Does Better Corporate Governance Encourage Payout? Idiosyncratic Risk, Agency Problem, and Dividend Policy

Transcription

1 Does Better Corporate Governance Encourage Payout? Idiosyncratic Risk, Agency Problem, and Dividend Policy by Debarati Bhattacharya a, Wei-Hsien Li b, and S. Ghon Rhee c Abstract Does stronger corporate governance encourage payout? The literature is divided on this question. Our empirical investigation reveals that corporate governance acts as a substitute for [complement to] a firm s dividend policy when its idiosyncratic risk is high [low], in effect managing the problem of underinvestment [overinvestment]. These findings are robust to various alternative model specifications and exogenous variations in corporate governance that result from state adoption of antitakeover laws. The dividend substitution [complement] effect is more pronounced in firms with possible CEO risk aversion [with higher free cash flow]. We observe a similar phenomenon in other types of payout decisions. JEL classifications: G34, G35 Keywords: Corporate Governance, Payout Policy, Idiosyncratic Risk. a Corresponding author. Palumbo Donahue School of Business, Duquesne University, 813 Rockwell Hall, Pittsburgh, PA, 15282, USA, phone: , fax: , bhattacharyad@duq.edu b Department of Finance, National Central University, Taoyuan 32001,Taiwan, R.O.C., weihsienli@ncu.edu.tw, c Shidler College of Business, University of Hawaii at Manoa, Honolulu, HI 96822, USA, rheesg@hawaii.edu We would like to thank Cheng-Yi-Shiu, Chia-Wei Huang, Yanzhi Wang, David J. Denis, Ying Huang, Jarrad Harford, Carl Hsin-han Shen, Shan Zhou, Taeko Yatsuake, and Joonho Kim for their comments. We wish to thank the participants at the 2015 National Central University mini-conference, the 2015 Taiwan Finance Symposium, the 12th Conference on the Development of Cross-Strait Financial Markets, the 23rd Conference on the Theories and Practices of Securities and Financial Markets, the 2016 Islamic Finance, Banking & Business Ethics Global Conference, the Society of Interdisciplinary Business Research 2016 Osaka Conference as well as the seminar participants at Hitotsubashi University and Deakin University for their comments. We also thank Bo-Shun Chen and Ya-Yun Kao for their research assistance. Bhattacharya gratefully acknowledges the financial support from the Palumbo Donahue School of Business Internal Research Grant, Summer Li gratefully acknowledges the financial support from the Ministry of Science and Technology of Taiwan (MOST H ). Rhee is grateful for the 2017 summer research grant from the Shidler College of Business, University of Hawai i. 1

2 1. Introduction The current literature offers two competing views on the relation between corporate governance and dividend payouts: (i) one view suggests that they complement each other; (ii) the other view theorizes that they are substitutes for one another. Strong governance may enforce a payout policy that reduces free cash flow and the associated agency costs, in which case corporate governance and dividend policy complement each other. Alternatively, firms with severe agency conflicts may make payout commitments to address the weakness of their governance mechanisms. In those cases, corporate governance act as a substitute for the dividend policy. Evidence documented by La Porta, Lopez-de-Silanes, Shleifer, and Vishny (2000), Francis, Hasan, John, and Song (2011) and Petrasek (2012) supports the view that corporate governance acts as a complement to dividend policy, whereas conflicting results are provided by Hu and Kumar (2004), Kalcheva and Lins (2007), Officer (2011), and John, Knyazeva and Knyazeva (2015). In a recent review, Farre-Mensa, Michaely, and Schmalz (2014) indicate that There are still significant disagreements in the literature regarding how corporate governance affects payout policy. In this paper, we reconcile this debate by studying the efficacy of governance mechanisms in addressing concerns about two types of distortions in investment decisions, underinvestment and overinvestment. The first is caused by managerial risk aversion, and the second is due to agency conflict of free cash flow. Fama and French (2001), Grullon, Michaely, and Swaminathan (2002), and DeAngelo, DeAngelo, and Stulz (2006) all advance the idea of a tradeoff between cash retention and distribution that is largely guided by profitability and investment opportunities and evolves over time. However, these studies do not address the issue of the possible effects of investment distortions, caused by managerial behavior, on the firm s 2

3 dividend policy. For example, in the absence profitable investment opportunities, managers motivated by rent-seeking behavior (rather than precautionary need) may want to stockpile cash, subsequently leading to overinvestment. Jensen (1986) and Stulz (1990) develop the free cash flow hypothesis, which predicts that shareholders will choose to limit the availability of free cash flow to managers to mitigate this type of dissention. This is the underlying motivation of the aforementioned studies that support corporate governance as a complement to the dividend policy hypothesis. Nevertheless, the question that is largely left unexplored by the literature on dividend payout, cash retention, and corporate governance, asks how the tradeoff between distribution and cash retention changes if risk-averse managers forego profitable investment opportunities, or underinvest, as firm-specific uncertainties increase. DeMarzo, Fishman, He, and Wang (2012) and Panousi and Papanikolaou (2012) develop theoretical models that predict decreasing investment with rising idiosyncratic risk. In particular, Panousi and Papanikolaou (2012) propose and empirically test a two-period model in which they demonstrate that risk-averse managers underinvest when idiosyncratic risk increases. We study how governance mechanisms affect the propensity to pay dividends in the face of the rising risk of underinvestment. When idiosyncratic risk is high, self-interested risk-averse managers, who are less inclined to make capital expenditures, choose between distributing and stockpiling cash. Recent studies, such as Faleye (2004) and Harford, Mansi, and Maxwell (2008) find that firms with weaker governance structures are less inclined to accumulate a large stockpile of cash for fear of drawing the unwanted attention of activist shareholders. Therefore, in absence of active monitoring, firms with higher idiosyncratic risk are more likely to disgorge cash by paying dividends. On the other hand, firms with high idiosyncratic risk that also have stronger governance mechanisms are less likely to pay dividends to preserve cash for profitable future 3

4 investment opportunities. To summarize, governance acts as a substitute for a dividend policy when the risk of underinvestment is high. We also study how governance structure affects the propensity of firms to pay dividends when idiosyncratic risk falls. When idiosyncratic risk is low, self-interested risk-averse managers are no longer less inclined to spend cash internally or to use it for external acquisitions; therefore, they may be less inclined to pay dividends. However, Masulis, Wang, and Xie (2007) document that poorly governed acquirers have lower announcement returns during acquisitions. Biddle, Hilary, and Verdi (2009) find that firms with poor financial reporting quality have worse investment efficiency. Hence, firms with stronger governance are more likely to pay dividends to reduce cash that might lead to value-destroying investments, when the risk of underinvestment is low or the risk of overinvestment is high. To summarize, governance acts as a complement to dividend policy when the risk of overinvestment is high. We begin by investigating whether idiosyncratic risk matters in the dividend payout decision. We report 4,396 non-payers and 4,018 payers among the firms with low idiosyncratic risk and 7,424 non-payers and 992 payers among the firms with high idiosyncratic risk. The substantially higher (lower) number of non-payers (payers) among the high-risk firms is consistent with the central argument of Hoberg and Prabhala (2009). Next, in a simple univariate framework, we find that the payers and non-payers in the high- and low-risk categories are significantly different from each other in terms of corporate governance. We use Industry Corporate Governance Quotients (ICGQs) obtained from Risk Metrics/Institutional Investors Services (ISS) as our primary governance measure. This database covers a large number of unique firms (5,232) from 2003 to Because Risk Metrics/ISS began to significantly restructure the manner in which the governance variables are developed in 2010, the sample 4

5 period is restricted. In robustness tests, we extend the sample period by introducing the following alternative proxy measures for corporate governance: inversed BCF index, inversed ATI index ( ), state-level governance index, and BC Law passages ( ). We report that the difference in median governance between payers and non-payers among the firms with lower idiosyncratic risk is 9.30 percentile points, which is substantially higher than the 2.70 percentile point difference observed for firms with higher idiosyncratic risk. The results are indicative of firms with stronger governance and lower idiosyncratic risk being more inclined to pay dividends. Among firms with high idiosyncratic risk, the considerably smaller percentile point difference in median governance quality between payers and non-payers may be indicative of governance being substituted for dividend policy when idiosyncratic risk is high. This leads us to examine the possible asymmetric effect of corporate governance on dividend payout decisions that hinge upon the level of idiosyncratic risk faced by the firm, in a multivariate setting. We find that a firm in the top governance quintile is 38.27% less likely to pay dividends than a firm in the bottom quintile when they both belong to the highest idiosyncratic risk quintile, supporting the aforementioned substitute theory. By contrast, a firm in the top governance quintile is 14.79% more likely to pay dividends than a firm in the bottom quintile when they both belong to the lowest idiosyncratic risk quintile, supporting the complement theory. Our results remain robust to alternative measures of idiosyncratic risk and corporate governance as well as a number of variations of the baseline model. The endogeneity of firm characteristics and outcomes poses a challenge to empirical corporate finance research, and our paper is no exception. We use exogenous changes of firm governance around the adoption of antitakeover legislation in their states of incorporation to address such issues. Our methodology closely follows Bertrand and 5

6 Mullainathan (2003). We do not find any evidence of bias due to potential endogenous concerns in our previously documented results. To further test the underinvestment and corporate governance as dividend substitute hypothesis, we study how managerial risk aversion affects the dividend policy of our sample firms experiencing varying degrees of idiosyncratic risk. We find that a firm with a high idiosyncratic risk and strong governance mechanisms is even less likely to pay dividends as the risk aversion of the CEO rises. In other words, the dividend substitution effect of corporate governance is more pronounced in firms that have more risk-averse CEOs. This evidence supports the underinvestment hypothesis. To supplement the initial investigation of the overinvestment and corporate governance as dividend complement hypothesis, we examine how free cash flow, a variable that is associated with conditions that foster value-destroying investment or overinvestment, affects the dividend policies of our sample firms. We find that firms with a low idiosyncratic risk and strong governance mechanisms show an even greater inclination to pay dividends when presented with a rising level of free cash flow. In other words, the dividend complement effect of corporate governance is stronger for firms that have more opportunities to engage in value-destroying investments. This evidence lends support to the overinvestment hypothesis. We extend our investigation to the level of payout and the propensity for other payout policy decisions, such as share repurchase and total payout. The results are consistent with our findings on the propensity to pay dividends. We find corroborating evidence in the level of dividend payout decisions as well. Additionally, untabulated results reveal the likelihood of dividend increases being attributed to firms with stronger governance and lower idiosyncratic risk. However, these firms are also more likely to omit dividends when idiosyncratic risk 6

7 increases. These results support the idea that firms with stronger governance exhibit greater prudence in shaping their dividend policy. In summation, we find evidence supporting the asymmetric role of corporate governance in payout decisions under different levels of idiosyncratic risk, which is instrumental in reconciling the debate over the complement and substitute theories. The rest of the paper is organized as follows. Section 2 reviews the related literature and develops our hypotheses. Section 3 describes the data sampling process and provides summary statistics. Section 4 examines the joint effect of governance and idiosyncratic risk on the propensity to pay dividends and tests the robustness of the documented results. Section 5 investigates the link between underinvestment and corporate governance acting as a dividend substitute through the lens of managerial risk aversion. Section 6 examines the association of overinvestment concerns with corporate governance acting as a dividend complement within the ambit of agency cost of free cash. Section 7 analyses the joint effect of governance and idiosyncratic risk on the level of payout and the propensity to repurchase shares and total payouts. Finally, Section 8 concludes. 2. Hypothesis Development and Literature Review An extensive body of literature studies the relation between governance quality and dividend policy through the ambit of agency conflicts. However, whether weakly governed firms are more inclined to pay dividends to reduce the suboptimality of their cash-holding and financial investments has been a topic of frequent debate. La Porta et al. (2000) arguably open the complement versus substitute debate between corporate governance mechanisms and dividend policy in disciplining managers. They propose and test the outcome model and the substitute model of dividends. In their outcome model, minority shareholder rights are associated with a 7

8 higher dividend payout, which implies that corporate governance complements dividend policy. Dividend payout can be the result of effective governance; in such cases, strong governance forces management to pay out to mitigate the overinvestment problem, ensuring that corporate governance and dividend payout complement each other. By contrast, other governance mechanisms may be employed to effectively control the behavior of managers; in these cases, corporate governance can act as a substitute for a dividend payout. The substitute argument predicts that the dividend payout is negatively related to the quality of corporate governance, whereas the complement argument predicts a positive relation. Evidence on the relation between dividend payout and governance mechanisms is mixed. In this paper, we reconcile these competing views by highlighting the impact of idiosyncratic risk on the relation between corporate governance and dividend policy. Idiosyncratic risk has been linked to underinvestment in the recent investment literature. 1 DeMarzo et al. (2012) and Panousi and Papanikolaou (2012) develop theoretical models that predict decreasing investment with increasing idiosyncratic risk under the assumption of managerial risk aversion. Galai and Masulis (1976) and more recently Cao, Simin, and Zhao (2008) have shown that idiosyncratic risk is related to corporate growth options. In particular, growth-option proxies have been documented to explain63% of the variation in idiosyncratic volatility. We test the following hypothesis that connects the risk of underinvestment, governance and the structure of a firm s dividend policy. Underinvestment and Corporate Governance as Dividend Substitute Hypothesis: Firm s governance structure responds to the risk of underinvestment driven by selfinterested and risk averse managers, by discouraging payouts. This hypothesis predicts that a 1 Idiosyncratic risk has also been linked to payout propensity (see Hoberg and Prabhala, 2009). 8

9 firm with strong corporate governance is less likely to pay dividends when the idiosyncratic risk is high. When faced with a higher idiosyncratic risk, managers are less inclined to invest. However, instead of stockpiling cash, these managers are eager to dispose of it in the form of payouts to avoid conflicts with shareholders over it (Faleye, 2004, Richardson, 2006, Harford et al. 2008). Therefore, under these circumstances, governance mechanisms discourage payouts to preserve funding for future value-enhancing investments because rising idiosyncratic risk increases the value of growth options while simultaneously distorting the optimal investment policy of riskaverse managers. The underlying assumption that the cash reserves, viz., retained earnings, will eventually be used towards profitable investment is founded on the extant literature that suggests a positive relation between governance, cash reserves, and investment efficiency. John, Litov, and Yeung (2008) report that better investor protection (corporate governance) leads to riskier (but value-enhancing) investments. Biddle et al. (2009) find that firms with poor financial reporting quality have worse investment efficiency. 2 The underinvestment and corporate governance as substitute hypothesis of dividend policy is based on the premise that idiosyncratic risk leads to underinvestment under the assumption of managerial risk aversion proposed by Panousi and Papanikolaou (2012). In a number of related studies, CEO risk aversion has also been linked to less risky policy decisions, increased debt value, and declining equity and enterprise values (Sundaram and Yermack, 2007, Wei and Yermack, 2011, and Cassell, Huang, Sanchez, Stuart, 2012). Managerial risk aversion and habit formation in a manager s utility function are also shown to lead to dividend smoothing 2 Even in the absence of profitable real investment opportunities in the near future, a firm ideally retains its earnings for precautionary needs or payouts to shareholders. However, some recent studies find that despite a firm s need to save cash for precautionary motives, poorly governed firms would suffer from a lower value of cash-holding, and they tend to have lower returns on their financial investments (Dittmar and Mahrt-Smith, 2007, Harford et al. 2008, and Duchin, Gilbert, Harford, and Hrdlicka, 2017). 9

10 (Lambrecht and Myers, 2012 and Lambrecht and Myers, 2017). In addition, Brav, Graham, and Michaely (2005) and Daniel, Denis, and Naveen (2009) report that managers would rather reduce investments than dividends. An analysis conducted by S&P Capital IQ shows that firms belonging to the S&P 500 index over the decade between 2003 and 2013 increased their spending on dividends and buybacks from 18% to 36% of their operating cash flow while reducing investment in plants and equipment to 29% from 33% of operating cash over the same period. 3 A recent study by Caliskan and Doukas (2015) also documents that risk aversion (proxied by CEO inside debt) induces CEOs to pay dividends, whereas convex CEO compensation decreases dividend payout. Therefore, we argue that as idiosyncratic risk rise, CEOs who are more risk averse will further magnify the risk of underinvestment by avoiding profitable and risky growth options and also by distributing cash, driven by their increased need to smooth dividends. Evidence consistent with this empirical prediction will lend further support to the underinvestment hypothesis. We also test the following hypothesis that connects the risk of overinvestment, governance and the structure of firms dividend policy. Overinvestment and Corporate Governance as a Dividend Complement Hypothesis When the risk of underinvestment falls, a firm s governance structure responds to the risk of overinvestment that arises from agency conflicts, by encouraging payouts. The hypothesis predicts that a firm with strong corporate governance is more likely to pay dividends when the idiosyncratic risk is low. Overinvestment problems arise when managers engage in value-destroying behavior, such as incurring lavish expenses or making poor acquisitions driven by empire building urges. 3 The Wall Street Journal 5/26/

11 Dividend payout has been shown to be a sign of effective governance in that strong governance forces management to make payouts in order to mitigate this type of overinvestment problem, ensuring that corporate governance and dividend payout complement each other. When idiosyncratic risk is low, the likelihood of underinvestment falls. Under these circumstances, governance mechanisms encourage payouts, thereby reducing the possibility of overinvestment. Since the study by Jensen (1986), free cash flow has often been linked to the severity of the overinvestment problem, such as wasteful managerial spending. Therefore, free cash flow lends itself well to acting as a proxy for overinvestment. When idiosyncratic risk is low and the likelihood of underinvestment is low as a consequence, rising free cash flow exacerbates the risk of overinvestment. A combination of these factors will prompt better governed firms to pay dividends. Evidence consistent with this empirical prediction will lend further support to the overinvestment hypothesis. Allen and Michaely (2003) note the growing importance of repurchase since Skinner (2008) reports that the positive relation between earnings and dividend payout observed in the past is now also observed for total payout, which is defined as dividend plus repurchase, suggesting that firms consider both dividend and repurchase as effective ways of paying out earnings. We extend our study by investigating whether our earlier finding that governance either complements dividend policy or substitutes for it based on the level of idiosyncratic risk is corroborated by models predicting the probability of both repurchase and total payout. Given that dividends are considered to be more effective than share repurchase in disciplining managers (Jagannathan, Stephens, and Weisbach, 2000; Guay and Harford, 2000), we expect our results to be weaker for share repurchase. The level of payout has been included in studies such as La Porta et al. (2000), Hu and Kumar (2004), and Brockman and Unlu (2009) that investigate the 11

12 governance as dividend substitute and complement theories. We extend our study to also include the level of payout and expect to find corroborating evidence of the predicted asymmetric role of corporate governance. 3. Data and Summary Statistics 3.1. Sample Selection We analyze the dividend decisions of firms from 2003 to Our sample period begins in 2003 and ends in 2009 because of the unavailability of consistent ICGQs, which represent the primary governance measure in this study. We also use multiple alternative governance measures based on firm-level shareholder rights, the E-Index (Bebchuk, Cohen, Ferrel, 2009), the ATI Index (Cremers and Nair, 2005) and state-level shareholder rights (John et al., 2015) to verify the robustness of our findings. 4 ISS started publishing ICGQs in 2001, but the coverage was not comprehensive until Moreover, ISS significantly restructured the manner in which its governance variables are developed after Our sample consists of the firms for which ISS collects data on governance standards and for which security market and accounting data are available in the Center for Research in Security Prices (CRSP) and Standard & Poor s COMPUSTAT databases. To ensure that the firms are publicly traded, we include firms with a CRSP share code of 10 or 11. We require that the sample firms have price and shares outstanding available for December of year t as well as at least 100 trading days for estimating the risk measure for the same calendar year t. We exclude utilities (SIC Codes ) and financial firms (SIC Code ). We also require that the firms have the accounting data used to construct the variables for our analyses in COMPUSTAT for fiscal year t. 4 Detailed definitions of all variables are in the appendix. 12

13 RiskMetrics/ISS provides ICGQs that measure sample firms percentile positions within their GICS industry groups. 5 ICGQs are the most discernable commercial governance ratings based on 64 governance variables constructed from several dimensions of board structure (composition, independence), executive and director compensation, ownership, corporate audit, charter, bylaw provision, takeover defenses, progressive practices (performance reviews and succession plans), and director education. ISS covers the vast majority of publicly traded firms, approximately 5,000 firms every year over our study period. Thus, ICGQ is multidimensional and the most comprehensive and corporate governance measure available for a broad set of firms as compared to several academic governance measures that cover only the S&P 1500 companies. Notably, this rating adjusts for industry and size, and its algorithm also accounts for broader changes in market conditions. Therefore, ICGQ clearly has an advantage over check-and-sum academic measures that disregard the variation in governance standards and practices across industries or time-varying market changes. 6 The governance rank for a firm in year t is the ICGQ as reported in December of that year. The final sample in the payout propensity analysis consists of 16,830 firm-years over the sample period Variable Definitions and Summary Statistics The idiosyncratic risk of a firm is measured by the standard deviation of residuals from a regression of its daily excess stock returns on the market risk premium, following Hoberg and 5 The definition of the 24 GICS industry groups can be found at 6 We acknowledge that the index only reflects the within-industry quality of corporate governance. Thus, it is not necessarily a fitting indicator of the cross-industry quality of corporate governance. However, under a more realistic assumption that investment decisions and dividend decisions are not independent, one could expect an industry effect to exist for dividend decisions. Many studies recognize such an effect, including those by Lintner (1956), Michel (1979), Marsh and Merton (1987), Smith and Watts (1992), and Christie (1994). In particular, Smith and Watts (1992) use industry-level data to test dividend policy. According to Brav et al. (2005), 38.3% of executives indicate that the dividend policies of their competitors and other companies in their industries play an important role in their own dividend decisions. Therefore, we believe that using ICGQ could provide useful insight into how governance affects dividend decisions. Later in the paper, we test our main results using different governance proxies from the literature that are not industry-adjusted and obtain consistent results. 13

14 Prabhala (2009). We also use the idiosyncratic risk estimated using the Fama and French 3-factor and Carhart 4-factor models as well as the Campbell, Lettau, Malkiel, and Xu (2001) approach adjusting for market and industry. Our results remain robust to these alternative measurements of idiosyncratic risk. The majority of firm-specific controls are motivated by Fama and French (2001), DeAngelo et al. (2006), and Hoberg and Prabhala (2009). These variables include firm size, market-tobook, asset growth and profitability. We use NYSE as our firm size proxy to make our results comparable to those reported in previous studies on dividend propensity. Market-to-book is measured as the ratio of the book value of assets minus the book value of equity plus the market value of equity, all scaled by the book value of assets. Asset growth from year t-1 to year t is defined as the ratio of the book value of assets in year t over the book value of assets in year t-1 minus 1. Profitability is defined as earnings before extraordinary items plus interest expense plus deferred income taxes, all scaled by the book value of assets. RE/TE is measured as the ratio of retained earnings to total common equity winsorized at the 5 th and 95 th percentiles. TE/TA is measured as the ratio of total common equity to total assets, a variable commonly used in the literature to account for the effect of the firm s capital structure on dividend policy. Both of these ratios are included in the analysis. We add four other control variables in our main logit models. Ln (Firm Age) is the natural logarithm of the number listing years of the firm; RD/TA is research and development expenses divided by the book value of assets; Neg. Earnings is a dummy that equals 1 if the firm s earnings are negative at year t, and 0 otherwise. Product Fluidity is the local product market fluidity defined in Hoberg, Philips, and 14

15 Prabhala (2014), which is a measure of the intensity of change in the product market around a firm each year. 7 In our later tests, we introduce several managerial and additional firm characteristics. We use three proxies of managerial risk aversion: CEO tenure (Berger, Ofek, and Yermack, 1997), CEO inside debt and CEO relative leverage (Sundaram and Yermack, 2007, Wei and Yermack, 2011, and Cassell et. al., 2012). CEO inside debt for year t is equal to the sum of the total aggregate balance in deferred compensation plans at the end of the fiscal year and the present value of accumulated pension benefits from all pension plans. CEO relative leverage is equal to the CEO leverage divided by the firm leverage where CEO leverage equals the CEO inside debt divided by CEO equity holdings, which is defined as the sum of value of the CEO s common stock holdings, the value of the CEO s unvested stock, and the value of the CEO s stock options. Finally, we introduce two measures of free cash flow (hereafter, FCF). Our first FCF proxy is defined as operating income minus total income taxes plus the change in deferred taxes minus total interest expenses, all scaled by the book value of assets, a pre-dividend measure reported by Lehn and Poulsen (1989). Our second FCF proxy is industry-adjusted internally financed growth, as defined by Leuz, Triantis, and Wang (2008). The proxy is defined as ROA/(1-ROA) minus the median asset growth of the firm s GICS industry group. 8 Firm characteristics of dividend non-payers and payers are presented in Table 1 further sorted into two groups based on idiosyncratic risk. The full sample consists of 16,830 firm-years from 2003 to The idiosyncratic risk of a firm is measured every calendar year. The low (high) idiosyncratic risk subsample consists of firms with idiosyncratic risk lower (higher) than the sample median for each calendar year. A firm-year observation is classified as a dividend 7 We thank the Gerard Hoberg and Gordon Philips for providing the data on their website 8 Sample and variable description are summarized in the Appendix. 15

16 payer if the amount of the dividend paid by a firm is positive in a fiscal year, thereby segregating the full sample into 4,396 non-payers and 4,018 payers among the firms with low idiosyncratic risk and 7,424 non-payers and 992 payers among the firms with high idiosyncratic risk. The substantially higher (lower) number of non-payers (payers) among the high-risk firms is consistent with the central argument of Hoberg and Prabhala (2009). We first examine the average governance index (ICGQ) of the payers and non-payers among the firms segregated by idiosyncratic risk. At first glance, the results of difference tests based on t-tests for equality of means and a Wilcoxon test for equality of medians indicate that the firms governance among the dividend payers is higher than governance among the non-payers, suggesting that better governed firms are motivated to return money to investors regardless of the risk category. This finding supports the hypothesis of governance acting as a complement to dividend policy. However, upon closer inspection, the difference in median governance between payers and nonpayers among the firms with lower idiosyncratic risk is 9.3 percentile points, which is substantially higher than the 2.7 percentile points observed for firms with higher idiosyncratic risk. The results are consistent with our empirical prediction that better governed firms are more inclined to pay dividends when their idiosyncratic risk is low. Among the firms with high idiosyncratic risk, the considerably smaller difference of 2.37 percentile points in median governance quality between the payers and non-payers may be indicative of governance substituting for dividend policy when idiosyncratic risk is high. Prior dividend literature documents that dividend payers and non-payers differ fundamentally across various marketbased and accounting variables. Other factors deemed critical in the dividend policy literature, such as market-to-book, asset growth, profitability, size (Fama and French, 2001), retained earnings to total equity and total equity to total assets (DeAngelo et al., 2006), differ significantly 16

17 between the payers and non-payers across the idiosyncratic risk categories in a manner consistent with the literature. In particular, the group of dividend payers comprises larger, older and more profitable firms with a higher earned/contributed capital mix (a proxy indicating the matured stage in the life cycle of a firm), whereas the non-payers comprise firms with a higher market-tobook ratio (a commonly used proxy for growth opportunities), a faster rate of asset growth, higher R&D expenses, and stronger product market threats. [Insert Table 1] 4. Corporate Governance, Idiosyncratic Risk, and the Likelihood of Dividend Payout We conduct a pairwise correlation analysis for all variables that are deployed in most of our multivariate analyses of dividend propensity. The Pearson correlation coefficients and Spearman rank correlation coefficients are reported in Table 2. [Insert Table 2] We inspect the correlation coefficients to ensure that our variables do not suffer from a severe case of multicollinearity that may affect the validity of our multivariate regression analyses. Firm size and ICGQ show a moderately high positive correlation at 48.4%, which is not surprising because larger firms, subject to greater market scrutiny, are more likely to adopt better corporate governance practices. However, the other pairwise correlation coefficients have expected signs but are too small in size to raise any concern. In particular, the correlation coefficients between ICGQ and idiosyncratic risk, our main measure of interest, range between % and -33.9%. This finding suggests that the two measures are closely related in that better governed firms are less risky in general. However, we find that the condition index is less than 10 in our diagnostic test, which indicates that multicollinearity is not severe Baseline Model 17

18 In this section, we test the asymmetric effect of corporate governance on dividend payout decisions hinged upon the level of idiosyncratic risk faced by the firm in a multivariate setting. The empirical model follows the dividend propensity literature, particularly the studies of Fama and French (2001), DeAngelo et al. (2006), Hoberg and Prabhala (2009), Hoberg et al. (2014). [Insert Table 3] Table 3 reports estimates of a logit model that predicts the propensity of a firm to pay dividends based on its other characteristics, as described in the data section. The dependent variable Payer Indicator is equal to 1 for dividend-paying firms, and 0 otherwise. 9 Previous studies document an industry effect in dividend payout decisions (Lintner, 1956, Marsh and Merton, 1987, Smith and Watts, 1992). CFOs also recognize the existence of these types of industry effects (Brav et al., 2005). To examine whether our results are robust to industry effects and market-wide shocks, our pooled logit regressions include year and 24 GICS industry fixed effects; the standard errors are clustered by firm. In column 1 of Table 3, the governance index predicts higher odds of dividend payout in the absence of idiosyncratic firm risk control, which supports the complement theory. However, in column 2, when we add idiosyncratic risk, the governance index demonstrates a marginally significant negative relation with the propensity to payout, evidence that supports the substitute theory. Additionally, idiosyncratic risk is significantly negatively related to the propensity to payout, supporting the findings of Hoberg and Prabhala (2009). This result also indicates that the way governance interacts with risk in dividend payout decisions merits further exploration. In this specification, we also add RE/TE, which is measured as the ratio of retained earnings to total 9 We obtain qualitatively similar results using Fama and MacBeth (1973) style estimates with Newey West adjustments. The results are available upon request. 18

19 common equity, winsorized at the 5 th and 95 th percentiles, and TE/TA. 10 We find that RE/TE is positively related to payout propensity, which supports the findings of life cycle theory advanced by Grullon et al. (2002), DeAngelo and DeAngelo (2006), and DeAngelo et al. (2006), who observe a negative association between dividend payouts and investment opportunities related to the firm s life cycle stage, which is proxied by earned/contributed capital mix. In column 3, we test our premise of the asymmetric likelihood of payout by firms at varied levels of corporate governance under changing incentives of investment distortions due to idiosyncratic risk by adding an interaction term between ICGQ and idiosyncratic risk. We find our variable of interest to be significantly negative. As previously discussed, the underlying idea is that lower idiosyncratic risk implies a lower likelihood of underinvestment and possibly even a likelihood of overinvestment, while higher idiosyncratic risk implies a higher underinvestment problem. The empirical predictions of the under- and overinvestment hypotheses, also discussed earlier, are that the relation between corporate governance and dividend payout propensity is positive when the idiosyncratic risk is low but becomes less positive or even negative as the idiosyncratic risk increases. Our results in column (3) support these predictions. 11 Additionally, similar to the univariate results reported in Table 1, larger, more profitable firms with a higher proportion of retained earnings show a higher likelihood of paying dividends, whereas firms with a higher market-to-book ratio (a commonly used proxy for growth opportunities) and a faster rate of asset growth show a lower likelihood of paying dividends, which is evidence consistent with the prior literature. 12 Finally, we add four other control variables: Ln(Firm Age), RD/TA, Neg. Earnings, 10 The results are similar if the RE/TE is winsorized at the 1 st and 99 th percentile. However, we have a fairly large number of negative RE/TE observations, which drives the mean to -0.68, significantly lower than the median of This may be responsible for the predicted probability being largely undervalued when every variable is at its mean. 11 We also test our corporate governance, idiosyncratic risk and dividend payout hypotheses using probit and linear probability models. All results remain consistent with our original findings. 12 In similar regression analyses, we find that a three-way interaction term between idiosyncratic risk, governance (ICGQ and some alternative measures used in Table 6), and M/B is significantly negative, supporting the notion that 19

20 and Product Fluidity in column (4). While our main findings from column (3) remain unaffected, older firms, firms with higher R&D expenses, and stronger product market threats show a lower likelihood of paying dividends. To determine the economic significance of our results, we rank each independent variable each year and partition the resulting ranks into quintiles labeled from 1 (lowest quintile) to 5 (highest quintile). We re-estimate our dividend payout model using the quintile ranks and report the coefficients in Panel A of Table 4. The results from these quintile regressions further confirm that the impact of governance on dividend payout propensity decreases with idiosyncratic risk. Moreover, the results suggest that our findings are not driven by the extreme values of any independent variable. [Insert Table 4] In the first two rows of Panel B in Table 4, we present the predicted probabilities of dividend payout using the model specified in Panel A. These values represent the probability of a firm paying dividends when its corporate governance belongs to the highest and the lowest ICGQ quintiles for the five idiosyncratic risk categories. The control variables are held constant at their medians to calculate the predicted probabilities. For both the strong and weak governance firms, we observe that the predicted probability declines with increasing idiosyncratic volatility, which highlights the importance of idiosyncratic risk in determining payout policy, which is consistent with Hoberg and Prabhala (2009). The third row of Panel B reports the percentage change in the predicted probability of dividend payment by a firm when it moves from the 1 st (lowest) to the 5 th (highest) quintile of the ICGQ ranking for each idiosyncratic risk quintile. We find that when idiosyncratic volatility faster growing firms have more investment opportunities and will therefore show a relatively greater propensity to preserve cash or a lower propensity to pay out dividends., particularly when the cost of underinvestment is high. These results are not reported for the sake of brevity. They are available upon request. 20

21 is at its lowest quintile, moving from the 1 st quintile to the 5 th quintile of governance increases the predicted probability of dividend payout by 14.79% (= ( )/21.49). However, when the idiosyncratic volatility is at its highest quintile, moving from the 1 st quintile to the 5 th quintile of governance decreases the predicted probability of dividend payout by 38.27% (= ( )/7.29). Collectively, these findings suggest an economically meaningful asymmetric impact of governance on dividend payout propensity under low and high levels of idiosyncratic risk. This supports the conjecture that governance acts both as a complement to and a substitute for dividend policy subject to the idiosyncratic risk of a firm Robustness In Table 5, we perform additional tests to ensure the robustness of our previously documented results. First, we investigate whether our results are maintained using an alternative proxy of idiosyncratic risk to address the concern that firm-specific risk may be systematically different across industries. In the spirit of Campbell et al. (2001), the alternative proxy of idiosyncratic risk is estimated as each firm s residual volatility after accounting for market and industry volatility. The interaction term remains negatively significant, suggesting that our results are robust to an alternative idiosyncratic risk measure. 13 To ensure that our results are not driven by smaller firms that are less likely to pay dividends, we test our model exclusively on S&P 1500 firms in column 2. Even though the statistical significance slightly decreases, the economic significance of our results remains unchanged. Bliss, Chen, and Denis (2015) find a spike in cash retention by firms in response to a supply of credit during financial crises. In column 3, we investigate whether our results merely reflect firms reaction to the recent financial crisis in that better governed firms choose not to pay dividends when risk is high, while poorly governed firms simply go out of business. We exclude the observations from 2007 to 2009 and find 13 We obtain consistent results when idiosyncratic risk is estimated using the multifactor Fama and French model. 21

22 stronger statistical significance of the interaction term between governance and risk. In column 4, we explore the impact of the Jobs and Growth Tax Relief Reconciliation Act (JGTRRA) of 2003, which reduces the maximum tax rate on dividends and therefore makes dividends more attractive to investors. We drop 2003 to ensure that our results are free of any effect of the JGTRRA. The interaction between governance and risk remains significantly negative, suggesting that our findings are not affected by the JGTRRA. 14 Furthermore, all variables in our main model belong to the same fiscal year, which makes our results comparable to those in the current literature. However, it is likely that dividend payouts affect the firm s risk as well as its governance. To account for such a concern of reverse causality, we measure governance and risk in the year prior to dividend distribution in column 5 and measure every other independent variable in the year prior to dividend distribution in column 6. Our results remain robust with an even stronger statistical significance than before, indicating that they are not driven by reverse causality. However, we are cautious about this interpretation given how persistent corporate governance is. Columns 1 through 6 produce the most notable result as ICGQ, despite independently having a significantly positive impact on the propensity to pay dividends, is subsumed by the interaction term between itself and idiosyncratic risk. This finding supports our prediction that the relation between corporate governance and dividend payout propensity changes from positive to negative as the idiosyncratic risk increases. [Insert Table 5] 4.3. Alternative Governance Proxies Academic researchers and industry experts often find it difficult to agree on the validity of measures of corporate governance (see Gompers, Ishii, and Metric, 2003; Core, Guay, and 14 In unreported analyses, we add a dummy equal to 1 if the observation is in 2003 in a panel regression with industry fixed effects and find the dummy to be significantly positive. This result suggests that a change in the dividend tax has a significant impact on the propensity of paying dividends, yet our results are not affected by it. 22

23 Rusticus, 2006; Bebchuk et al., 2009; Johnson, Moorman, and Sorescu, 2009; and Daines, Gow, Larcker, 2010). Given the contentious nature of this literature, we next investigate the extent to which our findings are dependent on our choice of corporate governance measure, ICGQ, and the restricted sample period over which the measure is available. To that end, we conduct tests in which we use samples collected over extended periods and alternative proxies of corporate governance borrowed from the prior literature. The inversed BCF index equals 6 minus the E-index where the E-index is the entrenchment index constructed by Bebchuk et al. (2009) and obtained from the RiskMetrics Governance database from 1989 to A higher E-index implies more entrenchment and potentially worse governance; thus, the higher the inversed BCF index is, less entrenched the managers are. The inversed ATI index equals 3 minus the ATI index where the ATI index is the alternative takeover defense index constructed by Cremers and Nair (2005) and also obtained from RiskMetrics Governance database from 1989 to A higher ATI index implies stronger takeover defense and potentially worse governance; therefore, we implement a linear transformation of the ATI index for ease of interpretation. [Insert Table 6] Columns 1 and 2 of Table 6 report the results of the dividend propensity model using an inversed BCF index as a proxy for corporate governance without and with the interaction term between governance and idiosyncratic risk. Columns 3 and 4 of Table 6 report the results of the dividend propensity model using the inversed ATI index as a proxy for corporate governance without and with the interaction term between governance and idiosyncratic risk. While the coefficient of the governance proxy is negative in column 1, it is positive in column 3. These 15 We do not use the G-index from Gompers et al. (2003) because the G-index cannot be constructed after 2007 due to a change in data collection by Riskmetrics, and it is critical that we use alternative governance proxies that can be continuously measured at least over our original sample period. 23

24 findings once again show the difficulties in finding consistent empirical evidence of the relation between governance and dividend propensity, as documented in the literature. However, we find that the coefficients of the interaction term between governance and idiosyncratic risk are negative in both columns 2 and 4. The other control variables behave as they did in the original dividend propensity model. To summarize, the alternative governance proxies show a negative joint effect of governance and idiosyncratic risk on dividend payout decisions, which suggest that our previous findings are not confounded by the use of alternative proxies of corporate governance over extended sample periods Endogeneity Concerns Despite the robustness of our proposed dividend propensity model, our interpretation of the results, which indicate an asymmetric impact of corporate governance on dividend payout dependent on the level of idiosyncratic risk, is subject to a host of endogeneity biases. For example, firms simultaneously select a corporate governance structure and financial policies, potentially confounding any analysis of a causal relationship. Despite the comprehensive nature of ICGQ as a measure of corporate governance, its restricted period of data availability poses challenges in identifying events that might result in any exogenous changes in its levels. Therefore, in this section, we use the adaption of antitakeover laws at the state level to identify exogenous shocks to corporate governance quality of firms. With the exception of the firms that lobby and effectively influence the formulation of the legal environment of their states of incorporation, state-level antitakeover laws generally cannot be endogenously changed by firms, which arguably makes the state-level governance index a measure that captures cross-sectional differences in corporate governance that are largely externally determined. We exclude lobbying and motivating firms in all of our tests using this measure. We construct the state-level governance index following John et al. (2015). The state-level governance index equals 5 minus the antitakeover laws index constructed 24