The dividend trend and changing determinants across periods

(1)

The dividend trend and changing determinants across periods Tony Chau

Finance and Organization student (10555269)

Abstract:

Between 1978 and 1999, the amount of dividend payers reduced from 66.5% to 20.8% (Fama & French, 2001). This one-sided reduction is not observed in the period 2002 to 2014. The propensity to pay dividends increased after the tax cut in 2003, decreased in the financial crisis, and increased again afterwards. In addition, the increase of the coefficient of investment opportunities in the recovery period shows that firms are more likely to signal when access to external funds is attractive. Moreover, the coefficient of profitability is significantly higher in 2011-2014 compared to 2003-2006. This suggest that firms became more risk averse after the financial crisis.

Keywords: dividend trend, determinants of dividend, risk aversion, signalling theory JEL Classification: G35

(2)

2

Statement of Originality

This document is written by Tony Chau who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for contents.

(3)

3

List of figures

1. Amount of dividend payers ... 20

2. Percentage dividend paying firms... 21

List of tables

1. Overview of previous studies ... 12

2. Explanatory variables, expected sign, theory ... 18

3. Descriptive statistics for dividends ... 19

4. Average and median for the independent variables ... 22

5. The dividend trend ... 23

6. Coefficients of determinants across periods ... 27

7. Difference in determinants across periods ... 30

(5)

5 1. Introduction

The dividend irrelevance theory of Miller and Modigliani (1961) is one of the most well-known theories about dividends. According to the dividend irrelevance theory, the value of a firm is independent of its dividend policy in a frictionless market. However, in the real market, there are market imperfections like taxes, information asymmetries, and transaction costs. These imperfections led to a development of various theories of dividend policies like the agency cost hypothesis and the signaling hypothesis.

The analysis of the determinants of dividend policy belongs to the core issues in modern financial theory (Breuer, Rieger & Soypak, 2014). Aside from the determinants of dividend payout, the dividend trend is also investigated. In 2001, Fama and French conducted research on dividends. They found a surprising reduction of the numbers of dividend paying firms from 1978 to 1999. This is known in the literature as the disappearing dividend puzzle (Fama & French, 2001; Fatemi & Bildik, 2011).

In this paper, an analysis will be conducted on the dividend trend and on the determinants of dividend payout policies. The focus will be on the period 2002-2014. This period includes the tax cut in 2003 and the financial crisis in 2007-2008. The period will be split into four sub periods: 2003-2006 as the tax cut period, 2007-2008 as the financial crisis, 2009-2010 as the recovery period, and 2011-2014 as the most current period. The research question: how did the dividend trend and determinants of dividend develop in the period 2002-2014?

The data needed for this analysis are retrieved from Compustat and the Center for Research in Security Prices (CRSP). We use a logit regression to examine the development of dividends and the development of the determinants of dividend payout. The variables we use are: profitability, investment opportunities, asset growth, size, life cycle, lagged profitability, and debt.

Our results reveal statistically significant relations between the propensity to pay dividends and its profitability, investment opportunities, asset growth, size, life cycle, and lagged profitability. The determinant debt is not significant in the tax cut period and the recovery period. In the financial crisis and the most current period, debt is significant but the coefficients are mixed. The coefficient of debt is positive in the financial crisis and negative in the most current period. Debt shows different effects on dividend payout policies across both

(6)

6

periods. In the financial crisis, high amount of debt is only attainable by more capable firms as interest obligations are higher and prospect on future earnings are lower. In the most current period, the high amount of debt is also attainable by less capable firms. However, as debt grows, the interest obligation increases, resulting in a higher risk of default. So, debt indicates the capabilities of firms in the financial crisis (positive); and in the most current period, debt indicates the default risk of a firm (negative).

In addition, the development of certain variables shows signs of behavioral aspects of the dividend payout policy. The variables investment opportunities and growth show signs of the signaling hypothesis which depends on the market environment. The coefficient of investment opportunities is significantly higher in the recovery period compared to the non-crisis periods. In the recovery period, access to loans is more difficult. In this case, if a firm has a high level of investment opportunities but insufficient cash, it is attractive to initiate dividend payout to appear healthy. By appearing healthy, it is easier to get access to external funds. The development of the growth rate adds support for the signaling hypothesis. In the recovery period, the coefficient of growth is significantly lower than in the non-crisis periods. This supports the signaling theory in the sense that firms which are able to grow, have less incentive to appear healthy by initiating dividends. Firms would rather use the money to invest in the increasing investment opportunities. Another behavioral element observed, is the risk aversion which can be seen by analyzing the variables profitability and lagged profitability. From 2002 to 2014, firms are becoming more risk averse. In the financial crisis, the determinant lagged profitability became a more important variable. Firms were less likely to pay if the results of last year were negative compared to other periods. Besides, in the most current period, the coefficient of profitability is significantly higher than the coefficient in the tax cut period. Profitability became more important after the financial crisis. Nowadays, firms which are unprofitable are less likely to pay dividends compared to the period before the crisis.

Furthermore, in contrast to the one-sided development of dividends in 1978 to 1999, we observe a volatile development of dividends in the period 2002 to 2014 after controlling for the independent variables. In the tax cut period, the propensity to pay dividends increased. In the financial crisis, the tendency to initiate dividends decreased. In the recovery period, the probability of firms paying dividends increased again. This development continued in the most current period.

(7)

7

The remainder of the paper is organized as follows. In section 2, we present the current state of research with respect to the trend and determinants of dividend payout policy. Section 3 describes our data sample. Section 4 provides the methodology used in this paper. Section 5 reports the evidence on the dividend trend and the evidence on the determinants of dividends. Section 6 is devoted to the robustness tests. Section 7 concludes.

2. Literature review

In this section, literature regarding dividends will be discussed. It will be split into three parts: the dividend trend, determinants of dividend, and behavioral influences on dividend payout policy.

One of the first researches documented on the dividend trend is from Fama & French (2000). They report a surprising finding of a reduced fraction of dividend payers, from 66.5% in 1978 to 20.8% in 1999. One of the explanations is the majority of newly listed firms with low profitability and strong growth opportunities. However, after controlling for profitability, investment opportunities and size, the percentage of dividend payers still declined. This trend is termed the ‘’disappearing dividends puzzle’’ (Fama & French, 2000). Although Fama & French (2000) conducted their research on firms in the US, the disappearing dividend trend is also found in other countries. Over the period 1989-2002, hints of reduction in the propensity to pay dividends were found in Canada, UK, France, and Japan (Denis & Osobov, 2008). In addition, Fatemi & Bildik (2011) have studied more than 17,000 firms from 33 different countries and found evidence in support of a significant worldwide decline in the propensity to pay dividends. These two papers show that most of the decline is due to the increase of firms which are smaller, less profitable and with more investment opportunities. Nevertheless, even after controlling for these determinants, the proportion of firms paying dividends has declined. These findings are in line with Fama & French (2000) supporting the disappearing dividend trend.

Contradicting the disappearing dividend trend, DeAngelo, DeAngelo & Skinner (2004) found increased dividends over the period 1978-2000, in both nominal and in real terms (224.6% and 22.7%). The reason is that the reduction in payers occurs almost entirely among firms that paid very small dividends. The top payers on the other hand increased their dividends. This resulted in a lower percentage of firms paying dividends but an increase in the aggregate supply. These findings are supported by other papers concentrated on firms outside

(8)

8

the US (Denis & Osobov, 2008; Fatemi & Bildik, 2011).

Thereafter, in the period 2000-2006, dividends reappeared (Julio & Ikenberry). The Jobs and Growth Tax Relief Reconciliation Act of 2003 (JGTRRA) lowered the tax rate on dividends to 15%. This reduced the long-standing tax disadvantage of dividends compared to share repurchases (Floyd, Li & Skinner, 2015). A large number of firms initiated or increased regular dividend payments after the reform. The proportion of firms paying dividends increased after two decades of continuous decline (Chetty & Saez, 2005; Blouin, Raedy & Shackelford). However, Julio & Ikenberry (2005) argued that the increase had started before 2003, questioning the causality of the tax cut. This critique is shared by Floyd, Li & Skinner (2015) and Edgerton (2013). To confirm the effect of a tax cut on dividends, Chetty & Saez (2005) applied an additional test using firms with high levels of non-taxable institutional ownership as a ‘’control group’’. The firms in the control group did not change payout policies while the firms in the normal group did, supporting the causality of the tax cut as determinant of dividend payments. A remark given by Chetty & Saez was the major accounting fraud scandals in 2000-2002. They argued that this might had created distrust among shareholders and increased the demand for dividends.

At last, the period with the financial crisis. Before the crisis, dividends increased at a pace that is impressive in absolute terms. In 2008 and 2009, dividends declined by a total of 5%. In the subsequent years, dividends rebounded to levels above the 2007 peak (Floyd, Li & Skinner, 2015). A similar result has been found by Hauser (2013). Even after controlling for profitability, size, and investment opportunities, the reduced probability of paying dividends is significant at the 1 percent level (Hauser, 2013).

The first determinants that will be discussed are profitability, investment opportunities, and size; these variables have been tested by several economists (Fama & French, 2000; Gill, Biger & Tibrewala 2010; Denis & Osobov, 2008; Fatemi & Bildik, 2011). The findings are all the same: larger firms, firms with higher profitability, and firms with lower growth opportunities have a greater propensity to pay dividends. The empirical determinants of the propensity to pay dividends appear to be similar across countries worldwide (Denis & Osobov, 2008; Fatemi & Bildik, 2011). Other papers, which tried to determine other relevant variables for dividend payout, often use these three variables in their models (Kuo, Philip & Zhang, 2013; Hoberg & Probhala, 2009; DeAngelo, DeAngelo & Stulz, 2006; Breuer, Rieger & Soypak, 2014).

(9)

9

Furthermore, DeAngelo et al. (2006) tested the life-cycle theory by assessing whether the probability of a firm paying dividends is related to its mix of earned and contributed capital. According to the life-cycle theory, young firms pay few dividends because their investment opportunities exceed their internally generated capital while they have limited access to external capital. In later years, internal funds exceed investment opportunities so firms pay out excess funds to mitigate the possibility that the free cash flows would be wasted (Denis & Osobov, 2008). The evidence of DeAngelo et al. (2006) indicates that the probability a firm pays dividend increases with the relative amount of earned equity in its capital structure. The effect of the life-cycle theory on dividends across different countries is empirically supported by Denis and Osobov (2008) and Ferris et al. (2009).

Moreover, risk as potential determinant of dividends is examined in previous papers. According to Hoberg & Prabhala (2009), risk is a significant determinant of the propensity to pay dividends and it explains roughly 40% of the disappearing dividends puzzle. Kuo, Philip & Zhang (2013) shares the same view indicating that risk plays a major role in firms’ dividend policy.

In addition, debt is a determinant of the propensity to pay dividends (Gill, Biger & Tibrewala, 2010). Dividend payments increase the risk of default by reducing the amount of cash that is available for debt obligations. Firms with higher debt-equity ratios should favor lower dividend rates. This is especially important for firms with higher risk (Breuer, Rieger & Soypak, 2014). Furthermore, debt is often used to complete models in order to determine other potential variables that influence the propensity to pay dividends (Breuer, Rieger & Soypak, 2014; Grullon & Michaely, 2002; Hobbs & Schneller, 2012; Kuo, Philip & Zhang, 2013). Nevertheless, Gill, Biger & Tibrewala found conflicting evidence for the relationship between dividend payout ratios and leverage. They found a positive relationship between debt and dividend for both the service and the manufacturing industries.

Besides the standard determinants, dividend initiations incorporate some behavioral influences. The signaling theory posits that firms convey their optimism for the future by initiating dividend payments. Lintner (1956) interviewed managers from 28 companies to determine which factors were the most important for payout policies. An important factor is the belief in the sustainability of company earnings over the long-term. Managers tend to increase or initiate payouts only when expected long-term profits would be high. 50 years later, Brav et al. (2005) conducted in-depth interviews with 384 financial executives to

(10)

10

determine factors that drive dividend and share repurchase decisions. The results of Brav et al. (2005) indicate that maintaining the dividend level is as important as investment decisions, while repurchases are made out of the residual cash flow after investment spending. Although managers express a strong desire to avoid dividend cuts, increases in dividends are considered only after investment and liquidity needs are met. Managers reject the notion to pay dividends as a costly signal to inform the firm’s true worth or to separate the firm from competitors. Another paper contradicting the signaling theory is from DeAngelo et al. (2004). The finding that dividends are highly concentrated among a small number of firms with high amount of earnings and a bigger size raises doubt on the signaling theory. If dividends are used to communicate with stockholders, dividend signaling should occur primarily in small, relatively unknown firms with limited information outlets. But the vast majority of dividends are paid by prominent corporations that enjoy coverage by analysts and journalists, whose managers should have little need to communicate with shareholders. Furthermore, the mix of earned/contributed equity (earned equity (retained earnings) divided by contributed equity (book value of equity)) as determinant of dividend policies casts doubt on the importance of signaling (DeAngelo et al., 2006; Denis & Osobov, 2008). Firms with low earned/contributed equity would be the ideal candidates for dividend signaling because these firms are less mature and it is therefore more difficult to gauge their future prospects. Yet, these firms are not likely to pay dividends. Another potential behavioral influence on dividend payout policy is the agency costs of free cash flow. According to the free cash flow hypothesis (Jensen & Meckling, 1976), managers invest in projects with negative net present value to increase personal utility. By investing, firms grow in power and company size which is more prestigious to control (empire building). Such an overinvestment problem can be counteracted by increasing dividend payments so that the free cash flows available to management diminish. This view is supported by Floyd, Li & Skinner (2015), suggesting that the free cash flow explanation is the most important reason dividends survive. On the contrary, Gill et al. (2010) who used the net cash flow variable in their model to test determinants of dividend in the US found no significant evidence regarding the relation between cash flow and dividend payout policy.

Furthermore, some researches have been conducted on the catering theory of dividends. Baker & Wurgler (2004) developed the catering theory, arguing that managers will modify corporate payout policy when the investors prefer the payment of dividends and omit

(11)

11

dividends when investors put a discount on dividend paying firms. The catering effect among US firms is further confirmed by Li & Lie (2006). However, Hoberg & Prabhala (2009) and Kuo, Philip & Zhang (2013) found no evidence for the presence of catering incentives after incorporating risk. The catering effect outside the US is also inconsistent. In the UK, Ferris et al. (2006) reported that catering incentives had held explanatory power in the propensity to pay dividends while the results of Denis & Osobov (2008) indicated the opposite. The evidence regarding the presence of dividend catering developed by Baker & Wurgler (2004) is mixed. Literature provides no consensus among scientists on this matter.

Nevertheless, uniform evidence on the relation between stock options and dividends has been found, indicating a clientele effect. Firms whose managers have stock options in their compensation package are less likely to pay dividends (Shapiro & Zhuang, 2014; Fenn & Liang, 2001; Kahle, 2002; Chetty & Saez, 2005). Additionally, managers with more restricted stocks have an incentive to increase dividends for liquidity reasons (Brown, Liang & Weisbenner, 2007).

At last, Breuer et al. (2014) empirically verified the relation between dividends and patience, loss aversion, and ambiguity aversion. They found that impatient investors prefer firms that pay out a large share of their earnings as dividends. Patient investors who want to save, but lack the willpower to do so, prefer companies who retain earnings. In both scenarios, the dividend payout policy should account for the time preference of the investors. In addition, dividends are certain (‘‘a bird in the hand’’), while retained earnings lead to uncertain future earnings so that ambiguity averse investors prefer dividends, even if retained and future earnings are completely reflected in the current stock prices. This is a pure psychological phenomenon, because investors can obtain the same amount of cash by selling their stocks. The study of Cyert and March (1993) (as cited in the Breur et al., 2014) argues that people prefer dividend payments to retained earnings because they try to avoid uncertainties as much as possible. Last of all, investors who are risk averse, prefer dividends due to a lower exposure of future losses. If a future shock has the potential to cause a negative return on the stock, dividends can be utilized in order to reduce the exposure to potential future losses.

(12)

12 Table 1 overview of previous studies

Author(s) Region Time

period

Method Control variables Results

.Fama & French (2001) US 1926-1999 Logit regression - Profitability - Size - Investment oppurtunities Significant decline of dividends from 1978 to 1999. Deangelo et al. (2004) US 1978-2000 Descriptive statistics Decrease of the amount of dividend payers but an increase of the total amount of

dividends.

Chetty & Saez (2005)

US 1980-2004 Descriptive

statistics

- 20% increase in dividend payments following the tax cut. - Firms with high levels of non-taxable institutional

ownership did not change pay-out policies, supporting the causality of the tax cut. Brav et al. (2005) 384 financial executives in the US

2002 Surveys - Maintaining the

dividend level is on par with investment decisions. Deangelo et al. (2006) US 1973-2002 Logit regression - Size - Profitability - Lagged profitability - Growth - Cash balance - Equity - Life cycle Life cycle as significant determinant of dividend payout policy.

Denis & Osobov (2008) - US - Canada - UK - Germany - France - Japan 1994-2002 Logit regression - Profitability - Investment opportunities - Growth - Size - Life cycle

- Doubts about the signaling, clientele, and catering explanations for dividends.

- Support for the life cycle variable - Propensity to pay dividends are fairly small.

(13)

13 Table 1 continued

period

Hoberg & Probhala (2009) US 1963-2004 Logit regression - Size - Profitability - Growth - Investment opportunities - Risk - Risk is a significant determinant of dividend payout, and it explains roughly 40% of the disappearing dividend trend of Fama and French (2001).

- After adjusting for risk, there is little support for the catering hypothesis of dividends. Gill, Biger &

Tibrewala (2010) US 2007 Descriptive statistics and regression - Profitability - Cash flow - Tax - Growth - Investment opportunities - Debt - Statistically significant relation between dividends and profitability, growth, and debt.

Fatemi & Bildik (2011) 33 countries 1985-2006 Logit regression - Profitability - Size - Investment oppurtunities - Significant worldwide decline in the propensity to pay dividends. - Aggregate

dividends are highly concentrated in that they are paid only by a small group of firms. Hauser (2013) US 2006-2009 Logit regression - Equity - Cash - Profitability - Growth - Size - Life cycle - Decline of the probability to pay out dividends in 2008 and 2009.

Kuo, Philip & Zhang (2013) 18 countries 1989-2011 Logit regression - Investment opportunities - Growth - Profitability - Size - Debt - Life cycle - Stock liquidity - Risk

- Risk and stock liquidity play a major role in the dividend payout policy of firms.

- After adjusting for risk, there is little support for the catering hypothesis of dividends.

(14)

14 Table 1 continued

period

Breuer, Rieger & Soypak (2014)

29 countries 1992-2012 Tobit and logit regression model - Size - Debt - Growth - Profitability - Life cycle - Institutional ownership - Country specific control variables

- Loss aversion and ambiguity aversion are significant and positively related to dividend payout. - Patience is significant and negatively related to dividend payout. Floyd, Li & Skinner (2015) US industrials and banks 1980-2012 Descriptive statistics - Dividend payouts of industrials and banks increased before the crisis at a pace that is

impressive both in absolute terms and relative to earnings. - The ability to address the free cash flow problem is a plausible

explanation for the survival of dividends.

3. Data

We analyze the dividend trend and the dividend decisions of industrial firms in the US from 2003 to 2014. Chetty & Saez (2005) found a significant effect of the tax cut in 2003 (JGTRRA) on dividend payout decisions. This tax cut is extended in 2012 by the American Taxpayer Relief Act. The analysis will be conducted on a period where the tax cut is still in place and we will therefore use 2003 as starting year in our data sample. The financial data are retrieved from Compustat, including firms with the following data: total cash dividend (TC), earnings before interest and taxes (EBIT), book value of assets (A), book value of equity (BE), market value of equity (ME), retained earnings (RE), and total debt (D). The holding period returns of shares are retrieved from the Center for Research in Security Prices (CRSP). The sampling procedure for this paper parallels those of Fama & French (2001), DeAngelo et al. (2006), and Denis & Osobov (2008). Specifically, we restrict analysis to nonfinancial and nonutility firms due to regulatory issues or uncommon capital strucutre in these industries. Utility firms and financial firms with SIC code 4900-4949 and 6000-6999 are excluded. To avoid survivorship bias, active

(15)

15

and non-active firms are included in the dataset. In addition, to be included in our sample for a given year, a firm must have non-missing values for the book value of assets and earnings before interest and taxes in year t and t – 1. The other financial data must be non-missing for year t. Furthermore, observations with negative cash dividends, a negative book value of equity, or negative book value of assets are excluded. At last, only fiscal year-ends data are included in the data sample.

4. Methodology

Following prior research on dividend policy (Fama & French, 2001; DeAngelo et al., 2006; Denis & Osobov, 2008; Hoberg & Prabhala, 2009; Fatemi & Bildik, 2011; Kuo, Philip & Zhang, 2013), a logit model will be employed to examine the determinants of dividend payout policy. The dependent variable Yi is one if the firm pays dividends in year t and zero otherwise. The explanatory variables are: profitability (P), investment opportunities (IO), asset growth (AG), size (S), life cycle (LC), lagged profitability (LP), and debt (D). To determine the significance of the determinants, the following logit regression will be used:

Y

i

=

𝑙𝑜𝑔𝑖𝑡 (β0 + β1(P) + β2(IO) + β3(AG) + β4ln(S) + β5(LC) + β6(LP) + β7(D))+ ε

β0 is the regression intercept, β1 β2 β3 β4 β5 β6 are the regression coefficients. If the regression coefficient is above zero, the independent variable has a positive effect on the propensity to pay dividend. Likewise, if the regression coefficient is below zero, the independent variable has a negative effect on the propensity to pay dividend.

Independent variables: - Profitability (P):

𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑏𝑒𝑓𝑜𝑟𝑒 𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑎𝑛𝑑 𝑡𝑎𝑥𝑒𝑠 (𝐸𝐵𝐼𝑇) 𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (𝐴)

This profitability measure is also used by Breuer, Rieger & Soypak (2014). When profitability increases, firms are more likely to pay dividends (Fama and French, 2001; DeAngelo et al., 2006; Hobbs & Schneller, 2012; Breuer, Rieger & Soypak, 2014).

(16)

16

- Investment opportunities (IO): 𝑀𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (𝑉)

𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (𝐴)

The market value of assets is derived the following way: the book value of total assets - book value of equity + market value of equity. This measure is used in several papers (Fama and French, 2001; Grullon & Michaely, 2002; DeAngelo et al., 2006; Denis & Osobov, 2008; Hoberg & Prabhala, 2009; Gill, Biger & Tibrewala, 2010; Hobbs & Schneller, 2012; Kuo, Philip & Zhang, 2013). High level of investment opportunities will result in lower propensity to pay out dividend.

- Asset growth (AG):

𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡 𝑖𝑛 𝑡ℎ𝑒 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑝𝑒𝑟𝑖𝑜𝑑 𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑝𝑒𝑟𝑖𝑜𝑑

This growth variable measures the actual growth of assets. This measure is also used in several papers (Fama and French, 2001; DeAngelo et al., 2006; Denis & Osobov, 2008; Hoberg & Prabhala, 2009; Kuo, Philip & Zhang, 2013). Dividend payments are less likely to be made by firms with high asset growth.

- Size (S):

𝑁𝑎𝑡𝑢𝑟𝑎𝑙 𝑙𝑜𝑔𝑎𝑟𝑖𝑡𝑚𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠

This measure is used by Grullon & Michaely (2002), Hobbs and Schneller (2012), and Breuer, Rieger & Soypak (2014). Bigger firms have a higher propensity to pay dividend (Fama and French, 2001; DeAngelo et al., 2006; Breuer, Fatemi & Bildik, 2011; Hobbs & Schneller, 2012; Rieger & Soypak, 2014)

- Life cycle variable (LC): 𝑅𝑒𝑡𝑎𝑖𝑛𝑒𝑑 𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 (𝑅𝐸) 𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑒𝑞𝑢𝑖𝑡𝑦 (𝐵𝐸)

This life cycle variable is tested by DeAngelo et al. (2006). The earned/contributed capital mix is a logical proxy for the life-cycle stage at which a firm finds itself because it measures the extent to which the firm is self-financing or reliant on external capital. Firms with a low RE/BE tend to be in the capital infusion stage, whereas firms with high RE/BE tend to be more mature with ample cumulative profits, hence good candidates

(17)

17

to pay dividends (DeAngelo et al, 2006). The life cycle variable is empirically supported by Denis & Osobov (2008) and Ferris et al. (2009).

- Lagged profitability

𝐸𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑏𝑒𝑓𝑜𝑟𝑒 𝑖𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑎𝑛𝑑 𝑡𝑎𝑥𝑒𝑠 (𝐸𝐵𝐼𝑇)

𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (𝐴) (𝑡 − 1)

The lagged profitability measure is derived the same way as the profitability measure, the only difference is the time period. The decision to pay dividends is not only based on the current profitability but also on past results. To account for this effect, the profitability of the past year is included. This measure is also used by DeAngelo et al. (2006).

- Debt

𝑇𝑜𝑡𝑎𝑙 𝑑𝑒𝑏𝑡 (𝐷) 𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑎𝑠𝑠𝑒𝑡𝑠 (𝐴)

This measure is used by Grullon & Michaely (2002) and Hobbs & Schneller (2012). Debt decreases the propensity to pay dividends (Breuer, Rieger & Soypak, 2014; Hobbs & Schneller, 2012). By increasing debt, the interest obligation increases, reducing the cash available for cash dividends.

Behavioral determinants of dividend payout policy will not be included in the model because no consistent results can be found in the literature. The empirical results of the signaling theory, agency costs, and catering theory are mixed. There is no consensus among researchers on behavioral determinants of dividend payout. Moreover, in this paper, we focus on the dividend trend and changing determinants across periods. By doing so, it is important to include determinants with a significant effect. As a result, behavioral determinants of dividend payout policy with mixed empirical results will not be included.

To account for potential input errors, values outside the range of ten times the standard deviation of the mean are excluded in our analysis. A negative impact of the exclusion is the elimination of real data. Nevertheless, the vast majority of the initial sample is still used (>95%). So, we still account for the vast majority and only eliminate the very extreme outliers. The outliers have a significant impact on the logit regression model. After

(18)

18

the exclusion, the empirical results are more similar to the results in preceding papers. In the robustness check section, the effect of the exclusion will be discussed in more detail. To account for multicollinearity, a variance inflation factor (vif) test is employed. The highest vif is 2.61 (see appendix A, table 1). Many practitioners use a vif of 10 as a sign of severe or serious multicollinearity (O’brien, 2007). A vif of 2.61 is below 10 and therefore acceptable, so multicollinearity is not a problem for our analysis.

Table 2: Explanatory variables, Expected sign, Theory

Independent variable Expected sign Theory

Profitability (P) + Profitable firms are more likely to

pay dividends

Investment opportunities (IO) - High level of investment opportunities result in lower dividend payout

Asset growth (AG) - Firms are less likely to pay dividends

when asset growth is high

Size (S) + Bigger firms are more likely to pay

dividend

Life cycle (LC) + Mature firms are more likely to pay

dividends than younger firms Lagged profitability (LP) + High profitability in the preceding

year increases the propensity to pay dividends

Debt (D) - Higher debt results in lower dividend

pay out

In order to investigate the dividend trend and the changing determinants of dividend, interaction variables have to be added in the logit regression model. In this way, we can compare two separate periods:

Y

i

=

𝑙𝑜𝑔𝑖𝑡 (β0 + β1(P) + β2(IO) + β3(AG) + β4ln(S) + β5(LC) + β6(LP) + β7(D) + D0 + D1(P) + D2(IO) + D3(AG) + D4ln(S) + D5(LC) + D6(LP) + D7(D)) + ε

(19)

19

In this case, β0 β1 β2 β3 β4 β5 β6 are the regression intercept and regression coefficients of the previous period (t-1). D0 is the difference in intercept between the current period and the previous period. If this variable is positive (negative) and significant, we can conclude that the propensity to pay dividend has increased (decreased) compared to the previous period, controlling for profitability, investment opportunities, asset growth, size, life cycle variable, lagged profitability, and debt. D1 D2 D3 D4 D5 D6 are the differences in regression coefficients between the current and previous period. If these interaction variables differ significantly, we can conclude that the effect of determinants of dividend payout policy differs across the two periods.

5. Results and discussion

In this section, the empirical results will be analyzed and discussed. At first, the descriptive statistics of dividends and the descriptive statistics of independent variables are reviewed. In the second part, we focus on the dividend trend while controlling for the independent variables. In the last part, the emphasis is on changing determinants across periods.

Table 3 Descriptive statistics for dividends

The reported values are the amount of dividend payers, amount of non-dividend payers, amount of total firms, and the percentage dividend payers from 2002 to 2014.

In table 3, we can see how dividends have been developing in the recent years. The amount of dividend payers increased after the tax cut in 2003. When the financial crisis started in 2007, the number of dividend paying firms decreased. From 2009 onwards, the amount of dividend

2002 (1) 2003 (2) 2004 (3) 2005 (4) 2006 (5) 2007 (6) 2008 (7) 2009 (8) 2010 (9) 2011 (10) 2012 (11) 2013 (12) 2014 (13) dividend payers 1021 1111 1216 1263 1230 1165 1100 995 1031 1069 1183 1161 1195 non-dividend payers 3471 3137 2992 2846 2783 2781 2568 2523 2409 2272 2083 2131 2160 total 4492 4248 4208 4109 4013 3946 3668 3518 3440 3341 3266 3292 3355 percentage dividend payers 0.227 0.262 0.289 0.307 0.307 0.295 0.300 0.283 0.300 0.320 0.362 0.353 0.356

(20)

20

payers increased except in 2013. When we account for the total amount of firms, a similar trend is observed. The percentage dividend payers increased after the tax cut and increased after the financial crisis. Nevertheless, a difference is perceived in 2008. While the number of dividend payers decreased, the percentage dividend payers increased. The similarities and differences between the percentage and amount of dividend payers can easily be observed in graph 1 and graph 2 below. The difference between the percentage and the amount of dividend payers is due to a higher decrease in the category total firms (table 3, row 4) relative to the decrease in the category dividend paying firms (table 3, row 1). In 2008, the total amount of firms was 3668, a reduction of 278 firms compared to 2007. The average reduction of total firms over the period 2002-2014 is 95 per year. The reduction of 278 firms indicates a higher amount of bankruptcies in 2007. Because the financial crisis started in the summer of 2007 (Reinhart & Rogoff, 2008), most firms that went bankrupt in 2007 could not survive the effects of the economic downturn for about four months. So, it can be assumed that the firms that went bankrupt in 2007 were the ‘’weaker’’ firms, leaving the more capable firms in 2008. From this we see that the dividend payout decision is based on certain company characteristics. This will be accounted for by using the independent variables, which is discussed next.

Graph 1 Amount of dividend payers

The graph shows the development of the amount of dividend payers from 2002 to 2014.

0 200 400 600 800 1000 1200 1400 2000 2002 2004 2006 2008 2010 2012 2014 2016 amou n t o f d iv id en d p ay ers year

(21)

21 Graph 2 Percentage dividend paying firms

The graph shows the development of the percentage dividend payers from 2002 to 2014.

The independent variables in this paper are profitability, life cycle, investment opportunities, size, growth rate, lagged profitability, and debt. The development of the determinants can be observed in table 4. The main focus is on the median values because the averages can be distorted by large positive values and large negative values. The independent variables profitability, life cycle, investment opportunities, and growth rate follow a similar trend as the amount of dividend paying firms: increase after the tax cut, decrease during the financial crisis, and an increase after the financial crisis. For the period 2011-2014, no consistent pattern can be found for these variables. Although most independent variables show the similar trend as the amount of dividend payers, some determinants of dividend payouts show a completely different development. First of all, the variable size grows every year, it does not even decrease during the financial crisis. Secondly, the variable lagged profitability is unstable, it decreases and increases in every period. At last, the variable debt ratio shows the opposite trend compared to dividend. It decreases after the tax cut, increases during the financial crisis, and decreases after the crisis.

0.000 0.050 0.100 0.150 0.200 0.250 0.300 0.350 0.400 2000 2002 2004 2006 2008 2010 2012 2014 2016 p erce n ta ge d iv id en d p ay ers year

(22)

22

Table 4 average and median for the independent variables

The reported values are the average and median values for profitability, investment opportunities, asset growth, size, life cycle, lagged profitability, and debt from 2002 to 2014.

2002 (1) 2003 (2) 2004 (3) 2005 (4) 2006 (5) 2007 (6) 2008 (7) 2009 (8) 2010 (9) 2011 (10) 2012 (11) 2013 (12) 2014 (13) Average profitability -0.064 -0.024 -0.014 -0.003 -0.015 -0.028 -0.033 -0.009 0.009 0.004 -0.010 -0.020 -0.039 Median profitability 0.035 0.043 0.058 0.062 0.060 0.055 0.055 0.043 0.063 0.063 0.059 0.054 0.050 Average lifecycle -2.599 -2.404 -2.184 -2.082 -1.993 -1.985 -2.286 -2.078 -2.112 -2.026 -2.293 -2.142 -2.168 Median Lifecycle -0.026 0.026 0.049 0.065 0.083 0.081 0.062 0.111 0.137 0.134 0.116 0.079 -0.001 Average investment opportunities 1.613 2.274 2.353 2.272 2.273 2.223 1.487 1.796 2.006 1.869 1.965 2.343 2.318 Median investment opportunities 1.180 1.594 1.713 1.696 1.722 1.616 1.126 1.346 1.462 1.346 1.403 1.673 1.687 Average size 4.901 5.077 5.223 5.352 5.463 5.543 5.579 5.689 5.766 5.860 5.945 5.990 6.035 Median Size 4.865 5.044 5.194 5.301 5.416 5.468 5.542 5.634 5.755 5.877 5.945 6.026 6.061 Average growth rate 1.038 1.149 1.238 1.212 1.245 1.257 1.056 1.068 1.179 1.171 1.129 1.194 1.211 Median growth rate 0.993 1.053 1.096 1.073 1.091 1.089 0.993 1.007 1.066 1.060 1.047 1.054 1.048 Average lagged profitability -0.068 -0.048 -0.025 -0.007 -0.006 -0.023 -0.002 -0.007 -0.019 0.009 0.006 -0.021 -0.035 Mean lagged profitability 0.029 0.040 0.044 0.060 0.062 0.059 0.062 0.062 0.043 0.063 0.065 0.058 0.052 Average debt ratio 0.165 0.161 0.150 0.149 0.152 0.158 0.167 0.157 0.150 0.157 0.166 0.171 0.182 Median debt ratio 0.104 0.101 0.089 0.087 0.092 0.095 0.100 0.089 0.083 0.099 0.110 0.116 0.000

(23)

23 Table 5 the dividend trend

The reported values without interaction variable D are the coefficients of profitability (P), investment opportunities (IO), asset growth (G), size (S), life cycle (LC), lagged profitability (LP), and debt (D) of the logit regression model of last year. The values with the interaction variable D represent the difference in coefficients of the current year and last year. The main focus is on variable D0 (D0<0 indicates a lower propensity to pay out dividends and DO>0 a higher propensity to pay out dividends in the current year compared to last year).

* Statistical significance at 10% level, ** Statistical significance at 5% level, *** Statistical significance at 1% level 2003 (1) 2004 (2) 2005 (3) 2006 (4) 2007 (5) 2008 (6) Int -3.033 (-15.85)*** -2.235 (-11.52)*** -2.404 (-13.36)*** -2.47192 (-13.62)*** -2.441 (-13.49)*** -2.199 (-11.80)*** P 1.215 (2.56)** 2.034 (3.51)*** 0.846 (1.74)* 1.448 (2.83)*** 1.219 (2.64)*** 1.768 (3.42)*** LC 0.038 (2.19)** 0.053 (3.11)*** 0.048 (3.26)*** 0.064 (3.80)*** 0.070 (4.01)*** 0.063 (3.47)*** IO -0.065 (-1.34) -0.136 (-3.57)*** -0.051 (-1.62) -0.060 (-1.87)* -0.074 (-2.05)** -0.098 (-2.78)*** S 0.398 (17.12)*** 0.371952 (16.54)*** 0.352 (16.06)*** 0.356 (15.77)*** 0.331 (14.58)*** 0.328 (14.20)*** G -0.340 (-2.68)*** -0.520 (-3.94)*** -0.351 (-3.67)*** -0.340 (-3.72)*** -0.289 (-3.50)*** -0.513 (-5.26)*** LP 2.578 (5.37)*** 1.351 (2.74)*** 1.783 (3.70)*** 1.278 (2.49)** 1.631 (3.40)*** 1.669 (3.27)*** D 0.141 (0.61) -0.557 (-2.28)** -0.051 (-0.21) 0.392 (1.66)* 0.600 (2.55)** 0.403 (1.73)* D0 0.798 (2.93)*** -0.169 (-0.64) -0.068 (-0.27) 0.031 (0.12) 0.242 (0.93) -0.503 (-1.82)* D1*P 0.819 (1.09) -1.188 (-1.57) 0.602 (0.85) -0.229 (-0.33) 0.549 (0.79) -1.957 (-3.10)*** D2*LC 0.014 (0.58) -0.005 (-0.20) 0.016 (0.70) 0.006 (0.25) -0.007 (-0.29) -0.024 (-1.03) D3*IO -0.071 (-1.15) 0.085 (1.72)* -0.009 (-0.21) -0.014 (-0.28) -0.024 (-0.47) 0.114 (1.94)* D4*S -0.026 (-0.79) -0.020 (-0.65) 0.004 (0.12) -0.024 (-0.76) -0.003 (-0.10) 0.026 (0.78) D5*GO -0.1802719 (-0.99) 0.169 (1.04) 0.011 (0.08) 0.051 (0.42) -0.224 (-1.76)* -0.022 (-0.14) D6*LP -1.226361 (-1.78)* 0.433 (0.63) -0.506 (-0.72) 0.352 (0.50) 0.038 (0.05) 2.359 (3.30)*** D7*D -0.698 (-2.07)** 0.506 (1.48) 0.443 (1.32) 0.208 (0.62) -0.197 (-0.59) 0.020 (0.06) obs 8740 8456 8317 8122 7959 7614 Pseudo R^2 0.1902 0.1735 0.1705 0.175 0.1814 0.1841

(24)

24 Table 5 continued 2009 (7) 2010 (8) 2011 (9) 2012 (10) 2013 (11) 2014 (12) Int -2.702 (-13.26)*** -1.578 (-5.97)*** -2.515 (-11.20)*** -2.263 (-10.22)*** -1.654 (-7.29)*** -2.140 (-9.86)*** P -0.189 (-0.52) 1.525 (2.90)*** 2.531 (5.31)*** 1.915 (3.54)*** 3.391 (5.62)*** 5.114 (7.66)*** LC 0.039 (2.73)*** 0.093 (4.21)*** 0.044 (2.98)*** 0.054 (3.18)*** 0.013 (1.34) 0.065 (3.59)*** IO 0.017 (0.35) -0.019 (-0.40) 0 (-0.00) -0.033 (-0.90) -0.069 (-1.74)* -0.151 (-4.23)*** S 0.354 (14.72)*** 0.343 (13.86)*** 0.354 (14.52)*** 0.363 (14.58)*** 0.328 (13.58)*** 0.345 (13.66)*** G -0.535 (-4.31)*** -1.583 (-7.06)*** -0.615 (-4.36)*** -0.691 (-4.91)*** -0.972 (-5.80)*** -0.416 (-3.29)*** LP 4.028 (8.06)*** 3.143 (6.29)*** 1.203 (2.86)*** 1.256 (2.52)** 1.659 (2.75)*** -0.009 (-0.02) Debt 0.423 (1.87)* 0.046 (0.18) 0.079 (0.31) -0.357 (-1.42) -0.168 (-0.69) -0.417 (-1.66)* D0 1.123 (3.37)*** -0.937 (-2.70)*** 0.252 (0.80) 0.609 (1.92)* -0.485 (-1.55) 0.214 (0.71) D1*P 1.714 (2.68)*** 1.006 (1.42) -0.616 (-0.85) 1.476 (1.82)* 1.722 (1.91)* -2.948 (-3.34)*** D2*LC 0.054 (2.06)** -0.049 (-1.83)* 0.010 (0.43) -0.041 (-2.08)** 0.052 (2.52)** -0.006 (-0.23) D3*IO -0.036 (-0.53) 0.019 (0.31) -0.033 (-0.63) -0.037 (-0.67) -0.082 (-1.53) -0.034 (-0.67) D4*S -0.011 (-0.31) 0.011 (0.32) 0.009 (0.25) -0.034 (-0.99) 0.016 (0.47) 0.002 (0.07) D5*GO -1.048 (-4.09)*** 0.968 (3.65)*** -0.076 (-0.38) -0.281 (-1.29) 0.556 (2.65)*** -0.062 (-0.36) D6*LP -0.884 (-1.25) -1.941 (-2.97)*** 0.053 (0.08) 0.403 (0.52) -1.668 (-2.04)** 3.019 (3.71)*** D7*Debt -0.377 (-1.11) 0.034 (0.09) -0.436 (-1.22) 0.189 (0.54) -0.249 (-0.71) -0.247 (-0.69) obs 7186 6958 6781 6607 6558 6647 Pseudo R^2 0.1864 0.1828 0.1739 0.1846 0.2072 0.2286

(25)

25

In this part, we will analyze the dividend trend. The main focus is on the variable d0 in table 5 because this measure indicates the difference in the payments of dividends after controlling for the independent variables. After the tax cut in 2003, a significant increase in the propensity to pay dividends can be observed (p<0.01). In the following two years, the probability of firms paying dividends decreased. However, the propensities to pay out dividends in 2004 and in 2005 are still significantly higher (p<0.05) than in 2002 (see appendix B). In 2006 and 2007, dividend payments among firms increased again. Although the increase in 2007 was insignificant, a decrease was expected because of the financial crisis. A possible reason is the timing of the crisis. In the summer of 2007, the US started experiencing a striking contraction in wealth and deterioration in credit market functioning (Reinhart & Rogoff, 2008). Firms which initiated dividends at the beginning of 2007 could have faced the consequences of the economic downturn at the end of the year. This could have resulted in a firm paying dividends at the beginning of the year while the financial data at the fiscal year end would expect otherwise, resulting in a higher propensity to pay dividends. In 2008, a significant decrease (p<0.10) can be found, confirming the expected effect of the crisis on dividends. In 2009, the probability of paying dividends increased significantly (p<0.01). The second year after the crisis, the tendency to pay dividends decreased again. While the dividend payments decreased in 2010 compared to 2009, it did not decrease compared to 2008 (see appendix B). From 2010 onwards, the propensity to pay out dividends has been increasing every year except in 2013. A possible explanation is the American Taxpayer Relief Act, which extended the tax cut of 2003. This act has been passed by the United States Congress on the first of January 2013. The tax cut of 2003 was going to expire in 2012. In order to benefit from the tax cut, firms had to initiate dividends before 2013. This resulted in a higher propensity to pay dividends in 2011 and even significantly higher in 2012 (p<0.10).

In table 5 and table 7, we observe results confirming previous research. After the tax cut in 2003, a higher tendency to pay out dividend is observed for the period 2003-2006 compared to 2002 (table 5 and appendix B). This corroborates the effect of the tax cut investigated by Chetty & Saez (2005). Furthermore, during the financial crisis, the propensity to pay out dividends decreased (table 7, column 1). At last, after the financial crisis, the probability to pay out dividends increased again (see table 7, column 4 and 5). These findings are similar to the empirical results of Floyd, Li & Skinner (2015) and Hauser (2013).

(26)

26

payout policies can be found. Although the tendency to pay dividends in the period 2003-2006 is higher compared to 2002, the increase in the probability to pay dividends is not cumulative. The propensity to pay dividends increased in 2003 but decreased in 2004 and 2005. A similar development is found for the period after the crisis, where the propensity to pay dividends increased significantly in 2009 (p<0.01) followed by a significant decrease (p<0.01) in 2010. These findings suggest that firms reacted immediately and significantly after a major event (the tax cut and the economic recovery) but the actions are not extended. Firms are not willing to keep paying dividends. According to Brav et al. (2005), managers have a strong desire to avoid dividend cuts. Instead of extending the significant increase in dividend initiations, managers tend to hold up to avoid the impression of becoming a consistent dividend paying firm. When a firm is not consistent in dividend payouts, dividend cuts are not relevant. The other holdup was in 2010 and significant at the 1% level. The period after the crisis includes an additional behavioral aspect. Managers are less eager to initiate dividends if company earnings over the long-term are uncertain (Lintner, 1956). Besides the avoidance of dividend cuts, the earnings uncertainty is an additional reason for the non-cumulative increase in dividend payments after the crisis.

To examine the changing determinants across periods, we have to evaluate the coefficients of the logit regression models. We start with table 6 where the significance of determinants in each period are tested. In table 7, the determinants will be compared in a direct regression model.

The independent variables profitability (P), life cycle (LC), size (S), growth rate (G), and lagged profitability (LP) are significant in all periods (see table 6). The determinant investment opportunities is significant in all periods except 2009-2010. The variable debt is the most interesting variable in table 6. This variable has a positive effect in 2007-2008 (p<0.05) and a negative effect in 2011-2014 (p<0.01). In previous papers, researchers found mixed results concerning debt. Some found a positive coefficient (Gill, Biger & Tibrewala, 2010) and some a negative one (Breuer, Rieger, & Soypak, 2014; Hobbs & Schneller, 2012). The results of table 6 and previous research suggest that the determinant debt has a different (dominant) effect on dividends across periods. During the financial crisis, the prospect on future earnings were lower. High amount of debt is only attainable by more capable firms as interest obligations are higher. In the most current period, high amount of debt is also attainable by less capable firms. In this case, debt shows the effect that we depicted before. As debt grows the interest

(27)

27

obligation increases, resulting in a higher risk of default. So, debt indicates the capabilities of firms in the financial crisis (positive); and in the most current period, debt indicates the default risk of a firm (negative).

Table 6 coefficients of determinants across periods

The reported values are the coefficients of profitability (P), investment opportunities (IO), asset growth (G), size (S), life cycle (LC), lagged profitability (LP), and debt (D) of the logit regression model across the periods 2003-2006 (period after the tax cut), 2007-2008 (financial crisis), 2009-2010 (recovery period), and 2011-2014 (the most current period).

2003-2006 (1) 2007-2008 (2) 2009-2010 (3) 2011-2014 (4) Int -2.433 (-26.95)*** -2.470 (-18.38)*** -2.195 (-13.09)*** -2.052 (-18.92)*** P 1.302 (5.19)*** 0.559 (1.80)* 2.061 (6.29)*** 2.962 (9.98)*** LC 0.058 (7.04)*** 0.050 (4.52)*** 0.064 (5.11)*** 0.041 (5.69)*** IO -0.077 (-4.53)*** -0.0452 (-1.70)* -0.006 (-0.19) -0.099 (-5.49)*** S 0.353 (31.61)*** 0.341 (20.55)*** 0.348 (20.11)*** 0.347 (27.98)*** G -0.341 (-7.15)*** -0.502 (-6.69)*** -0.917 (-7.54)*** -0.625 (-9.16)*** LP 1.567 (6.47)*** 2.938 (8.10)*** 1.973 (6.56)*** 1.482 (5.24)*** Debt 0.095 (0.80) 0.408 (2.52)** 0.064 (0.36) -0.350 (-2.83)*** obs 16578 7614 6958 13254 Pseudo R^2 0.1736 0.1819 0.1788 0.202

(28)

28

In table 7, we will compare the coefficients in a direct way to search for differences across the periods. The variables size and life cycle do not show any significant differences. The development of the determinants profitability and lagged profitability show that firms are becoming more risk averse. Profitability decreased significantly in 2007-2008 (p<0.10) while lagged profitability increased (p<0.01) (see table7, column 1). In the financial crisis, firms emphasized their dividend policy on results from the past instead of current results (see table 6, column 2). Even if a firm was profitable in the crisis, if the results in the past year were negative, the propensity to initiate dividends decreased. Another interesting observation is the significantly higher coefficient of profitability in 2011-2014 compared to 2003-2006 (see table, column 3). Nowadays, profitability is a more relevant determinant compared to the period before the crisis. If a firm is unprofitable, they are less likely to initiate dividends compared to the period before the crisis. All these empirical results indicate that the crisis caused firms to be more risk averse.

The development of the variable investment opportunities shows a signaling effect which depends on the market environment. The growth rate and investment opportunities decreased during the financial crisis and increased during the recovery period (see table 4). This result shows that firms are more likely to invest after the crisis. In addition, after the financial crisis, banks were heavily regulated. It was more difficult for firms to acquire loans. As a result, firms with a high level of investment opportunities had incentives to initiate dividends to appear healthy. When a firm appears healthy, it is easier to get access to external funds. In this way, firms could finance their investment opportunities. This behavior is confirmed by the empirical results in table 7. The statistics show that firms with a high level of investment opportunities were more likely to pay dividends in the recovery period than in the non-crisis periods (p<0.05) (see table 7, row 2 and 6). This finding could also explain the mixed statements of researchers concerning the signaling effects. If a researcher investigates the signaling effect of dividends over an extensive period, it is possible that significant results are not found, because the use of dividends as signaling device is affected by the market environment. Furthermore, the findings in table 7 (partially) refute the suggested contradicting evidence of DeAngelo et al. (2004), DeAngelo et al. (2006), and Denis & Osobov (2008). In the research of DeAngelo et al. (2004), the vast majority of dividends was paid by prominent corporations. This finding contradicts the signaling theory in the sense that bigger firms enjoy coverage by analysts and journalists, whose managers should have little need to

(29)

29

communicate with shareholders. However, we found a significant (p<0.10) increase in the coefficient of investment opportunities in the recovery period after controlling for size. So, in the recovery period, both bigger and smaller firms are more likely to pay dividends when the level of investment opportunities are high. Another point of critique was the finding of the empirical evidence for the life cycle theory (DeAngelo et al., 2006; Denis & Osobov, 2008). Firms with low earned/contributed equity would be ideal candidates for dividend signaling, because these firms are less mature and it is therefore more difficult to gauge their future prospects. Yet, these firms are not likely to pay dividends. Nevertheless, similar as size, a significant increase in the coefficient of investment opportunities in the recovery period is found after controlling for the life cycle variable. Young and mature firms are more likely to initiate dividends in the recovery period when investment opportunities are high. The development of the variable asset growth adds credibility to the signaling effects of dividend. During the financial crisis and recovery period, the coefficient of growth decreased significantly (p<0.10) (table 7, column 1 and 4). Firms with high level of asset growth were less likely to pay dividends in the financial crisis and recovery period. Growth rate is closely related to investment opportunities. The difference is that growth rate is realized and investment opportunities are not. Firms which are able to grow in the financial crisis and recovery period are not financially restricted. There is no incentive for these firms to appear healthy to get access to external funds. So, in the recovery period where investment opportunities are increasing, firms which are able to grow without restriction are more likely to use the cash for investments, instead of dividend initiations.

(30)

30 Table 7 difference in determinants across periods

The reported values without interaction variable D are the coefficients of profitability (P), investment opportunities (IO), asset growth (G), size (S), life cycle (LC), lagged profitability (LP), and debt (D) of the logit regression model for the first mentioned period. The values with the interaction variable D represent the difference in coefficients of the two periods.

* Statistical significance at 10% level, ** Statistical significance at 5% level, *** Statistical significance at 1% level 2003-2006 2007-2008 (1) 2003-2006 2009-2010 (2) 2003-2006 2011-2014 (3) 2007-2008 2009-2010 (4) 2007-2008 2011-2014 (5) 2009-2010 2011-2014 (6) Int -2.433 (-26.95)*** -2.433 (-26.95)*** -2.433 (-26.95)*** -2.470 (-18.38)*** -2.470 (-18.38)*** -2.195 (-13.09)*** P 1.302 (5.19)*** 1.302 (5.19)*** 1.302 (5.19)*** 0.559 (1.80)* 0.559 (1.80)* 2.061 (6.29)*** LC 0.058 (7.04)*** 0.058 (7.04)*** 0.058 (7.04)*** 0.050 (4.52)*** 0.050 (4.52)*** 0.064 (5.11)*** IO -0.077 (-4.53)*** -0.077 (-4.53)*** -0.077 (-4.53)*** -0.0452 (-1.70)* -0.0452 (-1.70)* -0.006 (-0.19) S 0.353 (31.61)*** 0.353 (31.61)*** 0.353 (31.61)*** 0.341 (20.55)*** 0.341 (20.55)*** 0.348 (20.11)*** G -0.341 (-7.15)*** -0.341 (-7.15)*** -0.341 (-7.15)*** -0.502 (-6.69)*** -0.502 (-6.69)*** -0.917 (-7.54)*** LP 1.567 (6.47)*** 1.567 (6.47)*** 1.567 (6.47)*** 2.938 (8.10)*** 2.938 (8.10)*** 1.973 (6.56)*** Debt 0.095 (0.80) 0.095 (0.80) 0.095 (0.80) 0.408 (2.52)** 0.408 (2.52)** 0.064 (0.36) D0 -0.037 (-0.23) 0.239 (1.25) 0.382 (2.70)*** 0.275 (1.28) 0.418 (2.42)** 0.143 (0.72) D1*P -0.743 (-1.86)* 0.760 (1.84)* 1.661 (4.27)*** 1.503 (3.33)*** 2.404 (5.60)*** 0.901 (2.04)** D2*LC -0.008 (-0.55) 0.007 (0.46) -0.017 (-1.54) 0.014 (0.86) -0.009 (-0.70) -0.024 (-1.63) D3*IO 0.032 (1.00) 0.071 (2.09)** -0.022 (-0.89) 0.040 (1.00) -0.054 (-1.67)* -0.093 (-2.70)*** D4*S -0.011 (-0.57) -0.005 (-0.25) -0.006 (-0.34) 0.006 (0.26) 0.006 (0.28) -0.001 (-0.03) D5*G -0.161 (-1.81)* -0.576 (-4.41)*** -0.284 (-3.41)*** -0.415 (-2.90)*** -0.123 (-1.22) 0.291 (2.09)** D6*LP 1.371 (3.14)*** 0.405 (1.05) -0.086 (-0.23) -0.965 (-2.05)** -1.456 (-3.17)*** -0.491 (-1.19) D7*Debt 0.312 (1.56) -0.031 (-0.15) -0.445 (-2.60)*** -0.344 (-1.43) -0.758 (-3.72)*** -0.414 (-1.91)* obs 24192 23536 29832 14572 20868 20212 Pseudo R^2 0.1763 0.1751 0.1891 0.1805 0.1966 0.1965

(31)

31 6. Robustness check

In this paper, several decisions regarding the data sample, methodology, and the regression model are made. In this section, empirical evidence will be provided to support and explain our choices.

Tabel 8 robustness checks

The reported values are the coefficients of profitability (P), investment opportunities (IO), size (S), asset growth (G), life cycle (LC), lagged profitability (LP), liquidity (L), debt (D), risk (R), and stock liquidity (SL) of the logit regression model for the period 2002-2014. The focus is on the significance of the determinants.

(1) (2) (3) (4) (5) Int -2.590 (-49.93)*** -2.417 (-45.20)*** -2.407 (-44.58)*** -2.156 (-39.51)*** -2.411 (-44.56)*** P 3.218 (33.81)*** 2.973 (28.58)*** 1.64 (11.93)*** 2.581 (24.44)*** 1.641 (11.92)*** IO -.064 (-6.80)*** -0.061 (-6.31)*** -0.065 (-6.70)*** 0.015 (1.49) -0.061 (-6.13)*** S 0.377 (64.60)*** 0.358 (59.77)*** 0.353 (58.81)*** 0.335 (55.32)*** 0.353 (54.73)*** G -0.449 (-14.69)*** -0.466 (-14.94)*** -0.475 (-14.83)*** -0.444 (-14.17)*** -0.479 (-14.87)*** LC 0.059 (13.18)*** 0.050 (11.51)*** 0.056 (12.83)*** 0.050 (11.39)*** LP 1.939 (14.35)*** 1.942 (14.30)*** 1.947 (14.36)*** L -1.396 (-20.48)*** D 0.018 (0.27) R SL obs 49813 49289 49098 48950 48896 Pseudo R^2 0.1771 0.1819 0.1855 0.1859 0.1862

(32)

32 Tabel 8 continued (6) (7) (8) (9) Int -3.093 (-74.06)*** -1.262 (-12.91)*** -1.705 (-16.98)*** -1.819 (-8.26)*** P 2.652 (24.91)*** 1.973 (7.57)*** 1.969 (7.49)*** 0.030 (0.15) IO -0.047 (-5.63)*** -0.176 (-10.47)*** -0.136 (-7.97)*** 0.185 (7.97)*** S 0.382 (61.51)*** 0.330 (36.79)*** 0.407 (39.79)*** 0.920 (18.84)*** G 0.000 (0.66) -0.861 (-13.27)*** -0.757 (-11.75)*** -.041 (-0.85) LC 0.001 (1.49) 0.137 (11.84)*** 0.128 (11.32)*** -0.007 (-1.13) LP 0.169 (2.55)** 2.421 (9.42)*** 2.621 (10.09)*** 1.564 (7.80)*** L D -0.155 (-2.45)** R -256.8423 (-13.84)*** -151.543 (-8.46)*** SL -2.045 (-17.08)*** obs 50585 22833 22833 14808 Pseudo R^2 0.1755 0.1973 0.2079 0.0554

* Statistical significance at 10% level, ** Statistical significance at 5% level, *** Statistical significance at 1% level

In table 8 (column 1, 2 and 3), the variables profitability, investment opportunities, size, growth rate, life cycle, and lagged profitability are significant from 0 at the 1% level. Profitability, investment opportunities, size, and growth rate as determinants of dividend

(33)

33

initiation are used in an extensive amount of papers (Breuer et al., 2014; Deangelo et al., 2006; Denis & Osobov, 2008; Fama and French, 2001; Fatemi & Bidlik, 2011; Gill, Biger & Tibrewala, 2010; Hoberg & Probhala, 2009; Kuo, Philip & Zhang, 2013). Life cycle variable and lagged profitability are used in the later papers (Breuer et al., 2014; Deangelo et al., 2006; Denis & Osobov, 2008; Kuo, Philip & Zhang, 2013). In column 5, debt is included as determinant of dividend payout. Although the variable is insignificant, several researchers (Breuer et al., 2014; Gill, Biger & Tibrewala, 2010; Grullon & Michaely, 2002; Hobbs & Schneller, 2012; Kuo, Philip & Zhang, 2013) used this variable in their models. Some of them found a positive coefficient and some found a negative one. In the results and discussion section we even found a significant positive coefficient in the financial crisis and a significant negative coefficient in the period 2011-2014. Because the effect of debt changes across periods, it is not unlikely to observe an insignificant coefficient if the regression is employed over the whole period. In column 4, the variable liquidity is added. We used cash and marketable securities over book value of assets as proxy for liquidity (Deangelo et al., 2006; Grullon & Michaely, 2002). Liquidity was expected to have a positive coefficient. If a firm is more liquid, it is less restricted by cash obligations, increasing the probability to pay dividends. However, the coefficient of liquidity is negative and significant at the 1% level. This result is also found by Deangelo et al. (2006). Because the empirical results contradicts with theory, we omitted the variable liquidity. In column 7 and 8, we added risk and stock liquidity as independent variables in our regression model. The proxy of risk is the variance of the stock return. This proxy is not used in previous papers. Hoberg & Prabhala (2009) and Kuo, Philip & Zhang (2013) used systematic and idiosyncratic risk as proxy for risk. To measure the risk of a specific firm, it is not necessary to distinguish between systematic or idiosyncratic risk. Therefore, we use one proxy that addresses the total risk of a firm. The proxy for stock liquidity is the average monthly trading volume divided by the amount of shares outstanding which is also used by Kuo, Philip & Zhang (2013). The coefficient of risk and stock liquidity are negative and significant. Nevertheless, the addition of the variable risk or stock liquidity distorts the external validity. By including the variables, 27752 observations are dropped. As a result, the average percentage dividend payers increased with 10% (see appendix c1). Last of all, distinction of industries could be made in the regression model. Some industries are more affected by an economic downturn than others. We neglected this distinction because we follow previous studies. Our methodology is based on previous papers and none of those used the industry distinction in

(34)

34

their regression models.

Moreover, instead of the robust standard errors, we used the normal standard errors. The robust standard error can often be used to address heteroscedasticity in a linear regression model. However, in non-linear regression models like a logit regression model, heteroscedasticity will lead to biased estimates. In this case, we would estimate the standard errors of inconsistent parameters. In addition, like the distinction of industries, the use of robust standard errors is not used in previous studies.

Furthermore, the decision was made to drop extreme outliers in our data sample. In column 6, the result of a regression is shown when the outliers are not omitted. The coefficients for growth and life cycle are not significant. However, in the literature, the support for these variables as determinant of dividends is consistent (Breuer et al., 2014; Deangelo et al., 2006; Denis & Osobov, 2008; Kuo, Philip & Zhang, 2013). When reviewing the summary statistics of growth and life cycle in appendix c2, we can observe outliers which are 100 standard deviations away from the mean. Extreme outliers can also be observed for profitability, investment opportunities, and lagged profitability. By omitting the extreme variables, more than 95% of the observations is still used. The average percentage dividend payers did not experience drastic changes. So, the external validity is not distorted by this decision while the coefficients are (more) significant.

At last, instead of a logit regression model, a fixed logit model could be used to account for potential omitted variable bias. In column 9, the statistics for the fixed logit model are shown. This model has several disadvantages compared to the logit regression model. In the first place, the use of a fixed logit model is not observed in previous researches while the use of a logit model is extensive (Deangelo et al., 2006; Denis & Osobov, 2008; Fama & French, 2001; Fatemi & Bildik, 2011; Hoberg & Probhala, 2009; Kuo, Philip & Zhang, 2013). Secondly, the variables profitability, growth, and life cycle are not significant while Breuer et al. (2014), Deangelo et al. (2006), Denis & Osobov (2008), and Kuo, Philip & Zhang (2013) found significant results. The last disadvantage is the loss of 35,777 (70.7%) observations. Observations with unchanged independent variables are dropped. So, firms which kept paying in the whole sample period or firms that did not pay in every period are omitted. This action will distort the external validity of the data sample.

The dividend trend and changing determinants across periods

Table of contents

List of figures

List of tables

Y

=

Y

=