Political uncertainty and dividend policy: Evidence from international political crises

Transcription

1 Political uncertainty and dividend policy: Evidence from international political crises Tao Huang, Fei Wu, Jin Yu, Bohui Zhang April 4, 2013 Abstract We examine the impact of political uncertainty on firms payout policy. Using a large international sample across 35 countries over the period from year 1980 to 2009, we find that past dividend payers are more likely to terminate dividends and that non-payers are less likely to initiate dividends during periods of high political uncertainty. These findings suggest a precautionary incentive of managers in response to political shocks. Furthermore, the impact of political shocks can be attenuated by stable political systems. In addition to identifying this precautionary incentive, we also find that firms with lower market valuation or less liquidity are more likely to initiate dividends during periods of high political uncertainty, which is consistent with the catering motive of managers. Jiangxi University of Finance and Economics ( taohuang@jxufe.edu.cn) Jiangxi University of Finance and Economics - International Institute for Financial Studies ( fwu@jxufe.edu.cn) University of New South Wales ( jin.yu@unsw.edu.au) University of New South Wales ( bohui.zhang@unsw.edu.au)

2 1 Introduction It is important to understand the determinants of corporate payout policy. Managerial perceptions regarding the uncertainty of firms future earnings, one of many driving factors, has been shown to have an influential effect on payout policy. 1 However, inferences about the importance of uncertainty on dividend payout policy in existing studies are generally limited by two factors: (1) financial measures of uncertainty can be contaminated by look-ahead bias or can be sensitive to omitted-variable errors; 2 and (2) the causal relationship between cash flow uncertainty and corporate dividend decisions is not completely clear. 3 To tackle these issues, we explore the impact of international political crises on dividend payout decisions. Considering international political crises is ideal for several reasons. First, political crises are exogenous to financial markets (see Berkman, Jacobsen, and Lee (2011)) and thus provide a clean identification strategy that allows us to analyze the true incentives for payout decisions. Second, political crises have a direct impact on the level of uncertainty of firms future earnings. 4 Third, because of the potentially substantial heterogeneity in how 1 See the literature based on survey evidence by Linter (1956), and Brav, Graham, Harvey, and Michaely (2005). The empirical evidence presented by Hoberg and Prabhala (2008) and Chay and Suh (2009) indicates that firm-level risk measures, as a proxy for cash ow uncertainty, explains a large part of both the crosssectional and the time-series variations in the probability of paying dividends. Examining the market-wide uncertainty (measured using the Chicago Board Options Exchange Volatility Index, VIX), Walkup (2011) documents a similar finding. 2 Managers could look ahead and forecast future economic conditions. For example, Edmans, Goldstein, and Jiang (2012) emphasize that most firm-related or financial variables are inappropriate for use in analyzing the true incentives for corporate decisions. Moreover, there are potential omitted variables that may simultaneously drive both the uncertainty of firms future earnings and dividend payout policy.for example, changes in the taxation affect both cash flow uncertainty and corporate dividend decisions 3 For example, reduction in dividend payments is associated with increase in the investment which leads to larger risk exposure to the business cycle 4 Political crises are likely to cause changes in the perceived rare disaster probabilities (Berkman, Jacobsen, 1

3 firms respond to exogenous shocks, political crises provide a unique platform for investigating the role of country characteristics and for determining firms various motives for dividend payout policy (see La Porta, Lopez-de-Silanes, Shleifer, and Vishny (2000) and Choy, Gul, and Yao (2011)). Using a comprehensive sample of 23,426 firms from 35 countries over 19 years, we examine whether and how the firms payout decisions are affected by international political crises. Our empirical designs are based on these firms dramatic payout events, namely dividend termination and initiation, in response to changes in the uncertainty level, as measured using several crisis indices. More specifically, we focus on firms that have been consecutively paying dividends over the past three years, i.e., past payers, and analyze how intensified political uncertainty affects these past payers probability of dividend termination. Analogously, we also examine how the likelihood of dividend initiation from firms that have never paid out dividends over the past 3 years, i.e., past non-payers, changes in response to intensified political uncertainty. Three findings emerge from this study. First, we show that past dividend payers tend to be more likely to terminate their dividend payouts and that past non-payers tend to be less likely to initiate dividend payouts during periods with more political crises and more and Lee (2011)) and the level of uncertainty associated with possible changes in government policy (see, e.g., Pastor and Veronesi (2012a) and Pastor and Veronesi (2012b)) and the business environment (see, e.g., Blomberg and Hess (2003)). Uncertainty has been documented as the key channel through which political events affect firm-level corporate decisions. For example, Julio and Yook (2012) and Durnev (2012) find that firms change their investments and investment sensitivity to price during election years when political uncertainty is presumably high. 2

4 severe crises. For example, our results show that for a past dividend payer, the ratio of the probability of terminating dividends to the probability of continuing to pay dividends increases by approximately 7.39%, on average, for a one-standard-deviation increase in the number of crises. Similarly, for a one-standard-deviation increase in the number of crises, for the average past dividend non-payer, the ratio of the odds of initiating new dividend payouts to the odds of continuing not to pay dividend decreases by 4.15%. Analyzing dividend termination and initiation defined based on a firm s past three-year dividend history may only lead us to identify dramatic and relatively rare changes in dividend policy. We have performed several robustness checks to relax this definition. Using three different dividend payout-level variables, we extend our analysis by examining the effect of political uncertainty on incremental changes in dividend policy. We find a significantly negative relationship between the intensity of political uncertainty and the change in the dividend ratio. Next, we use alternative definitions of past dividend-payers and non-payers based on different time intervals, i.e., the past one year and the past two years. We again obtain qualitatively consistent results. Because we construct an aggregate crisis severity index using six individual indices that capture different aspects of crisis severity, 5 we also show that our results are robust to the use of individual crisis severity indices. Last, we employ three methods of addressing the concern that our results may be driven by the time-varying composition of the sampled 5 See Appendix A for detailed variable definitions. 3

5 firms rather than by political crises. We split our sample into sub-samples based on economic development, restrict our sample to a group of continuously listed firms (for years 2000 to 2008), and control for a time-trend variable. Our results remain robust to these checks. Second, our international sample of firms and crises allows us to assess how a country s political system affects the sensitivity of dividends to political uncertainty. In our study, we document substantial cross-country variation. In particular, we find that dividend decisions become less sensitive to political uncertainty in countries with more stable political systems. This result highlights the important role of institutional settings in mitigating the adverse effect of political uncertainty on individual firms. 6 Finally, we show that under-valued firms and firms with illiquid shares exhibit dividend behavior that differs widely from that of the average firm in face of political uncertainty. In fact, firms in the former two categories are more likely to initiate dividends. In general, dividend-paying firms experience a higher premium during periods of high political uncertainty. It may thus be optimal for under-valued firms to cater the market demand for dividends. We also find deteriorated liquidity for all firms during periods of high political uncertainty and determine that firms with illiquid shares are more likely to initiate dividend payouts because a dividend may be viewed as a substitute for liquidity. These results are consistent with the catering hypothesis of dividends (see Baker and Wurgler 6 Johnson, Boone, Breach, and Friedman (2000) report analogous evidence emphasizing the role of macrocorporate governance in the relationship between stock market performance and the Asian financial crisis of

6 (2004)). However, firms that initiate dividend payouts during high uncertainty periods do not appear to subsequently perform better than other firms. Moreover, firms with higher information asymmetry do not appear to be more likely to initiate dividend payouts. These results together do not generate support for the signaling hypothesis of dividends (see, e.g., Bhattacharya (1979), John and Williams (1985), and Miller and Rock (1985)). Our study adds to the literature on how firms vary their payout policy in response to changes in the business environment in which they operate. For example, a firm could decrease its dividend distribution to enhance its bargaining position with organized labor (DeAngelo and DeAngelo (1990)). Chetty and Saez (2005) and Chetty and Saez (2006) document that dividend payments and dividend initiation have increased since the 2003 U.S. dividend tax cut. Amihud and Li (2006) argue that increased stockholdings by institutional investors over time increase the amount of information that is incorporated into the stock price. Because using dividends as a signaling tool is costly, a decline in the information content of dividend announcements reduces firms propensity to pay dividends. In addition to changes in taxation and institutional ownership, our evidence confirms that the precautionary motives of managers deter dividend payouts under high uncertainty. We also contribute to the literature that examines firms strategic dividend behavior. In the ideal world of Modigliani and Miller (1958), a firm s payouts should not affect firm value. However, much of the empirical evidence suggests that firms deliberately manage their payout strategies. For example, firms initiate dividends strategically to signal good 5

7 prospects for future performance (Bhattacharya (1979), John and Williams (1985), and Miller and Rock (1985)), or to cater to market demand for dividends (Baker and Wurgler (2004)). To avoid market penalties, managers may decide to maintain a stable level of dividends by practicing a dividend smoothing strategy (see, e.g., Fama and Babiak (1968), Brav, Graham, Harvey, and Michaely (2005)) or a partially pooling strategy (Guttman, Kadan, and Kandel (2010)). We find that firms with lower market valuation or less liquid shares are more likely to initiate dividends during periods of high political uncertainty. Our finding strongly supports the catering incentive of managers and is hard to reconcile with the stickiness view and the signaling hypothesis of dividend payments. Finally, because we examine time-varying political effects on payout decisions, our study relates to the literature that contends that a firm s dividend policy is, to a certain degree, time-varying. A large number of studies, e.g., DeAngelo and DeAngelo (1990), Fama and French (2001), Baker and Wurgler (2004), Amihud and Li (2006), and Hoberg and Prabhala (2008), have established the importance of the time-series dimension to payout policy. In an investigation of payout decisions in our international sample, we observe that at least 30% of payout variation is attributable to the time dimension. 7 The remainder of the paper is organized as follows. In Section 2, we review the theoretical foundation for our empirical study and develop several testable hypotheses. We then describe 7 Specifically, we construct a dividend payout dummy that is equal to one if a firm pays dividends and zero otherwise for our sample of 112,151 international firm-year observations. We then employ a variance decomposition approach to separate the sample s total variation into a within-firm component and a between firm component (see, e.g. Graham and Leary (2011), for a similar analysis of corporate leverage policy). 6

8 our data sources, the sample construction process, and the empirical models in Section 3. Section 4 presents and discusses the impact of political crisis on dividend payout policy and the role of political system. Robustness checks are also performed in Section 4. In section 5, we analyze why and which firms strategically initiate their dividend payments during periods of high political uncertainty. Section 6 concludes the paper. 2 Theoretical Motivation and Hypotheses Uncertainty is a key channel through which political factors affect financial markets. We base our first hypothesis on two assumptions. First, during periods that feature political instabilities, the uncertainties associated with possible changes in government policies and in the macro environment may dramatically alter managerial risk perception (see, e.g., Pastor and Veronesi (2012a) and Pastor and Veronesi (2012b)). Second, firms tend to make conservative dividend decisions, and these decisions are largely affected by managerial perceptions of risk related to the firm s future cash flows (see, e.g., Linter (1956), Brav, Graham, Harvey, and Michaely (2005), Hoberg and Prabhala (2008), Chay and Suh (2009), and Walkup (2011)). These conservative dividend decisions may be due to two reasons. First, a precautionary motive makes riskier firms more reluctant to initiate or increase dividend payments. Rather, such firms may retain more cash to better deal with the expected future financial shortfalls. Second, external financing is usually more costly than internal financing and is even more so during periods of high uncertainty. These two effects together yield our first hypothesis. 7

9 H1. During periods of high political uncertainty, firms are more likely to terminate dividend payments and are less likely to initiate dividend payments, ceteris paribus. Nevertheless, a stable and efficient political system may help to attenuate the adverse effect of negative political shocks on corporate dividend policy. Stable political systems act as a buffer against political uncertainty for two reasons. First, stable political systems reduce the risk (likelihood) and the adverse impact of changes in government policy and the business environment. Second, stable political systems are generally correlated with better law and institutional provisions for corporate governance. Better investor protection reduces the agency costs of outside shareholders 8 and promotes more transparent information environments, 9 which may reduce the negative impact of uncertainty shocks. Therefore, our second hypothesis is as follows: H2. The dividend decisions of firms operating in countries with more stable political systems are less sensitive to political uncertainty, ceteris paribus. Even though we hypothesize that firms on average reduce dividend payments during periods of high political uncertainty, there may be some firms that initiate dividend payments deliberately during these periods. Our third hypothesis consists of a set of testable predictions regarding whether and why these firms are incentivized to initiate dividends during 8 See, e.g., La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1998), La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1999), and La Porta, Lopez-de-Silanes, Shleifer, and Vishny (2002) for evidence of the role of the political system in controlling agency conflicts. 9 See, e.g., Bushman, Piotroski, and Smith (2004) 8

10 periods of high political uncertainty. There are two possibilities. First, firms may initiate dividends to signal that they are good firms. Political turmoil not only increases the costs of initiating dividend payouts, but also increases uncertainty about future cash flows and, hence, the risk premiums of firms. Despite the scarce support for dividend signaling in the empirical literature, 10 the signaling theory of dividends (Bhattacharya (1979), John and Williams (1985), and Miller and Rock (1985)) predicts that firms with good prospects for future performance will be willing to increase their dividends and that it will be too costly for firms with poor prospects for future performance to mimic these well-positioned firms. Moreover, given that dividend signaling is costly, the firms with less effective channels for revealing their information, i.e., those with higher information asymmetry, should have stronger incentives to use dividends as a signaling device. According to the signaling theory of dividends, we can make the following predictions regarding dividend initiation. H3a. During periods of high political uncertainty, firms that expect good performance in the future and/or firms with a higher degree of information asymmetry are more likely to initiate dividends, ceteris paribus. Second, corporate managers may attempt to initiate dividends due to a catering incentive (see Baker and Wurgler (2004)). If investors place a valuation premium on dividend-paying 10 See, e.g., DeAngelo, DeAngelo, and Skinner (1996) and Grullon, Michaely, Benartzi, and Thaler (2005) for empirical evidence and see Brav, Graham, Harvey, and Michaely (2005) for survey evidence 9

11 firms during political crisis periods, it then becomes optimal for managers to cater to market demand by issuing dividend payments. There may be two reasons why investors are willing to reward dividend payers with a premium. First, Baker and Wurgler (2004) argue that a dividend premium may reflect the belief that dividend-paying stocks are less risky (They represent a bird-in-the-hand vs. a bird-in-the-bush ). If the risk tolerance of the bird-in-the-hand investors changes during periods of high uncertainty, their preferences between dividend-payers and -non-payers will change accordingly. This type of a demand for dividends during periods of high uncertainty may be interpreted as a flight to safety. Second, the risk-averse investors who expect the earlier liquidation of part of their investments as a means to manage increasing uncertainty will experience a substantial decline in stock market liquidity. Dividends can be viewed as an alternative mechanism for gaining liquidity (Banerjee, Gatchev, and Spindt (2007)). This logic yields the following predictions based on the catering theory of dividends. H3b. During periods of high political uncertainty, dividend-paying firms experience higher premiums. Firms with a low market valuation and/or illiquid shares are more likely to initiate dividends, ceteris paribus. 10

12 3 Data and Empirical Design 3.1 Data and Sample Our source of political crisis data is the International Crisis Behavior Project (ICB) database. The ICB is a database of international political crises that occurred during the period from year 1918 to The database contains detailed information on more than 400 international political crises. To ensure reliable coverage of accounting balance sheet data, we restrict our sample to the period from year 1990 to 2008, which includes 98 international political crises. We use Worldscope from Datastream to obtain wordwide annual firm-level accounting data. For the primary tests, we only include the firm-year observations if a firm either pays dividends for all of the previous three years or does not pay dividends for any of the previous three years. 11 Brazil, Chile, Colombia, Greece, and Venezuela have mandatory dividend rules, so we remove these countries from our sample. We exclude financial and utility firms because their dividend policies are usually regulated and hence are quite different from those of other industrial firms. We truncate our continuous variables at the 1st and 99th percentiles. We truncate variables at the 99th percentile if they have a lower bound at zero, e.g., the dividend-yield ratio (dy), the dividend-to-sales ratio (dvs), and the total payoutto-sales ratio (tps). In the end, our sample consists of 19,117 unique firms and 112, See the definitions of our dividend payout variables in the next subsection and in Appendix A. We expand our sample accordingly when we use the firms past two year (or one year) history of dividend payments. 11

13 firm-years across 35 countries, ranging from emerging markets (14 countries) to developed ones (21 countries). Appendix B summarizes the distribution of firm-year observations across countries and years. It is not surprising that the U.S. accounts for approximately 35% of total observations, followed by Japan and the U.K.. In our sample, the number of firm-years increases over time given the improved data availability in recent years. 3.2 Variables In this subsection, we introduce the variables used in our subsequent empirical tests. The detailed variable definitions are summarized in Appendix A. We start with the dependent variables, i.e., the dividend policy variables. The extant dividend literature documents a general time trend in which dividends have declined in recent decades (see, e.g., Fama and French (2001)). The level of dividend policy may contain a time-trend that leads to potential measurement errors if the omitted time-trend is not properly addressed. Moreover, these firms dividend payouts are relatively persistent and our sample covers a large cross section of firms. These facts create problems when we estimate the dynamic changes in dividend policy in response to political crises. To address these concerns, we decide to focus on the changes in dividend policy rather than on the policy itself. Therefore, our main dependent variable is dramatic changes in dividend policy: dividend termination and initiation. More specifically, we define the dividend termination dummy (dt) as follows. We first form a sample of firms that made their 12

14 dividend payments in all of the last three years. Within this sample of past-payers, we assign a value of one to the dividend termination dummy if a past-payer stops paying any dividends in the current year; otherwise, we assign it a value of zero. Similarly, we define the dividend initiation dummy (di) for a sample of past-non-payers according to the firm last three years dividend payments. We then assign a value of one to the dividend initiation dummy if a past-non-payer pays dividends in the current year and assign the dummy a value of zero otherwise. 12 In addition to examining dramatic changes in dividend policy, we use several variables that capture the change in the dividend-yield ratio ( dy), the change in the dividend-to-sales ratio ( dvs), and the change in the total payout-to-sales ratio ( tps) as robustness checks. Our key explanatory variable is political uncertainty. Presumably, the political uncertainty in a particular year is positively correlated with the number of crises occurring during the year. Therefore, our first measure of political uncertainty is the crisis index (crisis), which is equal to the total number of political crises occurring in a particular year. However, some crises may be more devastating and may introduce greater risk than others. Following Berkman, Jacobsen, and Lee (2011), we examine six types of crises with varying levels of severity: 1) a crisis that begins with a violent act, 2) a crisis that involves either serious clashes or a full-scale war, 3) a crisis that involves a full-scale war, 4) a crisis that involves grave value threats, 5) a crisis that is part of a protracted conflict, and 6) a 12 In our robustness checks, we also define the dividend termination and initiation dummies on samples based on the firms last two years, or one year s, dividend payments. 13

15 crisis that involves great powers or superpowers on both sides of the conflict. To capture the full severity of the political crises that occurred during a particular year, we summarize the different aspects of crisis severity into one aggregate measure of crisis severity (sevidx). More specifically, for each individual crisis, we compute a severity score by summing the above six indices and adding one. For example, a war with great power involvement that is part of a protracted conflict will have an individual severity score of four (one for being a crisis, one for being a war, one for having great power involvement, and one for being part of a protracted conflict). We aggregate the individual severity scores to obtain the severity index for all crises (sevidx) that occurred during the year. Because market-wide uncertainty is presumably positively correlated with the cumulative severity of the crises that occurred during a particular year, sevidx is our key measure for political uncertainty. We summarize our crisis variables in Appendix C. In our sample period, which includes the years , 98 crises occurred. The average number of crisis per year is The year 2008 has the lowest incidence of political crises with only one crisis while 1991 has the highest incidence of political crises, with ten crises. The severity index (sevidx) has a mean of Similarly, 2008 has the lowest severity index (2) while 1991 has the highest (34). Figure 1 shows the time-series patterns for the political uncertainty variables and the average dividend termination and initiation variables. Consistent with our first hypothesis, these graphs show that there is a positive relationship between dividend termination (dt) and political uncertainty (crisis/sevidx) and that there is a negative 14

16 relationship between dividend initiation (di) and political uncertainty (crisis/seridx). [Insert Figure 1 about here] In accordance with the extant dividend literature (see, e.g., Fama and French (2001), Baker and Wurgler (2004), and Chay and Suh (2009)), we control for a set of firm characteristics when we examine the impact of political crises on corporate dividend policy. The control variables used in our primary regressions include Tobin s q (q), asset growth (dta), firm size (mv), the life cycle (rete), return on assets (roa), cash holdings (cash), closely-held ownership (ch) and stock return volatility (std). We report the summary statistics for our key variables in Table 1. The first two rows of the table show that dramatic changes in corporate dividend policy are rare events, which is consistent with the sticky pattern of dividend payouts. For example, the mean of the dividend termination dummy is 0.044, which indicates that only 4.4% of firm-years include dividend termination. The percentage of firm-years that include the initiation of dividend payouts is approximately 5.7%. 13 [Insert Table 1 about here] Table 2 presents the correlation matrix for our primary variables. In particular, we show that dividend termination is positively correlated with our crisis measures. On the other 13 We include the summary statistics for the political crisis variables based on firm-year observations in the table for the sake of completeness. A detailed description of the crisis distribution is provided in Appendix C. 15

17 hand, dividend initiation is negatively correlated with our crisis measures. For example, the Pearson (Spearman) correlation coefficient between dt and crisis is 0.019(0.016) and that between di and crisis is (-0.010). In addition, the two crisis measures are themselves highly correlated, as we expect. We therefore decide to use the crisis measures separately in the regression analysis. Although the correlation matrix in Table 2 provides us with a preliminary view of the relationship between dividend policy and political crises, we also address the relationship in a multivariate regression context in the next section. [Insert Table 2 about here] 3.3 Empirical Specification In this paper, we attempt to analyze how dramatic changes in dividend payout policy occur in response to political crises. Therefore, we use Logit models to estimate the dividend termination and initiation decisions. Let a dividend decision Y {dt, di} be a binary response variable, let z {crisis, sevidx} be a political uncertainty measure, and let X be a vector that contains firm characteristics, country- and industry-level fixed effects, and a constant. A Logit model assumes that the odds ratio of the dividend termination decision takes the form P (dt = 1) 1 P (dt = 1) = exp(αz + X β), (1) 16

18 where α and β are coefficient estimates, exp(.) is the exponential function, and the odds ratio is the ratio of the probability of terminating the dividend payout (P (dt = 1)) to the probability of paying dividends (1 P (dt = 1)). Replacing dt with di in equation 1, we can similarly estimate a Logit model of the dividend initiation decision. As robustness checks, we also estimate multivariate linear regressions for various dependent variables that represent the change in the payout ratio as follows. For example, if we want to study how a change in dividend yields dy responses to political risk z {crisis, sevidx}, after controlling for other factors (X), we can employ the following linear regression model: dy = αz + X β + ɛ. (2) Throughout the paper, the firm characteristics that we use as control variables include Tobin s q (q), asset growth (dta), firm size (mv), the life cycle (rete), return on assets (roa), cash holdings (cash), closely held ownership (ch), and stock return volatility (std). The standard errors are robust to heteroskedasticity and firm-level clustering unless we state otherwise. 4 The Effect of Political Crises on Dividend Policy In this section, we examine a few of unsettled empirical questions on corporate dividend policy. First, do firms become more likely to terminate dividend payouts and less likely to 17

19 initiate dividend payouts during periods of high political uncertainty? As a robustness check for our first set of results, we next examine whether the dividend-payers reduce their levels of dividend payments during these periods. We also perform several other robustness checks using sub-samples and different dividend payout measures. In particular, we attempt to address whether our results are driven by the time-varying composition of the sampled firms. We also analyze how the macro-level stability of political system reshapes the relationship between political uncertainty and dividend policy. 4.1 Dividend Termination, Initiation, and Political Crises Table 3 shows our Logit regression results for dividend termination (Models 1 and 2) and initiation (Models 3 and 4) decisions. The explanatory variables of interests are crisis and sevidx in the first two rows of the table. We first examine the impact of political crises on dividend termination decisions. Models 1 and 2 show that crisis has a positive coefficient of and that sevidx has a positive coefficient of 0.010; both are highly significant at the 1% level. These results indicate that there is a positive relationship between termination decisions and political uncertainty. In particular, our results imply that, on average, the ratio of a past dividend payer s probability of terminating these payouts to the probability of its continuing to pay dividends increases by approximately 7.39% (= [exp( ) 1] 100%) in response to a one-standard-deviation (= 2.229) increase in crisis. Similarly, the ratio of the odds of 18

20 terminating the dividend payouts to the odds of continuing to pay the dividends increases by 7.92% (= [exp( ) 1] 100%) in response to a one-standard-deviation (= 7.626) increase in sevidx. Therefore, our results suggest that past dividend payers tend to be more likely to terminate their dividend payments during periods of high political risk (measured by both the quantity and the severity of political crises). Next, we document the existence of a negative relationship between dividend initiation decisions and political risk (see Models 3 and 4 of Table 3). crisis has a negative coefficient of , and sevdix has a negative coefficient of ; both are highly significant at the 5% level. That is, given a one-standard-deviation increase in crisis, for the average past dividend non-payer, the ratio of the odds of initiating a new dividend payout to the odds of continuing not to pay the dividend decreases by 4.15%. For a one-standard-deviation increase in sevidx, for the same non-payer, the ratio of the odds of initiating a new dividend payout to the odds of continuing not to pay dividend decreases by 4.47%. Thus, past dividend non-payers tend to become less likely to initiate dividend payments during periods with a high degree of political uncertainty. In addition to examining the crisis variables, we control for the firm-level characteristics that are commonly used in the literature (see, e.g., Fama and French (2001), Baker and Wurgler (2004), and Chay and Suh (2009)). In particular, we document that large firms (measured by mv) tend to be more likely to initiate dividends and to be less likely to terminate dividends. Like large firms, mature firms (measured by rete) are on average more 19

21 likely to pay dividends and are on average less likely to cut dividends. Profitability (measured by roa) is positively associated with dividend initiation and is negatively associated with dividend termination. Cash holdings (measured by cash) tend to encourage dividend payouts and tend to discourage terminating such payouts. Last, firm-level uncertainty (measured by std) is positively correlated with dividend termination and is negatively associated with dividend initiation. These findings are generally consistent with those documented in the literature (see, e.g., Baker and Wurgler (2004) and Chay and Suh (2009)). The positive (negative) impact of political uncertainty on dividend termination (initiation) decisions merits more discussion. First, political uncertainty has significant explanatory power even after firm-level risk (std) is controlled for. Chay and Suh (2009) document that cash flow uncertainty captures the cross-section variation in dividend policy, and our results complement their finding by showing that uncertainty also accounts for the time-series variation in dividend policy. Second, unlike risk measures based on financial market variables, a political crisis is exogenous to the individual firm policy process. That is, a political crisis can affect an individual firm s dividend policy through the uncertainty channel but the reverse is very unlikely to be true. Therefore, to the best of our knowledge, we are the first to establish a causal relationship, rather than a correlation, between risk and corporate dividend policy. [Insert Table 3 about here] 20

22 4.2 Robustness Checks The first robustness check that we perform in this subsection allows us to study the impact of political uncertainty on the alternative dividend payout measures. These dependent variables are (in order) dividend termination variables based on shorter past periods (dt 2 and dt 1 ), dividend initiation variables based on shorter past periods (di 2 and di 1 ), and the change in dividend payout ratios ( dy, dvs, and dtps). Detailed definitions of these variables are provided in Appendix A. The regression results are presented in Table 4. The first and second rows of the table show that crisis variables (crisis and sevidx) have positive coefficients when dividend termination decisions are used and they have negative coefficients when dividend initiation decisions and continuous payout ratios are used. In particular, the coefficients of the political crisis variables in the payout ratio regressions (Models 9 to 14) are all significant at the 1% level. Consistent with our findings on dividend termination and initiation decisions, we find that firms tend to reduce their dividend payments in response to a high degree of political uncertainty. Although the results are weaker in the regressions based on alternative dividend termination/initiation decisions (Models 1 to 8), they are qualitatively similar to our primary results in Table 3. [Insert Table 4 about here] Second, we study the impact of the individual crisis index (greatp, vbreak, vcrisis, war, gthreat, 21

23 and protracted) on corporate dividend policy. Dividend termination regressions are provided in Models 1 to 6 of Table 5, and dividend initiation regressions are provided in Models 7 to 12 of the same table. In accordance with our primary results in Table 3, past dividend payers become more likely to terminate their dividend payments when an individual crisis index is high. In particular, we find that four out of six coefficient estimators of the individual crisis variables are significantly positive at the 1% level, although the other two are insignificant. Moving to Models 7 to 14 of Table 5, we find similar support when dividend initiation decisions are analyzed. All six crisis variables have negative coefficient estimates, and three out of the six coefficient estimates are significant at the 5% level, indicating that past non-payers tend to be less likely to initiate new dividend payments in response to more severe political crises. Overall, our primary results in Table 3 are robust to the individual components of our main explanatory variable (sevidx) rather than to sevidx itself. [Insert Table 5 about here] Third, our sample covers firms from both developed markets and emerging markets. If there is a time-varying composition of the sampled firms from developed markets and emerging markets and if this time-varying composition is highly correlated with our political uncertainty measures, our results regarding the relationship between the change in dividend payout and political uncertainty may be driven by the degree of economic development. To address this concern, in Table 6, we report estimation results for Models 1 to 4 that are based on the respective developed and emerging sub-samples (dt in Panel A and di in Panel 22

24 B). Overall, the results are qualitatively unchanged, although the magnitude of the crisis coefficient estimates tend to be larger for the firms in emerging markets. Interestingly, this finding also reveals that the negative impact of political turmoil on the individual firms dividend policy tends to be weaker in developed markets. 14 [Insert Table 6 about here] Fourth, a decline in the dividend payouts of the listed firms may be attributable to the time-varying characteristics of the listed firms. For example, the population of publicly traded firms tilts increasingly toward small firms with low profitability and high investment opportunities - characteristics that are typical of firms that have never paid dividends (see Fama and French (2001) for empirical evidence and DeAngelo, DeAngelo, and Stulz (2010) for a theoretical justification). To partially eliminate the effects of the time-varying properties of the sampled firms, we limit our sample to those firms that are continuously listed on the market over the second half of the sample period (during the years ), i.e., those firms that have valid observations for that period. This choice of a limited sample period is based on our observation that a large portion of the sample firms are not continuously listed in the early years. We estimate Models 5 and 6 of Table 6 (dt in Panel A and di in Panel B) for the period. The results remain qualitatively unchanged. Last, the extant dividend literature reports a general time trend of decline in dividends for U.S. firms in recent decades (see, e.g., Fama and French (2001)). The solid lines in Figure 14 In Section 4.3, we provide a more detailed discussion on the impact of the political system on the sensitivity of dividends to political crises. 23

25 1 confirm the existence of a similar time-series pattern based on our international sample. This time trend may drive the negative relationship between dividend payouts and political crises if the number of crises and the severity of the crises increase over the same period. However, as shown in Figure 1, there is actually a declining pattern of crises over the sample period (the dotted lines). Therefore, the documented negative relationship between dividend payouts and political uncertainty is unlikely to be driven by some omitted latent variables. Even more compellingly, when we add a time-trend variable (trend = year 1989) to our primary tests and report the estimation results obtained using this variable in Models 7 and 8 of Table 6 (dt in Panel A and di in Panel B), we find that our results remain robust when controlling for the time-trend variable. 4.3 Political System and Political Crisis For the sake of simplicity, from now on we will use sevidx as the only measure of political uncertainty because it captures not only the number of crises but also the severity of each crisis. Our prior discussions in Section 2 hypothesized that dividend policy in countries with more stable political environments are less sensitive to adverse political shocks. We test this hypothesis by interacting sevdix with different country-level political stability measures, respectively. We use three common political stability variables: 1. checks: A checks and balances variable provided by the Database of Political Institutions. This variable measures the effectiveness of checks and balances in each political 24

26 system on an annual basis. More specifically, we determine the number of decision makers whose agreement is necessary for the approval of policy changes (see, e.g., Keefer (2010) and Julio and Yook (2012)). 2. stability: Kaufmann s political stability measure captures the perceived likelihood that a government will be destabilized or overthrown by unconstitutional or violent means, including politically motivated violence and terrorism (see Kaufmann, Kraay, and Mastruzzi (2010)). 3. polista: A composite political stability variable that combines several indicators that measure the perceived likelihood that the government in power will be destabilized or overthrown by possibly unconstitutional and/or violent means, including domestic violence or terrorism (see A greater value for the three political stability variables indicates a more stable political system. We present our estimation results in Table 7. The first row of the table shows the coefficients of the interactions between sevidx and the political stability variables. Our second hypothesis states that corporate dividend decision are less sensitive to political uncertainty in more stable political environments. Therefore, we expect to find negative coefficient estimates of the interactions (in contrast to the positive coefficient estimates for seridx itself) for dividend termination, and we expect to find positive coefficient estimates for the interactions 25

27 (in contrast to the negative coefficient estimates for seridx itself) for dividend initiation. We find consistent results in the first row. In particular, those firms relative to their counterparts in countries with less stable political systems are less likely to terminate dividend payouts and are more likely to initiate dividend payouts during periods of high political uncertainty. [Insert Table 7 about here] 5 Strategic dividend initiation Thus far, based on an international sample, our results show that firms tend to be more likely to terminate dividend payouts during periods of high political uncertainty. This evidence indicates that there is a precautionary motive for managers deliberate dividend policy. However, an examination of our sample reveals that when we sort our sample into five groups based on sevidx, more than 5% firm-years include instances in which dividend payouts were initiated in the highest severity index group. This result yields an interesting question: that of why firms initiate dividends even when there is high uncertainty in the market. One possible explanation is that some firms may strategically manage their dividend policy in response to changes in the macro-environment. Although the number of firms initiating dividends during periods of high political uncertainty is relatively small, analyzing the behavior of these firms helps us to better understand firms strategic dividend behavior. In this section, we analyze why some firms respond to political uncertainty differently 26

28 from others, which firm characteristics drive this difference, and how we can use existing dividend theories to explain the difference. The exogenous nature of political crises provides us with a clean test of several existing explanations for corporate payout decisions. We focus on two likely channels through which firms may be encouraged to initiate dividends even in the presence of a high degree of uncertainty. Given that our focus is dividend initiation decisions, the subsequent analysis will be restricted to the dividend initiation measure (di). 5.1 The signaling hypothesis of dividends We start by testing the signaling hypothesis. First, we test for the consequence of signaling. That is, firms that make dividend payments under high political uncertainty tend to have better future performance. In Table 8, we regress the future net sales growth (growth sales,+1 and growth sales,+2 ), respectively, on the interaction between the crisis measure (sevidx) and the dividend initiation variable(di), controlling for the current year net sales growth (growth sales ), the crisis measure, and the dividend initiation variable. In addition, we examine alternative performance measures including the change in net income (growth earnings,+1 and growth earnings,+2 ) and the change in the return on total assets (growth roa,+1 and growth roa,+2 ). We fail to find any significantly positive coefficients for the interaction terms (the first row in the table) for any of the examined future performance measures. Although there is one case (Model 2) in which we document a significantly negative coefficient, our results indicate deteriorating rather than improved future firm performance 27

29 following crises. [Insert Table 8 about here] Next, we examine the incentives for signaling. The firms with a higher degree of information asymmetry should a have greater incentive to initiate dividend payouts to reveal their information. To test this theory, we add the interaction term between our main crisis variable sevidx and a firm-level information asymmetry measure. We use four different information asymmetry variables (firm age (age), accounting accruals (acc), the probability of informed trading (pin), and analyst coverage (ana)) separately in each model. It is evident that information asymmetry is positively correlated with pin and acc and that is negatively associated with age and ana. The definitions of these information asymmetry variables are summarized in Appendix A. If the signaling hypothesis is correct, the interactions between sevidx and age and ana should be negatively significant, whereas the interactions between sevidx and acc and pin should be positively significant. The first row of Table 9 summarizes the regression coefficients of the interaction terms. The interaction terms of all four models are insignificantly different from zero. In summary, the results in Table 8 and Table 9 are very difficult to reconcile with a theory of rational signaling in which corporate managers decide to initiate dividend payouts during periods of high political uncertainty. [Insert Table 9 about here] 28

30 5.2 The catering hypothesis of dividends Next we test the catering theory of dividends. To motivate firms to initiate dividends during high risk periods, financial markets must compensate them with a premium, i.e., they must value dividend payers higher than their otherwise identical counterparts. In Table 10, we provide evidence that the financial markets reward dividend payers more under higher political uncertainty. Following Baker and Wurgler (2004), we construct two aggregate timeseries of dividend premiums, i.e. we examine the difference between the average market-tobook ratios of the dividend payers and the non-payers. We denote the premium by dpremew (dpremvw) if the average market-to-book ratios are equally (value) weighted. The first row of Table 10 shows that the country-level dividend premiums are positively correlated with sevidx. [Insert Table 10 about here] Having established that financial markets appear to more intensely demand and better reward dividend payouts under high political uncertainty, we then study whether those under-valued stocks (using low q as a proxy) tend to be more likely to be associated with new dividend payouts. To test this potential relationship, we show in Model 3 of Table 11 that q decreases with sevidx, i.e., that firms are, on average, under-valued during periods of high political uncertainty. In addition, we expect to see a negative coefficient for the interaction between sevidx and q in the dividend initiation regression if under-valued firms 29

31 (i.e., low q) have more incentive to initiate dividend payments. Unsurprisingly, we find a significantly negative coefficient for the interaction in Model 1 of Table 11. This finding is consistent with the catering hypothesis of dividends, in which firms cater to investors by making dividend payments when those investors value dividend payers at a premium, i.e., when political uncertainty is high. Lastly, we study whether liquidity acts as a channel underlying the market demand for dividend payers. Following Amihud (2002), we use the Amihud (il)liquidity measure, amihud. In Model 4 of Table 11, we find that firm-level liquidity tends to decrease dramatically (or that illiquidity tends to increase substantially) under high political crisis. This result is consistent with what has been documented in financial crises. In Model 2 of Table 11, we add an interaction term for sevidx and amihud to our primary regression specification. The interaction term has a significantly positive coefficient. That is, although the average firm tends to be less likely to initiate dividend payouts during high political crisis, an illiquid firm may find that it is optimal to initiate dividend payments to cater to market demand for dividend payers during periods of high political uncertainty. This finding suggests that dividends and liquidity may be substitutes for one another (see Banerjee, Gatchev, and Spindt (2007)). [Insert Table 11 about here] 30

32 6 Conclusion Using a comprehensive sample of 112,152 firm-year observations from 35 countries over two decades, this paper explores firms intertemporal dividend decisions by examining the impact of political uncertainty. This paper has two distinguishing features that differentiate it from existing studies. First and most importantly, our measure of uncertainty is based on exogenous shocks triggered by international political crises. We are therefore able to establish the causal relationship rather than the correlation between uncertainty and dividend policy. Second, examining time-varying political risk allows us to study the time-series variation rather than the cross-sectional variation in dividend decisions. We have obtained three sets of novel findings. First, we show that past payers tend to be more likely to terminate their dividend payments and that past non-payers tend to be less likely to initiate dividend payments during periods with more (or more severe) political crises. These results are robust to the use of a variety of sub-samples, alternative political crisis measures, and alternative dividend decision variables. Second, we find that dividend decisions become less sensitive to political uncertainty in countries with more stable political systems. The result highlights the influential role of the macro-environment in alleviating the adverse effect of political uncertainty on individual firms. Third, we show that the dividend decisions of under-valued firms and firms with illiquid shares are less sensitive to political uncertainty because it is optimal for them to cater to 31

33 the market demand for dividends under high political uncertainty. Overall, our study provides new evidence that firms make deliberate dividend payout decisions. In particular, our results show that the precautionary motives of managers tend to be one of the important factors that explain the primary pattern of observed dividend payouts. In addition, the country-level heterogeneity in the sensitivity of dividends to political uncertainty indicates that political stability appears to be able to partially internalize the negative externalities triggered by political events. Finally, the firm-level heterogeneity in the sensitivity of dividends to political risk appears to be consistent with the catering hypothesis of dividends and difficult to reconcile with the signaling hypothesis of dividends. 32

34 Table 1: Summary Statistics. This table presents the summary statistics of firm-year observations in our sample across 35 countries over the period from year 1990 to dt is the dividend termination dummy, di is the dividend initiation dummy, crisis is the number of crises, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. N.obs. Mean Std. dev. P10 Q1 Median Q3 P90 dt 68, di 43, crisis 112, sevidx 112, q 112, dta 112, mv 112, rete 112, roa 112, cash 112, ch 112, std 112,

35 Table 2: Correlation matrix. This table presents the correlation coefficients of our key variables of our sample across 35 countries over the period from year 1990 to Pearson s correlations are given in the lower triangular matrix. Spearman s correlations are given in the upper triangular matrix. dt is the dividend termination dummy, di is the dividend initiation dummy, crisis is the number of crises, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is the return volatility. Detailed variable definitions are given in Appendix A. dt di crisis sevdix q dta mv rete roa cash ch std dt di crisis sevdix q dta mv rete roa cash ch std

36 Table 3: Logit regressions of dividend termination/initiation decision. This table presents Logit regression results of the dividend termination decision and initiation decision for our sample across 35 countries over the period from year 1990 to dt is the dividend termination dummy, di is the dividend initiation dummy, crisis is the number of crises, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. dt di Variable Model 1 Model 2 Model 3 Model 4 crisis 0.032*** ** (3.72) (-1.99) sevdix 0.010*** ** (3.91) (-2.08) q ** ** *** *** (-2.12) (-2.08) (-4.19) (-4.19) dta *** *** 0.114** 0.112* (-3.91) (-3.87) (1.98) (1.94) mv *** *** 0.182*** 0.182*** (-20.03) (-20.00) (12.43) (12.40) rete *** *** 0.064*** 0.064*** (-4.82) (-4.84) (4.43) (4.46) roa *** *** 2.636*** 2.634*** (-7.26) (-7.27) (6.01) (6.01) cash *** *** 0.849*** 0.854*** (-3.73) (-3.74) (2.60) (2.61) ch *** 0.625*** (1.41) (1.43) (6.14) (6.17) std 2.764*** 2.743*** *** *** (23.40) (23.15) (-11.13) (-11.06) Country FEs Yes Yes Yes Yes Industry FEs Yes Yes Yes Yes Clustering Firm Firm Firm Firm NObs 68,816 68,816 43,335 43,335 R % 19.9% 12.5% 12.5% 35

37 Table 4: Regressions of alternative dependent variables. This table presents regressions results of alternative dependent variables for our sample across 35 countries over the period from year 1990 to dt1 (dt2) is the dividend termination dummy based on last one (two) year(s), di1 (di2)is the dividend initiation dummy based on last one (two) year(s), dy is the change of dividend yield, dvs is the change of dividend-to-sales ratio, tps is the change of total payout-to-sales ratio, crisis is the number of crises, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. dt2 dt1 di2 di1 dy dvs tps Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 Model 13 Model 14 crisis 0.020*** 0.011* * *** *** *** (2.79) (1.76) (-1.35) (-1.83) (-15.91) (-5.46) (-4.91) sevdix 0.006*** *** *** *** (2.71) (1.25) (-0.92) (-0.46) (-18.51) (-8.41) (-7.79) q *** *** *** *** *** *** 0.032*** 0.032*** 0.049*** 0.049*** (-1.18) (-1.15) (0.84) (0.86) (-4.81) (-4.83) (-6.45) (-6.50) (-15.34) (-15.56) (13.03) (13.05) (9.16) (9.17) dta *** *** *** *** ** 0.090** 0.061*** 0.058*** *** *** *** *** (-4.36) (-4.34) (-4.19) (-4.20) (1.17) (1.16) (2.30) (2.29) (8.93) (8.40) (-7.63) (-7.83) (-15.14) (-15.34) mv *** *** *** *** 0.178*** 0.178*** 0.187*** 0.187*** *** *** 0.012*** 0.012*** 0.031*** 0.031*** (-23.66) (-23.64) (-27.34) (-27.33) (13.94) (13.93) (16.83) (16.83) (-4.00) (-4.16) (8.50) (8.42) (11.41) (11.35) rete *** *** *** *** 0.075*** 0.075*** 0.101*** 0.101*** *** ** *** *** *** *** (-5.65) (-5.67) (-6.32) (-6.32) (5.84) (5.85) (8.75) (8.75) (-2.93) (-2.29) (-3.77) (-3.51) (-3.89) (-3.65) roa *** *** *** *** 2.243*** 2.241*** 2.108*** 2.104*** 0.193*** 0.197*** 0.216*** 0.219*** 0.458*** 0.465*** (-7.55) (-7.56) (-6.13) (-6.12) (4.94) (4.94) (5.52) (5.51) (8.73) (8.89) (6.31) (6.40) (6.48) (6.57) cash *** *** *** *** 1.335*** 1.338*** 1.487*** 1.490*** 0.347*** 0.351*** 0.223*** 0.224*** 0.487*** 0.489*** (-4.21) (-4.21) (-5.95) (-5.95) (3.95) (3.96) (5.06) (5.08) (14.43) (14.62) (6.67) (6.71) (7.18) (7.21) ch *** 0.467*** 0.466*** 0.465*** *** *** (1.34) (1.35) (1.20) (1.20) (5.34) (5.34) (6.27) (6.26) (-3.20) (-3.12) (0.57) (0.60) (-1.08) (-1.05) std 2.586*** 2.575*** 2.351*** 2.347*** *** *** *** *** 0.035*** 0.046*** * (25.86) (25.66) (28.54) (28.41) (-12.45) (-12.40) (-15.61) (-15.56) (3.50) (4.63) (-1.79) (-1.36) (-1.45) (-1.00) Industry FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Clustering Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm NObs 80,637 80,637 93,677 93,677 54,600 54,600 68,573 68, , , , , , ,210 R2 20.6% 20.6% 21.0% 21.0% 13.5% 13.5% 15.8% 15.8% 0.9% 1.0% 0.8% 0.8% 0.7% 0.7% 36

38 Table 5: Logit regressions of alternative independent variables. This table presents Logit regressions of alternative independent variables for our sample across 35 countries over the period from year 1990 to greatp is the number of crises involving at least one great power or superpower on both sides of the conflict vbrak is the number of crises beginning with a violent act, vcrisis is the number of crises with either serious clashes or full-scale wars war is the number of crises with full-scale wars gthreat is the number of crises involving a territorial threat, a threat of grave damage, or a threat to existence, protracted is the number of crises in protracted conflicts, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. dt di Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10 Model 11 Model 12 greatp 0.089*** ** (4.96) (-2.18) vbreak 0.066*** ** (3.33) (-2.00) vcrisis (-1.07) (-1.09) war (0.02) (-0.75) gthreat 0.041*** (3.01) (-1.01) protracted 0.068*** ** (5.33) (-2.12) q ** ** ** ** ** ** *** *** *** *** *** *** (-2.16) (-2.04) (-2.05) (-2.05) (-2.05) (-2.15) (-4.23) (-4.22) (-4.23) (-4.27) (-4.21) (-4.19) dta *** *** *** *** *** *** 0.112* 0.110* 0.114** 0.112* 0.113* 0.111* (-3.85) (-3.96) (-4.11) (-4.07) (-3.96) (-3.83) (1.96) (1.92) (1.99) (1.95) (1.96) (1.94) mv *** *** *** *** *** *** 0.183*** 0.182*** 0.182*** 0.182*** 0.182*** 0.182*** (-20.11) (-20.05) (-20.12) (-20.12) (-19.98) (-19.98) (12.47) (12.39) (12.44) (12.43) (12.40) (12.41) rete *** *** *** *** *** *** 0.063*** 0.064*** 0.064*** 0.064*** 0.064*** 0.064*** (-4.73) (-4.84) (-4.82) (-4.84) (-4.87) (-4.79) (4.41) (4.45) (4.46) (4.45) (4.45) (4.44) roa *** *** *** *** *** *** 2.612*** 2.626*** 2.645*** 2.628*** 2.639*** 2.626*** (-7.13) (-7.24) (-7.20) (-7.26) (-7.31) (-7.21) (5.96) (5.99) (6.02) (6.00) (6.02) (5.99) cash *** *** *** *** *** *** 0.872*** 0.860*** 0.843** 0.858*** 0.851*** 0.862*** (-3.84) (-3.81) (-3.78) (-3.74) (-3.69) (-3.79) (2.67) (2.63) (2.57) (2.63) (2.60) (2.64) ch *** 0.624*** 0.621*** 0.621*** 0.623*** 0.624*** (1.40) (1.42) (1.39) (1.40) (1.43) (1.42) (6.15) (6.17) (6.14) (6.14) (6.16) (6.17) std 2.781*** 2.741*** 2.786*** 2.771*** 2.748*** 2.783*** *** *** *** *** *** *** (23.47) (23.12) (23.11) (22.52) (23.13) (23.51) (-11.09) (-11.03) (-11.08) (-10.85) (-11.08) (-11.13) Industry FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Clustering Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm Firm NObs 68,816 68,816 68,816 68,816 68,816 68,816 43,335 43,335 43,335 43,335 43,335 43,335 R2 20.0% 19.9% 19.9% 19.9% 19.9% 20.0% 12.5% 12.5% 12.5% 12.5% 12.5% 12.5% 37

39 Table 6: Logit regressions of sub-samples and with the trend dummy. This table presents Logit regressions for a variety of sub-samples and with the trend dummy. DEV is the developed countries sub-sample, EM G is the emerging markets sub-sample, Constant is a sub-sample of a balanced panel, i.e., every included firm has been continuously listed from year 2000 to 2008, trend is a year dummy equal to year dt is the dividend termination dummy (Panel A), di is the dividend initiation dummy (Panel B), q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. Panel A: dt DEV EMG Constant trend Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 crisis 0.024** 0.088*** 0.051** 0.023** (2.53) (3.82) (2.43) (2.46) sevdix 0.008*** 0.025*** 0.015** 0.007** (2.87) (3.53) (2.46) (2.27) trend *** ** (-2.67) (-2.20) q ** ** ** ** ** ** (-2.10) (-2.07) (-0.43) (-0.36) (-2.10) (-2.07) (-2.12) (-2.09) dta *** *** *** *** (-3.54) (-3.49) (-1.25) (-1.34) (-1.30) (-1.27) (-3.80) (-3.80) mv *** *** *** *** *** *** *** *** (-18.57) (-18.53) (-6.94) (-6.92) (-10.37) (-10.33) (-19.88) (-19.89) rete *** *** ** ** *** *** *** *** (-4.42) (-4.44) (-2.36) (-2.35) (-3.93) (-3.94) (-4.87) (-4.88) roa *** *** * *** *** *** *** (-6.87) (-6.89) (-1.61) (-1.65) (-3.26) (-3.26) (-7.33) (-7.33) cash *** *** *** *** *** *** (-2.72) (-2.73) (-4.63) (-4.60) (0.54) (0.52) (-3.60) (-3.62) ch (1.28) (1.29) (1.23) (1.25) (-1.53) (-1.51) (1.47) (1.47) std 2.945*** 2.932*** 2.000*** 1.927*** 3.367*** 3.323*** 2.759*** 2.746*** (20.85) (20.73) (9.45) (9.08) (12.80) (12.77) (23.31) (23.15) Industry FEs Yes Yes Yes Yes Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Yes Yes Yes Yes Clustering Firm Firm Firm Firm Firm Firm Firm Firm NObs 62,064 62,064 6,752 6,752 17,383 17,383 68,816 68,816 R2 19.3% 19.3% 22.1% 22.0% 22.3% 22.3% 19.9% 19.9% 38

40 Panel B: di DEV EMG Constant trend Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 crisis *** *** ** (-0.91) (-3.24) (-2.90) (-2.47) sevdix *** *** *** (-0.98) (-3.52) (-3.75) (-3.26) trend ** *** (-2.35) (-3.07) q *** *** ** ** *** *** (-3.35) (-3.35) (-2.29) (-2.42) (-0.51) (-0.48) (-4.27) (-4.28) dta 0.121** 0.120** ** ** 0.118** 0.117** (2.04) (2.02) (-0.49) (-0.58) (-2.16) (-2.17) (2.07) (2.05) mv 0.158*** 0.158*** 0.349*** 0.351*** 0.200*** 0.199*** 0.184*** 0.184*** (10.10) (10.08) (8.37) (8.39) (5.22) (5.17) (12.58) (12.58) rete 0.054*** 0.054*** 0.156*** 0.157*** 0.110** 0.109** 0.063*** 0.063*** (3.54) (3.56) (3.69) (3.71) (2.28) (2.28) (4.35) (4.37) roa 2.712*** 2.710*** *** 3.460*** 2.620*** 2.613*** (5.97) (5.97) (1.54) (1.60) (2.88) (2.83) (5.96) (5.95) cash *** 4.608*** *** 0.851*** (1.13) (1.14) (3.81) (3.75) (1.26) (1.28) (2.59) (2.60) ch 0.558*** 0.560*** 0.515* 0.522* 0.605** 0.610** 0.611*** 0.613*** (5.17) (5.18) (1.72) (1.74) (2.20) (2.21) (6.03) (6.06) std *** *** *** *** *** *** *** *** (-10.19) (-10.17) (-4.61) (-4.47) (-7.54) (-7.30) (-11.16) (-11.06) Industry FEs Yes Yes Yes Yes Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Yes Yes Yes Yes Clustering Firm Firm Firm Firm Firm Firm Firm Firm NObs 39,400 39,400 3,935 3,935 7,935 7,935 43,335 43,335 R2 11.8% 11.8% 16.4% 16.5% 17.0% 17.2% 12.5% 12.5% 39

41 Table 7: Logit regressions of dividend termination/initiation decision - political stability. This table presents Logit regressions with an interaction between sevidx and a political stability variable for our sample across 35 countries over the period from year 1990 to psv is a generic political stability variable that could be any one of checks, stability, and polista, checks is the checks and balances measure, stability is a political stability measure from Kaufmann, Kraay, and Mastruzzi (2010), polista is a political stability measure from dt is the dividend termination dummy, di is the dividend initiation dummy, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. dt di checks stability polista checks stability polista Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 psv sevdix ** *** *** *** 0.019*** (-2.41) (-3.55) (-2.95) (0.88) (3.69) (3.24) psv 0.221*** *** *** (5.09) (-0.63) (-0.63) (-1.56) (-4.28) (-4.28) sevdix 0.027*** 0.017*** 0.024*** *** *** (3.01) (3.00) (4.73) (-1.20) (-4.50) (-4.23) q *** *** ** *** ** *** (-2.68) (-2.69) (-2.26) (-4.02) (-2.37) (-3.30) dta *** *** *** 0.117* 0.139** (-4.07) (-2.83) (-3.16) (1.94) (2.00) (1.15) mv *** *** *** 0.178*** 0.189*** 0.186*** (-19.48) (-16.26) (-17.88) (11.90) (10.90) (11.77) rete *** *** *** 0.061*** 0.057*** 0.066*** (-4.78) (-3.79) (-3.67) (4.14) (3.42) (4.36) roa *** *** *** 2.598*** 2.201*** 2.482*** (-7.44) (-6.35) (-6.73) (5.83) (4.37) (5.33) cash *** * * 0.746** ** (-3.40) (-1.66) (-1.91) (2.26) (1.20) (2.15) ch *** 0.588*** 0.656*** (1.16) (-0.34) (0.15) (6.04) (4.88) (6.06) std 2.798*** 2.594*** 2.613*** *** *** *** (22.57) (17.65) (19.12) (-10.49) (-9.13) (-9.58) Industry FEs Yes Yes Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Yes Yes Clustering Firm Firm Firm Firm Firm Firm NObs 67,322 44,152 53,122 42,161 31,076 38,923 R % 20.1% 20.5% 12.3% 12.0% 12.8% 40

42 Table 8: Panel data regressions of firm performance. This table presents panel data regressions of firm performance for our sample across 35 countries over the period from year 1990 to Dependent variables are growth rates of sales, earnings, and return on assets, respectively. di is the dividend initiation dummy, sevidx is the aggregate crisis severity index, Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. growthsales,+1 growthsales,+2 growthearnings,+1 growthearnings,+2 growthroa,+1 growthroa,+2 Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 di sevdix *** (-0.87) (-2.91) (0.62) (1.42) (-0.38) (-1.27) di ** ** *** *** (1.36) (2.45) (-2.52) (-1.18) (4.02) (-4.37) sevidx 0.005*** (14.34) (-0.44) (-0.37) (-1.41) (-1.27) (0.35) growthsales 0.099*** 0.025*** (10.11) (2.61) growthearnings (-0.61) (-0.98) growthroa *** *** (-20.02) (-6.28) Industry FEs Yes Yes Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Yes Yes Clustering Firm Firm Firm Firm Firm Firm NObs 37,647 34,429 40,471 37,258 39,704 36,169 R2 3.5% 1.9% 0.0% 0.0% 5.3% 0.4% 41

43 Table 9: Logit regressions of dividend initiation - information asymmetry. This table presents Logit regressions of dividend initiation with an interaction between sevidx and an information asymmetry variable for our sample across 35 countries over the period from year 1990 to inf oasy is a generic information asymmetry variable that could be any one of age, acc, pin, and ana, age is firm age, acc is accounting accruals, pin is the probability of informed trading, ana is the analyst coverage, di is the dividend initiation dummy, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-to-total equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. age acc pin ana Model 1 Model 2 Model 3 Model 4 inf asy sevdix (-0.40) (-1.03) (-0.72) (0.05) inf asy ** (1.32) (2.07) (0.17) (-1.53) sevdix ** * (-0.63) (-2.14) (-0.15) (-1.68) q *** *** *** ** (-4.02) (-3.99) (-3.17) (-2.24) dta 0.119** 0.103* (2.07) (1.73) (-0.40) (0.78) mv 0.179*** 0.184*** 0.238*** 0.165*** (12.16) (12.33) (10.06) (6.40) rete 0.063*** 0.067*** 0.110*** 0.062*** (4.43) (4.52) (5.11) (3.00) roa 2.618*** 2.487*** 4.315*** 2.952*** (5.96) (5.43) (5.90) (5.17) cash 0.862*** 0.875** 1.391** (2.63) (2.54) (2.51) (0.88) ch 0.634*** 0.625*** 0.614*** 0.708*** (6.23) (6.11) (3.97) (5.37) std *** *** *** *** (-10.98) (-10.78) (-11.52) (-10.20) Industry FEs Yes Yes Yes Yes Country FEs Yes Yes Yes Yes Clustering Firm Firm Firm Firm NObs 43,335 42,231 15,935 25,933 R % 12.6% 14.8% 15.0% 42

44 Table 10: Country-level dividend premium regressions. This table presents countrylevel dividend premium regressions for our sample across 35 countries over the period from year 1990 to sevidx is the aggregate crisis severity index, dpremew is the equally weighted country-level dividend premium, dpremvw is the value-weighted country-level dividend premium. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. dpremew dpremvw Variable Model 1 Model 2 sevidx 0.004*** (4.46) (1.31) dpremew *** (6.03) dpremvw *** (8.54) Country FEs Yes Yes Clustering Country Country NObs R % 39.4% 43

45 Table 11: Logit regressions of dividend initiation decision - Tobin s q and liquidity. This table presents Logit regressions of dividend initiation decision with an interaction between sevidx and q or an interaction between sevidx and liquidity for our sample across 35 countries over the period from year 1990 to di is the dividend initiation dummy, amihud is the amihud (il)liquidity measure, sevidx is the aggregate crisis severity index, q is Tobin s q, dta is the growth rate of assets, mv is firm size, rete is retained earnings-tototal equity ratio, roa is return on assets, cash is cash holding, ch is the fraction of closely held shares, and std is stock return volatility. Detailed variable definitions are given in Appendix A. T-statistics are given in parentheses. ***, ** or * next to coefficients indicate that coefficients are significantly different from zero at the 1%, 5%, or 10% levels, respectively. di q amihud Variable Model 1 Model 2 Variable Model 3 Model 4 q sevdix * sevidx ** 0.025*** (-1.91) (-1.98) (48.92) amihud sevdix 0.003*** mv 0.273*** *** (3.41) (17.19) ( ) amihud bm *** 0.018* (-1.60) (-29.49) (1.69) sevdix ret 0.448*** 0.135*** (0.19) (-1.52) (21.84) (16.20) q *** std 0.857*** *** (-0.80) (-3.48) (15.90) (-12.10) dta 0.113** adr * *** (1.97) (1.48) (-1.85) (-3.89) mv 0.182*** 0.183*** ana *** *** (12.42) (6.68) (-7.09) (-33.12) rete 0.064*** 0.062*** (4.47) (4.22) roa 2.636*** 2.566*** (6.03) (5.67) cash 0.858*** 0.768** (2.63) (2.29) ch 0.623*** 0.633*** (6.16) (5.46) std *** *** (-11.08) (-10.92) Industry FEs Yes Yes Industry FEs Yes Yes Country FEs Yes Yes Country FEs Yes Yes Clustering Firm Firm Clustering Firm Firm NObs 43,335 41,364 NObs 25,458 25,458 R % 12.4% R2 34.1% 34.1% 44

46 Figure 1: For our sample period from year 1990 to 2008, the top left graph plots the time-series of the number of crises (crisis) and the average dividend termination (dt), top right graph plots the time-series of the severity index (sevidx) and the average dividend termination (dt), bottom left graph plots the time-series of the number of crises (crisis) and the average dividend initiation (di), and bottom right graph plots the time-series of the severity index (sevidx) and the average dividend initiation (di). Data sources are the ICB database and Datastream. 45