Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia

Transcription

1 IJE : Volume 6 Number 2 December 2012, pp Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia Moncef Guizani * & Ezzeddine Abaoub ** Abstract: The aim of this paper is to investigate the disciplinary role of dividend in the Tunisian context. Based on the agency theory predictions, we consider the effect of two conflicting model of dividend: the outcome and the substitute model. Using a sample of 440 firm-years listed on the Tunisian Stock Exchange over the period , our results highlight the deficiency of the disciplinary role of dividend. Empirical evidence shows that dividend policy is the result of large shareholders preferences. We find a positive relationship between dividend payout ratios and the voting powers of financial institutions and the second largest shareholder. In contrast, the control stakes of family and the largest shareholder are negatively related to the payout ratios. We also find a positive association between the free cash flow, the return on assets, the business sector and dividend to earnings ratio. Finally, we document a negative relationship between the debt ratio and dividend payouts. Taken together, our results are consistent with the outcome model of dividend policy. Keywords: dividend, corporate governance, controlling shareholders, voting power, agency theory. 1. INTRODUCTION Since the pioneering works of Gordon (1959), Lintner (1956), and Miller and Modigliani (1961) there is an ongoing debate on dividend policy, which has been a controversial issue to this day. According to Fisher Black (1976), there is no convincing explanation to why do corporations pay dividends, and why do investors pay attention to dividends. The author concludes to the so-called dividend puzzle. «The harder we look at the dividend picture the more it seems like a puzzle with pieces that just don t feet together». During the three last decades, economists have proposed a number of explanations of the dividend puzzle. Of these, the idea that dividend payments are expected to attenuate agency problems between corporate insiders and outside shareholders. According to Easterbrook (1984), Jensen (1986) and DeAngelo, DeAngelo and Stulz (2006), dividend can reduce agency conflict for many reasons. First, dividend generates external monitoring by forcing managers into the capital market to raise funds. Second, dividends reduce free cash flows that could otherwise be spent by managers on their private benefits rather than maximizing shareholders wealth. * Kairouan University, Department of Management, Higher Institute of Computer Science and Management of Kairouan, ( guizani_m@yahoo.fr) ** Tunis Carthage University, Department of Management, Faculty of Economics Science and Management of Nabeul, ( ezzeddine.abaoub@planet.tn)

2 274 Moncef Guizani & Ezzeddine Abaoub Recent empirical research shows that many publicly traded firms around the world have large shareholders in control (La Porta et al., 1999). Large owners (blockholders) may play a valuable role by reducing the familiar agency problems between shareholders and managers, but recent research has emphasized that large block holdings give rise to a second agency problem between blockholders and minority investors (Shleifer and Vishny, 1997). Large shareholders prefer to generate private benefits of control that are not shared by minority shareholders. They can implement policies that benefit themselves at the expense of outside shareholders. Exploring the field of literature on corporate governance shows that the providers of funds are obsessed with their personal interests. Every shareholder seeks to maximize its profit which creates agency conflicts. Dividend policy as a main component of shareholders wealth would be affected by the interests sought by large shareholders. Inspired by agency theory arguments, La Porta et al. (2000) have established two models of the relation between a firm s corporate governance quality and its payout policy: the outcome model and the substitute model. Under the first model, dividends are the outcome of an effective system of legal protection of shareholders. The authors claim that under an effective system, minority use their legal power to force firms to pay out dividends. As suggested by Adjaoud and Ben-Amar (2010), according to this model, corporate governance quality should be positively related to dividend payouts since better governed-firms offer stronger protection rights to shareholders. Contrary to this view, the second model argues that dividend payout policy is a substitute for governance problems in a firm (La Porta et al. 2000; Gomes, 2000; Maury and Pajuste, 2002). Under this model, corporate governance quality should be negatively related to dividend payouts in a way that better governed-firms are less likely to use dividends as a device to mitigate agency conflicts opposing managers to shareholders (John and KnyAzeva, 2006; Adjaoud and Ben-Amar, 2010). The present paper contributes to this debate as it assesses empirically the disciplinary role of dividend in concentrated firms. In particular, the paper derives and tests a set of hypotheses pertaining to the impact of controlling shareholders on dividend payout ratios and the free cash flow allocation to dividend. It complements the existing literature in two ways. First, we investigate the relationship between the dividend payout ratio and the voting power enjoyed by large shareholders. We analyse the impact of the largest shareholder, the second largest shareholder and different types of shareholders coalition on the payout ratios. This allows us to test a set of hypotheses derived from the two agency models of dividend. Second, we address the problem of control measurement and advocate the use of Banzhaf indices, derived from game theory, as a relevant measure of voting power in the analysis of dividend policy choices. Empirically, regression results of econometric tests on a sample of 44 Tunisian firms observed over the period provide a positive relation between corporate governance quality and dividend payout ratios. Consistent with the outcome model, we find that dividend to earnings ratio and dividend to free cash flow ratio are positively associated with the control stakes of the second largest shareholder, financial institutions and outside individuals. While dividend payouts are negatively related to the voting power of family and the largest shareholder. The remainder of this paper is organized as follows. Section 2 reviews relevant literature on the two agency models of dividend and develop hypotheses. Section 3 describes the data

3 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 275 and explains our research methodology. Section 4 reports our empirical results and section 5 concludes the paper. 2. RELATED LITERATURE AND HYPOTHESIS DEVELOPMENT 2.1. Two Models for the Relation between Agency Costs and Dividend Policy As outlined in La Porta et al. (2000), there are two models describing the relation between agency costs and dividend policy. Following Officer (2006), we adapt each of these models to a setting where firms are in a common legal system (Tunisia) but have attributes that offer minority shareholders more protection ( strong governance ) or weak protection ( weak governance ) The Outcome Model The hypotheses derived from the outcome model are based on the cash flow hypothesis of Jensen (1986). Dividend payouts are the outcome of the firm s governance quality. Managers of firms with weak governance are more likely to retain cash within the firm as it allows them to consume more perquisites and invest in projects that generate more personal benefits at the expense of shareholders wealth. By contrast, in firms with strong governance, managers are less likely to spend the free cash flow in their own interests. They act thus in the interests of shareholders and pursue value-maximizing policies, such as paying out cash to shareholders. Under this model, a positive association between corporate governance quality and dividend payouts is expected. Several empirical studies find evidence supporting the outcome model. Using data of 4000 companies from 33 countries around the world, La Porta et al. (2000) find consistent support for the outcome model agency model of dividends. Firms operating in countries with better protection of minority shareholders pay higher dividends. Maury and Pajuste (2002) find empirical evidence in favour of the outcome agency model of dividends in Finnish listed firms. Specifically, they find that the concentration of voting rights is associated with lower levels of dividends as a proportion of earnings. As noted by the authors, large controlling shareholder or a coalition of large shareholders has the preference and the ability not to pay out profits to all shareholders, but rather to pay themselves only in form of private benefits of control. Michaely and Roberts (2006) confirm the predictions of the outcome agency theory put forth by La Porta et al. (2000). In particular, when there no agency problems (the wholly owned sample), dividends are highly sensitive to changes in investment. In contrast, when agency problems become relevant (private dispersed and public samples), governance mechanisms that protect public firms investors force managers to disgorge cash when growth opportunities are low. According to Kowalewski et al. (2007), there is a strong positive correlation between the Transparency Disclosure Index (the measure of corporate governance) and dividend payouts. The results are in line with the outcome model assuming that when shareholders have greater rights, they can use their power to influence dividend policy. Adjaoud and Ben Amar (2010) investigate the relationship between corporate governance quality and dividend policy in Canada. Their results confirm that effective corporate governance mechanisms attenuate agency conflicts between managers and shareholders and limit managers opportunistic behaviour in payout policy, thus

4 276 Moncef Guizani & Ezzeddine Abaoub supporting the predictions of the outcome model of dividends. Jiraporn et al. (2011) find empirical evidence in agreement with the prediction of outcome hypothesis. Specifically, firms with better governance quality exhibit a stronger propensity to pay dividends and those that pay dividends pay larger dividends The Substitute Model According to La Porta et al. (2000), this view relies crucially on the need for firms to come to the external capital markets for funds, at least occasionally. As suggested by Easterbrook (1984), managers of firms need to establish a reputation with public capital markets for not expropriating wealth from shareholders because the firm will need to tap such markets to raise new external capital in the future. In line with this view, Gomes (2000) argues that controlling shareholders can implicitly commit not to expropriate outside shareholders. More specifically, the author claims that managers can develop a reputation for treating outside shareholders well. One way to establish such a reputation is by paying dividends, which reduces what is left for expropriation (La Porta et al, 2000). The substitute model stipulates that dividend policy is a substitute for other governance mechanisms in reducing agency costs. Therefore, better-governed firms should be less likely to use dividends as a device to mitigate agency conflicts opposing managers to shareholders (La Porta et al, 2000; Knyazeva, 2007 and Adjaoud and Ben-Amar, 2010). Empirically speaking, some studies support the substitute agency model of dividends. Officer (2006) finds that firms with weak internal and external governance (large, insider dominated boards, entrenched managers, and low ownership levels by insiders and important external monitors) are more likely to pay dividends. Such firms also experience significantly more positive stock price reactions to dividend initiation announcements. These results show that dividend policy is used to compensate for other characteristics that create agency conflicts inside the firm. Knyazeva (2007) examines the effect of corporate governance on dynamic dividend behaviour. The results show that weakly governed managers engage in more dividend smoothing, have lower dividend variability, and fewer dividend cut. According to the author, governance quality has a negative effect on changes in the dividend level, which is in line with the substitute model predictions. Based on a sample of 8279 firms from 37 countries around the world, Truong and Heaney (2007) examine the interaction between the largest shareholders and dividend policy. They find a convex relation between large shareholding and dividend payout. The authors interpret this result as it reflects the classical agency view that the largest shareholder may act as a substitute for dividends in mitigating agency costs. Setia-Atmaja et al. (2009) investigate the role of dividends, debt, and board structure in the governance of the family controlled firms. Their results indicate that in such firms, dividends and debt are used as a substitute for independent directors. According to the authors, the findings are consistent with the notion that some governance mechanisms (e.g., the market for corporate control, institutional investor and compensation) play a less important role in mitigating agency conflicts, making the governance role of dividends and debt in family firms more effective Hypotheses Development In order to identify their governance quality, we have classified firms into two classifications with regard to the voting power of largest shareholders 1. First, we distinguish between

5 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 277 concentrated control and shared control. Several recent studies show that firms with strong governance are those with shared control. Gomes and Noveas (2005) argue that sharing control protects minority shareholders while preserving valuable private benefits of control. The governance role of shared control stems from two sources: the equity effect and the bargaining problems. The equity effect implies that fewer minority shares have to be sold to satisfy financing needs. Thus, controlling groups internalize firm value to a greater extent. It then follows that there are fewer incentives to inefficiently dilute minority shareholders. The bargaining problems may prevent business decisions that are in the collective interest of a controlling group, but harm minority shareholders. Bloch and Hege (2001) suggest that the presence of multiple shareholders reduces private benefits through competition for control. Laeven and Levine (2008) find that firms with multiple blockholders have a higher performance than firms with only one big blockholder. Second, we distinguish between the types of the controlling group. Firms are classified into family controlled, institutional controlled, outside individuals and state controlled. In an agency perspective, the problems opposing insiders to outside shareholders are affected by the type of the controlling group. Andersen and Reeb (2004) suggest that founding families may engage in self dealing by reducing firm risk, enriching themselves at the expense of minority investors, and generally representing their interests over those of the firm s other stakeholders. Filatotchev et al. (2005) argue that family control over the board may lead to greater executive entrenchment and potential agency conflicts with outside investors. Similarly, Villalonga and Amit (2006) suggest that controlling families have greater incentives for both monitoring and expropriation. The agency conflicts between insiders and outside shareholders may be duplicated in state-controlled corporations. As argued by Gugler (2003), in such firms, a double principleagent problem exists. The state acts on behalf of the citizens who do not control the firm directly. Agency problems can arise between citizens and state representatives and between state representatives and managers who look to their own interest. Contrary to these controlling shareholders, institutional investors are seen to be active in corporate governance. The literature recognizes that institutional investors serve a significant role as monitors in the stock market (Agrawal and Mandelker, 1992). The efficient monitoring hypothesis initiated by Bathala, Moon and Rao (1994) provide that institutional investors, by their expertise, can mitigate information asymmetry between insiders and outsiders. Li and Huang (2009) argue that these investors have more incentive to influence the corporation manager s decision compared to common investors because of their advantage of capital, technology and human resources. In another hand, financial literature recognizes the crucial role of outside blockholders in reducing the opportunistic behavior of managers (Shleifer, 1986). By their important holdings, large outside shareholders can reduce the problem of collective action stemming from dispersed ownership. In sum, firms with strong governance are those with majority control of outside blockholders or institutional investors and those with weak governance are firms controlled by families or the state. Hence, according to this discussion and having that the outcome and substitute models have opposite predictions about the relation between governance and dividend policy, we state our hypotheses as follow:

6 278 Moncef Guizani & Ezzeddine Abaoub H1 (outcome): dividend payout ratio increases with the voting power of the second largest shareholder, institutional investors and outside blockholders and decreases with the voting power of the largest shareholder, families and the state. H2 (substitute): dividend payout ratio increases with the voting power of the largest shareholder, families and the state and decreases with the voting power of the second largest shareholder, institutional investors and outside blockholders. 3. DATA AND METHODOLOGY 3.1. Sample Selection, Data Sources and Summary Statistics The base for the selection of our sample was the list of issuers of listed securities admitted to trading on a regulated market from the Tunisian securities market commission. The data were gathered from the annual reports of each company registered in the official bulletins of the Tunisian stock exchange (TSE) and the financial market council (FMC). The sample selection procedure is as follow: from the initial sample, we have eliminated banking and insurance companies because of their accounting specificity. We have also excluded companies which never have paid dividend during the period of analysis. We therefore end up with a final sample of 440 firm-year observations covering the years 1998 to It includes 12 financial firms, 23 industrial firms and 9 service firms. Table 1 summarizes the key characteristics of the sample firms. 2 The results show that the average dividend to earnings and dividend to free cash-flow ratios are 50% and 18%, respectively. The average free cash flow is 11.7% of total assets which indicate that the funds available to managers of Tunisian firms are relatively high. The existence of these funds may lead management to undertake sub-optimal investment projects. The sample mean values of debt, ROA and growth proxies equal 18.1%, 5% and 11.4%, respectively. Table 1 Summary Statistics for Pooled Sample (440 firm-years) Variable Mean St.dev. Min. Max. PAYOUT FCFDI FCF DEBT ROA GROWTH Ownership Concentration and the Measurement of Voting Power Panel A of Table 2 illustrates the distribution of equity blocks across different classes of shareholders. The most important categories of blockholders are families and financial institutions. Their average cumulative stakes are about 23% and 16%, respectively. Table 2 also shows that the other groups of investors (outside individuals and the state) held smaller fractions of equity outstanding (about 5% and 7%, respectively).

7 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 279 In addition to the dispersion of blocks across various types of shareholders, panel B of Table 2 analyses also ownership concentration. The average sizes of the largest and the second largest blocks equal 46% and 10% of the equity outstanding, respectively. This result indicates that ownership of Tunisian firms is highly concentrated and as the focal point of this paper is to investigate the relation between dividend policy and the voting power of specific types of shareholders, we construct various measures of voting power. Following Renneboog and Trojanowski (2006), we analyse a two-stage voting game. In the first stage, we assume that all the shareholders of a particular type form a coalition. In the second game, such coalitions participate in a voting game with the objective to influence the dividend policy. Table 2 Distribution of Equity Blocks Panel A- Distribution of equity blocks across different classes of shareholders Variable Mean St. dev. Min Max Family and executive directors Financial institutions Outside individuals State Panel B- Sizes of the largest blocks Largest block nd largest block The analysis of control, though it is crucial for an understanding of corporate governance, is still very much an open research area both empirically and theoretically. A naïve, yet quite often followed in the literature, approach uses just the size of the stake controlled by different blockholders. According to Trojanowski (2004), those stakes are assumed to be a (crude) proxy for the strength of a particular investor. For instance, a shareholder with 25% of votes in a dispersed ownership is very likely to exercise an effective control over the company, while a block of 30% of votes in a company with a majority shareholders does not give its holder significant influence unless supermajority requirements are imposed. The main problem with such an approach is that it ignores the stakes controlled by other shareholders. Crespi and Renneboog (2003) suggest that, it is the relative rather than the absolute voting power of a given investor, which determine his ability to extract private benefits of control. Hence, it seems simplistic to consider solely the percentage of shares held by the largest shareholder. A more general approach might take account of the possibility of coalitions of large shareholders being formed for the purpose of control. The approached adopted here is to use a game-theoretic approach to study the formal power represented by shareholder votes. The idea is to model shareholders as players in a voting game, and to use classical power indices to measure the extent of their control over a target company. Intuitively, such power indices reflect the relative ability of each player (shareholder) to impose his will to the target company through coalitions with other players. As Crama et al. (2003), we propose to use the Banzhaf index, which measures the ability of a voter

8 280 Moncef Guizani & Ezzeddine Abaoub to swing the decision in his or her own favor. More precisely, the Banzhaf index of a player can be defined as the probability that the outcome of the voting process changes when the player changes her mind unilaterally, under the assumption that all vectors are equally likely (see Banzhaf, 1965). To compute the Banzhaf indices, we have considered an oceanic game 3 which can be obtained as the Banzhaf indices for the modified, finite game consisting only of the major players M with weights w 1, w 2, w m and quota 4 q (1-w(M))/2 (Leech, 1999). We have developed an algorithm for the calculation of these indices (see appendix). We have used two procedures to compute the voting power of the largest shareholders. A one-stage voting game which measures the absolute Banzhaf indices of the largest and the second largest shareholder, and a two-stage voting game which measures the absolute Banzhaf indices of shareholder coalitions. Four coalitions are distinguished: Family, Financial institutions, outside individuals and the State. Table 3 (panel A and B) illustrates the distribution of voting power (as measured by Banzhaf indices, Z) among the two largest shareholders and different categories of controlling shareholders. The results confirm the considerable potential of families and financial institutions to influence corporate policies. Their voting powers are about 40% and 30%, respectively. Interestingly, panel B of Table 3 shows that the voting power of the largest shareholder is quite high (0.76) making him very powerful. The second largest shareholder has on average 0.09 of the voting power. These results confirm the concentrated ownership of Tunisian firms and indicate that the control is often in the hand of the first largest shareholder. Table 3 Voting Power of the Largest Blockholders Panel A- Two-stage voting game (absolute Banzhaf indices for shareholder coalitions) Variable Mean St. dev. Min Max Family and executive directors Financial institutions Outside individuals State Panel B- One-stage voting game (absolute Banzhaf indices for the largest shareholders) Largest block nd largest block Dividend Measures Consistent with prior research in finance (Jensen et al. 1992; Farinha, 2003; Poulain-Rehm, 2005 and Kowalewski et al, 2007), we use two measures of dividend payout ratios, the ratio of cash common dividends to net income, and dividend to free cash flow Control Variables A review of the literature revealed that dividend payout ratio is also determined by other variables. Table 4 summarizes the variable description of the study Free Cash-flow The free cash-flow hypothesis initiated by Jensen (1986) suggests that if firms have cash in excess of their requirement of investment in positive-npv projects, it is better to pay these

9 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 281 cashes as dividend in order to reduce managerial discretionary funds and thus avoid agency costs of free cash-flow. Jensen et al. (1992) and Mollah et al. (2000) find a support of this hypothesis, thus we predict a positive relationship between free cash-flow and dividend payout ratio. The ratio of free cash flow to total assets is considered as the proxy of free cash flow for agency cost arises for free cash flow Debt According to Jensen and Meckling (1976), Jensen (1986) and Stulz (1988) debt policy has an important role in monitoring managers thus reducing agency costs arising from the shareholdermanger conflict. Moreover, as argued by Kalay (1982) and Smith and Warner (1979), some debt contracts include protective covenants limiting the payout. Therefore, we expect a negative relationship between payout ratios and the level of debt. This variable is defined as the long term debt deflated by total assets (Jensen et al, 1992) Performance The financial literature documents that a firm s performance is a significant and positive explanatory variable of the dividend payout. Jensen et al. (1992) find a positive relation between the return on assets and the dividend payout ratio. Similarly to Farinha (2003) and Adjaoud and Ben-Amar (2010), we expect a positive relation between performance and dividend payout ratios. Following Jensen et al. (1992), we use the return on assets as a measure of firm performance Past Growth Our model predicts a negative relationship between the past growth and dividend payout ratio since firms prefer to avoid transaction costs due to external financing. According to the pecking order theory, we can expect firms to pay fewer dividends if they experienced past growth. Last studies such as the Rozeff s study find that dividend policy is negatively influenced by the past growth of the firm. As Rozeff (1982), we use the average of the historical sales growth over the period as a measure of past growth Market Listing Wallgren (2006) finds that listed companies have significantly higher payout ratios than nonlisted companies. This finding is supported by the outcome and the substitute model. On the one hand, in listed firms, minority shareholders are able to put pressure on the controlling shareholder to increase dividend payments as suggested by the outcome model. On the other hand, with conformity to the substitute model, in listed firms controlling shareholder has more incentives to signal to the outside investors that he is not extracting private benefits of control by employing a generous dividend policy. In agreement with these models, we expect a positive relation between market listing and dividend payout ratios. This variable takes the value 1 if the firm is listed and 0 otherwise Business Sector Poulain-Rehm (2005) suggests that dividend policy should be evaluated relatively to their competitor in the same business sector which leads to neutralize the effects of the

10 282 Moncef Guizani & Ezzeddine Abaoub economic conjecture. Two groups of firms are distinguished: Firms from the financial sector and other firms. Table 4 Variable Description Expected signs Variable Abbreviation Description Outcome Substitute Dependent variable Dividend payout ratios PAYOUT cash common dividends / net income FCFDI Dividend paid/free cash flow Governance variables Control structure Z j Z1: Banzhaf index of the largest shareholder - + Z2: Banzhaf index of the 2 nd largest shareholder + - ZFAM: Banzhaf index of families - + ZINST: Banzhaf index of financial institutions + - ZIND: Banzhaf index of outside individuals + - ZSTAT: Banzhaf index of the state - + Control variables Free cash-flow FCF (net income + amortization and + + depreciation dividends)/total assets Debt policy DEBT Long term debt/total assets - + Performance ROA Net income/total assets + + Past growth GROW Ratio of sales growth - - Market listing LIST 1 if the firm is listed in year t and 0 otherwise + + Business sector SEC 1 if financial sector, 0 otherwise +/- +/- 4. EMPIRICAL RESULTS 4.1. Univariate Analysis Governance Characteristics In order to specify the disciplinary role of dividend suited to various governance variables, we have carry out two series of tests over two firms classifications. The first classification consists in decomposing the full sample into three groups according to the level of dividend to earnings ratios 5. The first group (PAYL) contains firms whose mean payments are inferior to the first quartile of the full sample (34% in our study). The second group (PAYH) includes firms having mean distributions superior than the second quartile of the sample (50%). Finally, the intermediate position (between 34% and 50%) contains firms pertaining to the third group (PAYM). The second classification is based on the level of dividend to free cash flow ratios. Two groups are distinguished. The first group (HFCFDI) is composed by firms with dividend to free cash flow ratios higher than the median. The other group (LFCFDI) includes firms with dividend to free cash flow ratios lower than the median.

11 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 283 Panel A of Table 5 illustrates the mean and median tests of governance variables for the three groups of firms. To compare continuous variables means, we use the Kruskal-Wallis test. For the discrete variables, we present the frequency and the chi-2 test. The results show that firms of the group PAYH have, on average, a family control degree (27%) lower than those of the group PAYM and PAYH which have 52% and 60% respectively. In contrast, the financial institutions and outside individuals control degrees of firms belonging to the group PAYH are higher than those of firms belonging to the groups PAYM and PAYL. The mean voting power of financial institutions (outside individuals) is about 42% (10%) in firms paying higher dividend to earnings ratios against 24% (6%) and 6% (4%) in firms paying medium and lower dividends, respectively. Moreover, the voting power of the state seems to be lower for the group PAYH (13%) comparatively to the groups PAYM and PAYL in which the state holds 16% and 26% of the voting power, respectively. In another hand, as shown by panel A of Table 5, firms of the group PAYH have an important control degree of the second largest shareholder (13%) than those of the group PAYM (6%) and PAYL (4%). However, the largest shareholder has, on average, a voting power of 69% in the group PAYH against 82% for the group PAYM and 88% for the group PAYL. The mean comparison tests between the groups PAYL, PAYM and the group PAYH show significant differences for all the governance variables except for the variable ZSTAT. Table 5 Descriptive and Univariate Statistics for the Three Sub-samples Panel A: continuous variables Mean St. dev. Median comparaison Test Variable PAYL PAYM PAYH PAYL PAYM PAYH PAYL PAYM PAYH Mean 1 Median 2 ZFAM *** *** ZIND *** *** ZINST *** *** ZSTAT ** Z *** 8.47 ** Z *** *** FCF *** *** DEBT *** *** ROA *** *** GROW Kruskal-Wallis Test 2. Median test Panel B : discrete variables Frequency Variables PAYL PAYM PAYH Chi-2 Test PAYL-PAYM-PAYH LIST *** SEC *** ***, **,*: significant at 1%, 5% and 10% level.

12 284 Moncef Guizani & Ezzeddine Abaoub Panel A of Table 6 presents the results of mean comparison tests between the governance variables for the two groups HFCFDI and LFCFDI. They show significant differences between these groups concerning the voting power of families, financial institutions, the state, the largest shareholder and the second largest shareholder. Family shareholders hold a more important voting power in firms having lower dividend to free cash flow ratios. This voting power is, on average, about 47% for the group LFCFDI against 32.8% for the group HFCFDI. In contrast, it seems that the control degrees of financial institutions and the second largest shareholders are higher for the group HFCFDI. Their respective values are 44.1% and 11.3% against 16.7% and 7.1% for the group LFCFDI. Concerning the state, it appears that his control degree is largely higher for the group of firms that pay fewer amounts of their free cash flow as dividends. These results give some preliminary evidence suggesting that the disciplinary role of dividend depend on the governance characteristics of the firm. Table 6 Descriptive and Univariate Statistics for the Two Sub-samples Mean Panel A : continuous variables Variables HFCFDI LFCFDI Difference Probability ZFAM ZIND ZINST ZSTAT Z Z FCF DEBT GROW ROA Panel B : discrete variables Frequency Variables HFCFDI LFCFDI Chi-2 TestHFCFDI- LFCFDI LIST *** SEC ** Control Variables In our study, we have introduced some variables which can affect dividend policy. Some of them constitute a motivation for higher payment such as free cash flow and performance. Other variables constitute a constraint for higher payout ratios. Panel A of Table 5 shows that the group PAYH illustrates a proportion of free cash flow (18%) higher than their counterpart of the groups PAYM (9%) and PAYL (10%). Moreover, it seems that firms paying higher dividends are less leveraged and more profitable. On average,

13 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 285 their debt levels are 15% against 21% and 22% for the groups PAYM and PAYL, respectively. The return on assets is, on average, about 7% for the group PAYH and 3% for the two other groups. Kruskal-Wallis tests reveal significant differences between the three groups on the free cash flow, the debt and the performance variables. For the discrete variables, panel B of table 5 provides a clear disparity between the three groups of firms. The frequency of listed firms is 72% for the group PAYH, 41% for the group PAYM and 59% for the group PAYL. As for the business sector, firms of the groups PAYH and PAYM present similar frequencies (33% and 36%). However, none of the firms of the group PAYL belong to the financial sector. The chi-2 tests are all significant. For the second classification, table 6 suggests that the two groups of firms differ largely in the level of free cash flow, the level of debt and the return on assets. The comparison tests for these variables are significant. Similarly, for the discrete variables, HFCFDI group and LFCFDI group show a disparity in the frequency of listed firms and business sector Multivariate Analysis Model Specifications and Estimation Techniques The overall impression left by the univariate analysis of the three groups is that firms belonging to the group PAYH have, on average, governance characteristics characterized by low involvement of family shareholders and a strong implication of institutional shareholders and the second shareholder in comparison with the other two groups PAYM and PAYL. To further support this observation, we conduct multivariate tests linking the level of dividend distribution to the firms governance characteristics. In order to investigate the disciplinary role of dividend suited to the voting power of various large shareholders, we used panel data. According to Sevestre (2002), panel data presents an essential feature because of their double dimension, individual and temporal, unlike other types of data, time series and cross sections. In the context of panel data, it is often useful to identify the effect associated with each individual (an effect that does not vary over time, but varies from one individual to another). This effect can be fixed or random. Two critical tests provide the model specification. The first is the test of the presence of individual effects. The second is the Hausman test that determines if the coefficients of the two estimates (fixed and random) are statistically different. If both hypotheses are rejected, the fixed effects model will be retained. To reach our objective, we conducted a series of tests through which we regress the dividend to earnings ratio and the dividend to free cash flow ratio on the governance and control variables. Inspired from previous studies (Rozeff, 1982; Jensen, 1986; La Porta et al. 2000; Maury and Pajuste, 2002; Officer, 2006; Adjaoud and Ben Amar, 2010), we propose the two following models: PAYOUT it = Z jit + 2 FCF it + 3 DEBT it + 4 GROW it + 5 ROA it + 6 LIST it + 7 SEC i + it FCFDI it = Z jit + 2 FCF it + 3 DEBT it + 4 GROW it + 5 ROA it + 6 LIST it + 7 SEC i + it

14 286 Moncef Guizani & Ezzeddine Abaoub Verification of the Absence of Multicollinearity between Explanatory Variables Before the linear regression analyses will be performed, there will first be examined if there is multicollinearity between the variables of the research model. To reach this goal, we have calculated the Pearson Correlation Coefficients between the independent variables and the Variance Inflation Factor (VIF). As the panel A of Table 7 indicates, problems of multicollinearity between the variables of control structure are detected. The FIV tests reveal high values for the variables ZFAM, ZIND, ZINST and ZSTAT. Similarly, from panel B of Table 7, the VIF tests provide a multicollinearity problem between the Z score of the largest and the second largest shareholder. These results lead us to introduce each of these variables separately in estimating our models. For each of the independent variables, we estimate six models where we introduce the variables ZFAM, ZIND, ZINST, ZSTAT, Z1 and Z2 for each dependant variable. Table 7: Panel A: Correlation Matrix and VIF between Variables of the First Regression ZFAM ZIND ZINST ZSTAT FCF DEBT GROW ROA LIST SEC VIF ZFAM ZIND ZINST ZSTAT FCF DEBT GROW ROA LIST SEC Table 7: Panel B: Correlation Matrix and VIF between Variables of the Second Regression Z1 Z2 FCF DEBT GROW ROA LIST SEC VIF Z Z FCF DEBT GROW ROA LIST SEC Fixed Effects versus Random Effects Test of the presence of individual effects According to Ouellet et al. (2005), verifying the presence of individual effects consists to test the null hypothesis H 0 : i = 0 in the regression models above. The result is an F Fischer statistic with (N-1, NT-NK-1) degree of freedom. If we reject the null hypothesis, then we must include individual effects in the model.

15 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 287 In our study, the F statistics obtained by the STATA program are summarized in the following table: Table 8 Test of the Presence of Individual Effects Model Model 1.1 : Dep var.: Payout Model 1.2 : Dep var: FCFDI Model 2.1 Model 2.1 Retain. Var. ZFAM ZIND ZINST ZSTAT ZFAM ZIND ZINST ZSTAT Z1 Z2 Z1 Z2 Fisher F Prob ind. Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes According to these results, all probabilities have values of which strongly rejects the null hypothesis. Thus, we are able to include individual effects in the models to be tested. Hausman Test The Hausman test is a specification test for determining whether the coefficients of the two estimates (fixed and random) are statistically different. The idea of this test is that under the null hypothesis of independence between errors and explanatory variables, both estimators are unbiased, so the estimated coefficients should differ little. The Hausman test compares the variance - covariance matrix of the two estimators (Ouellet et al. 2005): W = ( f a ) var ( f a ) -1 ( f a ) The result follows a chi-2 with a K-1 degree of freedom. If we cannot reject the null hypothesis, that is to say if the p-value is greater than the level of confidence, then we use the random effects that are effective. In the case of our study, the results of Hausman tests of the proposed models allow us to retain the estimates with random effects for all models when all the probabilities (p-value) are above the confidence level of 10 %. Table 9 Hausman Test Model Model 1.1 : Dep var.: Payout Model 1.2 : Dep var: FCFDI Modèle 2.1 Modèle 2.2 Retain. Var. ZFAM ZIND ZINST ZSTAT ZFAM ZIND ZINST ZSTAT Z1 Z2 Z1 Z2 Chi Prob Effect Random Rand Rand Rand Rand Rand Rand Rand Rand Rand Rand Rand Regression Results The estimation results of our models (Tables 10 and 11) are satisfactory both econometrically as that of economic and financial interpretation. Expectations about the meaning of relationships, as revealed by the sign of the estimated values are all satisfactory. Moreover, the R 2 between (the R 2 most relevant for random effects) which constitutes a measure of the portion of inter-individual variability of the dependent variable explained by explanatory variables are satisfactory: they range from 38.57% to % for models whose dependent variable is PAYOUT, and between % and 59.48% for models whose dependent variable is the FCFDI.

16 288 Moncef Guizani & Ezzeddine Abaoub The probability of Breush-Pagan test shows that random effects are globally significant at 1% level. The results analysis will be divided into three parts: first, we examine the effect of the control power of various types of shareholders, then the effect of control concentration, and finally, the effect of control variables on the dividend payout ratios. Shareholders Coalitions Type The type of shareholders coalitions is described by the relative voting power of families, outside individuals, financial institutions and the state as measured by the Banzhaf index. Voting power of family shareholders According to the predictions of the outcome model (hypothesis 1), the voting power of family shareholders exerts a negative effect on the dividend payout ratios. From Table 10, the coefficient associated with ZFAM is negative and significant (p = 0.008), implying that the more corporate control structures are dominated by families or individuals involved in management, more they tend to distribute fewer dividends. This result challenges the disciplinary role of dividend policy when the supervisory power of the family shareholders is important. Moreover, firms with low distribution have, on average, higher scores ZFAM. This could be explained by several reasons. First, as emphasized by Fama and Jensen (1983), the family shareholders involved in management, because of the lack of diversification of their risk, will focus on financial resources that have low risk. This financial behavior is reinforced by their aversion to a dilution of their control via external funding. Therefore, they promote a higher retention of profits at the expense of a dividend distribution (Guizani et al. 2008). On the other hand, concerned about the sustainability of their business and conservation to their offspring, the family shareholders try to reduce the risk of their business while avoiding risky investment and accumulating more cash reserves. In addition, this type of business presents a specific feature: the remarkable presence of family members in the management and board of directors. This could cause a risk of excessive salaries and lack of competence of these members. These results are supported by multivariate analysis of the second regression whose dependent variable is the FCFDI. From Table 11, it appears that the coefficient associated with the variable ZFAM is negative (-0.09) at a significance level of 5%. This shows the reluctance of family shareholders to distribute free cash flows as dividends. It should be noted that the results of earlier writings are quite mixed. Calvi-Reveyron (2000) shows that French family firms appear less generous than the managerial or controlled firms. Contrasting with this result, Maury and Pajuste (2002) and Oreland (2006) found a positive relationship between family ownership and the level of dividend in the context of Finland and Sweden respectively. The authors explain their result by a tax effect. Voting power of outside individuals Our hypotheses resulting from the outcome model (the substitute model) stipulate that the level of dividend and the allocation of free cash flow for dividend distribution is positively (negatively) related to the degree of external control of private persons. From Tables 10 and 11, it appears that the coefficients of the variable ZIND are not significant. The voting power of outside individuals not part of the management team does not appear to increase their vigilance

17 Can Dividend Serve as a Disciplinary Mechanism? Evidence from Tunisia 289 with respect to the dividend distribution. This is not surprising given that outside individuals do not seem decisive in the voting process. Moreover, no company is majority controlled by individuals who are not leaders. This result is not consistent with the earlier writings. Trojanowski (2004) and Renneboog and Trojanowski (2006) reported a positive relationship between the proportion of control held by outside individuals and the dividend distribution level. - Voting power of financial institutions According to the outcome model, we hypothesized that the level of dividend payout is positively related to the control degree of financial institutions. The estimation results reported in Table 10 indicate that the coefficient associated with the variable ZINST is positive and significant, confirming the hypothesis of the outcome model and contrasting that of substitute model. To exercise strict control over the firm s management, financial shareholders adopt a generous dividend policy. This could be explained by the effectiveness of control exercised by financial institutions on the managers in accordance with the efficient monitoring hypothesis initiated by Bathala et al. (1994). By their ownership, these shareholders guide the firm s management in a way that promotes the interests of shareholders and reducing the extent of conflicts of interest that might arise within the company. More specifically, these shareholders can deter opportunistic decisions of managers and ensure the shareholder wealth maximization. This empirical evidence shows that dividend policy is the result of controlling shareholders preferences. Our result is consistent with those of Eckbo and Verma (1994), Trojanowski (2004) and Li and Huang (2009) who argue that institutional ownership has a positive effect on the level of dividend payment. However, Elston et al. (2004) show that neither institutional ownership nor bank control have a significant effect on the rate of dividend distribution of German companies. In addition to their positive effect on the dividend to earnings ratios, financial institutions promote distribution of free funds to the service of dividend policy. In accordance with the hypothesis 1, the control degree of financial institutions ZINST is associated with a positive coefficient (0.137) at a significance level of 1% (Table 11). Given the risk of conflict between shareholders and managers associated with the presence of free cash flow as postulated by Jensen (1986), companies with strong presence of financial institutions make large dividend distributions to reduce excess cash. This disciplinary effect can limit the investment in value destroying project. Voting power of the State Contrary to our hypotheses, Table 10 shows that the state voting power has no significant effect on the dividend payout despite the higher average Z score of the state in the group PAYL (26%) compared to groups PAYM (16%) and PAYH (13%). Similarly, the regression results of the model whose dependent variable is the FCFDI show no significant effect of the state voting power over the allocation of free cash flow for dividend distribution despite the significant difference of state control degree, as measured by the Banzhaf index, among firms with low and high dividend to free cash flow ratios (0.230 against 0.101, respectively). Control Concentration - Voting power of the Largest Shareholder As indicated by table 10, the influence of the voting power enjoyed by the largest shareholder on dividend to earnings ratios is significantly negative in accordance with the outcome