pressure to maintain good governance since sale of the firm s shares is the easiest solution for the institutional investor in the case of an agency conflict. Shareholder demand for monitoring and governance is also expected to be affected by the value of managerial discretion given the nature of the industry. In industries where valueenhancing managerial effort is much less observable, monitoring (disciplining) mechanisms can frequently fail to identify inefficient investment projects. The higher level of noise and false negatives associated with the monitoring structure could reduce overall shareholder demand for strong governance. One could argue that high-tech industries fulfill this criterion due to extensive presence of innovative projects with a high degree of asymmetric information and investments whose return depends considerably on managerial skill, hence the value placed of managerial discretion is high. The high-tech industry dummy is defined according to Kasznik and Lev (1995): companies in pharmaceutical (SIC codes ), computer (SIC codes ), electronics (SIC codes ), programming (SIC codes ), and R&D services (SIC codes ) industries. Second, managerial ability to thwart shareholder efforts aimed at maintaining good governance is conditional on the practice related to governance in a given industry and state. If most firms follow sound governance practices, the manager s ability to negotiate an inferior governance structure is hampered. A manager that disagrees with the prevalent governance and monitoring level is more likely to be fired and replaced with a manager that works within the accepted governance standards. An individual firm s manager is unlikely 23

25 to be able to influence the distribution of governance quality in the industry / location, which defends the instrument s exogeneity. Presence of state anti-takeover laws represents another exogenous proxy for the ease of weakening the governance structure at the firm level. A firm s internal and external governance is expected to be correlated with median governance quality and state laws. The advantage of the last two proposed instruments for governance quality is that a strong argument can be made for their exogeneity to an individual firm s dividend policy. However, these measures are also potentially noisier and can have a low correlation with the underlying endogenous firm-level governance variables. Therefore, a firm-level proxy for managerial resistance to governance is used. It is expected that managers in firms with strong market power are in a better position to provide a steady stream of earnings for shareholders and thwart attempts to impose additional monitoring (entrenchment effect). Market power of the firm is captured by its share in industry sales in a given year. Finally, market environment conditions at the time of the firm s entry affected the bargaining power of the shareholders and the management when the early firm governance structure was formed. Governance mechanisms evolve slowly over time, so entry-year determinants of governance would arguably predict today s governance quality. The proposed instrument focuses on the level of market for corporate control activity in the year of entry into CRSP database (the fraction of firms delisted due to merger). Unlike the year of entry variable, which could be responsible for the direct age effect on dividends, more aggressive takeover market activity in the year of entry is likely to specifically affect formation of a better governance mechanism. Use of the CRSP historical data has advantage over Compustat due to longer time series, hence more accurate measurement for more mature firms. Addressing selection bias 24