Effects of dividend payout on stock returns during
and surrounding the financial crisis
Tjard de Wolf S1694286
Rijkuniversiteit Groningen, MSc Finance Supervisor: Dr. J.H. Von Eije
June 26th 2015
Word count: 12,150
Abstract:
Using returns from European firms listed in the STOXX 600 returns, it is found that the amount of dividends divided by the average amount of total assets, or dividend asset ratio, has a significant influence on future total returns in the period surrounding the financial crisis. This is in line with signaling and agency theories, which say that dividends stabilize firm returns because of the information dividends signals to investors regarding expected future returns and the avoidance of cash wasted on non-essential projects. No statistically significant results were found for the dividend asset ratio during the crisis year 2008, though its correlation coefficient with returns in 2008 was negative. A possible reason for this negative coefficient of the dividend asset ratio with respect to returns during the crisis lies in the
fact that firms had to write-off part of their assets due to the crisis, which has an impact on both total assets and market capitalization. These results are robust to control variables such as market-to-book ratios, total assets, the leverage ratio and to robustness checks through winsorisation and changes in crisis
years.
Keywords: Dividend policy, information asymmetry
1
1.
Introduction
The search for a firm’s optimal dividend payout structure remains a topic that is much discussed in literature. Many theoretical models have been created in order to account for the effects of dividend policy on stock returns. The irrelevance theorem of Miller & Modigliani (1961) states that in perfect markets with investors behaving rationally and with perfect certainty, the dividend policy of a firm has no effect on firm value. While this theorem would imply that paying dividends should have no effect on expected returns, other models -with different assumptions- show that dividend policy can increase returns and therefore (future) firm value. For imperfect markets, tax considerations (Black, 1976), waste of corporate funds (Jensen, 1976) and signaling properties of dividends (Bhattacharya, 1979) -among other reasons- show that there are factors that imply dividend policies may affect firm value. Empirical studies by Fama & French (1988) and Boudoukh, Michaely, Richardson and Roberts (2007) show a positive correlation between dividend yields and future firm value or returns, implying that in practice dividend policy is not irrelevant with respects to firm value.
Bhattacharya (1979) argues with his signaling theory that dividends signal investors about firms’ expected returns. Other economists, such as Fama & French (2001) and DeAngelo, DeAngelo & Stulz (2006) disagree with Bhattacharya (1979), and claim that paying dividends implies an impossibility to find interesting investment opportunities. This implies that when a firm pays out most of its earnings in dividends, it cannot strategically grow and cannot create more (future) firm value. This could lead to lower abnormal returns of these firms relative to other firms, which implies lower firm value (Hakansson, 1982). Jensen & Meckling (1976) and Jensen (1986) argue with this agency theory that dividends are paid out in order to avoid the waste of free cash flow by management on non-essential investments, thereby increasing shareholder value. Other reasons include the principles of risk aversion (Easterbrook, 1984) which deals with risk aversion by management and implies that this aversion destroys shareholder value. Theories in the field of behavioral finance include the theories of self-control and aversion of regret (Kahneman & Tversky, 1982; Shefrin & Statman, 1984) show that investors prefer receiving dividends to selling part of their equity for reasons of regret. If investors have to sell part of their equity in order to receive income and the share prices increase, they regret having sold part of their portfolio. Finally the catering theory (Baker & Wurgler, 2004) shows a relation between the demand for dividends in the market and the price of dividend-paying firms. When dividend demand is high, investors prefer dividend-paying firms and when demand is low, investors prefer non-dividend paying firms.
2 reverting markets, dividends give more stability which would relate to higher abnormal returns of dividend-paying firms. It would be interesting to see whether this “high dividend equals higher stock return” theory holds in the case of a crisis. While Blume (1980), Fama & French (1988), Bali (2003) and Boudoukh et al. (2007) have all found direct correlations between dividend yields and firm value, these studies were done in time periods after the “Great Depression” in 1929 and before the financial crisis of the twenty-first century in 2008. This last period was the strongest crisis since the Great Depression with equity index prices such as the S&P500 going down by as much as fifty per cent in a year, which makes it an interesting period to study. It would be interesting to see whether effects of dividends on returns are significant for the period during and surrounding the financial crisis. Furthermore, most research in the field has been done in the United States, as this is the largest public equity market in the world. Studying Europe, a completely different region, can add to an available pool of dividend research.
This raises the question: does dividend payout have an influence on stock returns in a period containing a crisis? And more importantly for this thesis, does dividend payout have an influence on stock returns during the recent financial crisis? These questions are the foundation for this thesis’ main research question:
Is there a significant positive relationship between the amount of paid dividends and total return for firms in Europe during the recent financial crisis?
3 negative correlation between dividends over assets and returns adds additional possibilities for asset management.
The results found in this thesis indicate that firms with high dividend asset ratios outperform firms with low dividend asset ratios in the period surrounding the financial crisis, which is in line with signaling theory (Bhattacharya, 1979) and the agency theory of Jensen (1986). However, it is also found that during the crisis year of 2008 the dividend asset ratio has a negative, but statistically insignificant, impact on stock returns. This suggests that the crisis year reactions to dividends may differ from non-crisis years. These results hold when control variables for risk, such as market-to-book ratios, the logarithm of total assets, a leverage ratio and a catering variable are included into the models. From the results of the control variables, it can be deduced that firm returns are negatively correlated to market-to-book ratios and to higher leverage ratios. These results are robust to one per cent winsorisation. The results suggest that while dividends over assets have a positive effect on returns in the period surrounding the crisis, the dividend asset ratio has a negative effect on returns during the crisis itself.
The remainder of this thesis is structured in the following way. The literature overview can be found in section two. Section three will illustrate data and methodology. Section four will present the results of this thesis and section five will show the conclusions and future research possibilities.
2.
Literature overview
There are many different theories regarding dividends and their ability to predict future returns. This thesis will discuss different theories, some of which fall in line with the neoclassical way of thinking and some of which are from a behavioral finance point of view. Neoclassical theories in the field of dividends are theories that are built around investor demand and supply and are about maximizing investors’ utility or profits. Neoclassical theories all assume investor rationality. This is different from behavioral finance theories, where decisions by investors are based on psychological, social and economic factors. In this part of the thesis, these two sides of literature are explained through different theories.
2.1 Neoclassical theories
4 Black (1976) wrote about the tax imperfection theory and discussed the so called “dividend puzzle”. He writes that dividends are a form of payment to investors by the firm. In the world of perfect markets without taxes or transaction costs as described by the Miller & Modigliani (1961) theorem, dividend policy would have no effect on firm value. In such a world, investors would be indifferent between a firm paying out dividends and a firm using these same funds for investments. However, in a world where dividends are taxed more heavily than capital gains and where capital gains are not taxed until they are realized, Black (1976) argues that a firm that does not pay dividends will be preferred over a similar firm that does pay dividends. This argument would imply dividend paying firms would have lower valuations and thus lower expected returns. Black (1976) further argues that in the presence of transaction costs, an investor who holds non-dividend paying stock will have to sell (part of) his investment or has to borrow against these shares in order to raise cash. If the investor had been paid dividends, this transaction would not have been necessary and therefore receiving dividends would have lowered transaction costs. This argument seems to be in favor of firms paying out dividends. Black (1976) further argues that it is also possible for the firm to buy back (part of) its own shares from investors willing to sell or for investors to sell on the open market, which would decrease transaction costs, thus making this argument somewhat trivial. There is little reason to think that this argument would change in a crisis.
5 to shareholders. In crisis times, it is expected that the effects of dividends would be even stronger. As market values of firms are getting lower and lower, firms paying out stable dividends would signal even more strongly to the market that nothing has changed about the expected cash flows of the firm, which implies that firm value and market price should remain constant.
A third neoclassical theory that goes against this theory of dividends signaling positive future returns is the idea of paying dividends due to low growth possibilities developed by -among others- Fama & French (2001) and DeAngelo et al. (2006). This idea states that firms with high profitability, but without interesting investment opportunities will pay out most of the excess cash as dividends. On the other hand, firms with low profitability and/or high growth tend to pay fewer dividends and retain profits. These findings imply both positive and negative impacts of the dividend asset ratio on expected returns. The negative aspect lies in the investment opportunities: if a firm pays out dividends, it has trouble finding attractive investment opportunities, implying this firm has maximized its growth potential (DeAngelo et al., 2006). The reason the dividend asset ratio would have a positive influence on stock returns is, in their eyes, due to the fact that companies paying out high dividends have high earnings and are therefore more stable than companies that have no dividend plans. The more stable the firm, the higher the demand of its shares will be by risk-averse investors. This idea implies that increases in dividends would signal to investors that the firm that the firm has fewer interesting investment opportunities, which would imply that dividends have a negative impact on future returns. In times of crisis however, an argument of the implications being reverted can be made. When firms have to downsize due to economic pressures, less and less interesting investment projects are being executed, as there is less financing in the market. This would imply that all firms have this expected reduction in investments, which is not due to dividend payments anymore. Any effect of dividends on investments would then be trivial and this argument would not hold during a period where investments are not made, such as during a crisis.
6 pay out dividends with the cash used for these investments. Jensen (1986) further argues that this waste of free cash flows serves the interests of management, but not the interests of shareholders. By implementing control mechanisms such as (internal) monitoring, but also by directly paying out dividends to shareholders, Jensen (1986) concludes that information asymmetry can be decreased and that corporate waste can be reduced. This implies that high dividend asset ratios would have a negative effect on information asymmetry and therefore a positive effect on the certainty of profits and therefore a positive effect on net returns. This theory holds even more during crisis times. When total investments are reduced, only the best projects should be financed. By paying out dividends, management has even less opportunity to waste available corporate funds on bad projects.
Easterbrook (1984) has extended the agency theory model and argues that there are two sources of agency costs, namely monitoring costs of agents and risk aversion by managers. Easterbrook (1984) argues that, for a publicly listed company, any one shareholder will not wish to incur all monitoring costs by himself, as he would only proportionally prosper from these costs. This argument can be trivialized by having an external or internal monitor appointed by some person or firm that represents the shareholders. However, these monitoring costs will be lower when dividends are paid, as this would reduce the risk of bankruptcy due to investments in bad projects. Therefore paying out high dividends instead of investing in risky projects could be beneficial to investors. This theory holds even more in the case of high investment uncertainty, during crises for example. Easterbrook (1984) his second source of agency costs lies with the fact that agents are self-interested in the sense that they benefit from both company stability and growth, as their personal compensation is based on these factors. This encourages risk-averse agents to take riskless instead of risky projects: lower risk means lower chances of losing their jobs or losing their personal wealth tied up in the firms’ stock. As shareholders are going for their own value maximization and thus prefer high-risk high-reward projects. This risk-aversion from management could create a difference in interests, which could increase information asymmetry. Managers would then pay out dividends, based on the signaling and agency theory, in order to decrease this information asymmetry. During a crisis, these effects could be even stronger. As firms have to choose their expenditures carefully, due to a lack of available cash, only the best projects can be taken. This would imply that with less cash, the same expected cash flows must be achieved in order to keep the market price at the same level. Management would therefore have to invest more heavily in high-risk high-reward projects
2.2 Behavioral theories
7 has first been developed by Kahneman & Tversky (1982) and has been extended with the theory of regret aversion by Shefrin & Statman (1984). A second clientele theory has been described by Allen, Bernardo & Welch (2000). The third and final behavioral theory is called the catering theory for dividends by Baker & Wurgler (2004).
A first behavioral argument for paying out dividends lies in prospect theory by Kahneman & Tversky (1982) and regret aversion by Shefrin & Statman (1984). Prospect theory is the psychological theory of decision-making for investors in the case of certain and uncertain losses or uncertain gains. The idea behind it is that investors who have suffered paper losses will often choose an unfair gamble in order to gain back the losses they have had, instead of taking their losses directly. An investor is faced with the choice of taking his loss now or taking a gamble that could either make good for the losses that were made or that would increase the losses that the investor has to incur. Shefrin & Statman (1984) used this theory in application to dividends and extended the model with the regret aversion theory. Regret aversion theory describes the feelings of investors with regards to selling part of their portfolio. One conclusion from the irrelevance theorem by Miller & Modigliani (1961) is that dividend policy are irrelevant, due to the fact that investors are able to sell part of their stock in order to create their own home-made dividend as there are no transaction costs and shares are easily tradable. Shefrin & Statman (1984) argue against this view, as selling part of a portfolio could lead to regret for the investor. In case the stock that was sold goes up, the investor would regret his choice of selling part of his stock as he would have had higher gains by keeping the stock and selling it later. Dividends would reduce this regret factor, as the investor would already have (part of) the necessary cash for his personal investment. Regret aversion would therefore be an argument in favor of paying out dividends. This theory is expected to hold in times of crises. As stock prices are going down, investors will want to receive dividends as this is a bird-in-hand payment. In other words, despite their portfolio decreasing in value, investors in firms with dividend payments will still receive their dividends. As receiving something is better than receiving nothing, it can be concluded that dividend-paying stocks hold lower regret levels for investors. This would imply that during the crisis, receiving more dividends will lower total investor regret, which implies that firms that pay out dividends during a crisis are more interesting for regret-averse investors.
8 pay out funds to their clients every period, receiving dividends helps lower transaction costs and satisfies this demand for periodic cash. The clientele theory shows an argument in favor of paying out high dividends, especially during crises. The investor would still receive their periodic payments in terms of dividends, which would help in paying their customers.
The last behavioral theory is the catering theory of dividends developed by Baker & Wurgler (2004) and is based on management incentives to pay out dividends to shareholders. The catering theory by Baker & Wurgler basically states that the amount of dividend paid out to investors is based on the demand for dividends by these same investors. This has been tested using dividend and non-dividend paying stocks: when dividend paying stocks are less in demand, investors implicitly ask for less or even no dividends, whereas if non-dividend paying stocks are less in demand investors ask for more dividends or dividend-payers. They further found evidence of a positive relationship between the rates of dividend initiation and the relative stock prices of dividend and non-dividend paying stocks. Dividends are omitted when non-dividend paying stocks are priced higher and dividends are increased or initiated when dividend-paying stocks are priced higher. In their view, the demand for dividend is mostly based on investor sentiment. Catering theory implies that increases and decreases in dividend asset are dependent on market demand for dividends. Baker & Wurgler (2004) have expressed this demand for dividends, or dividend premium, as the logarithm of the ratio of the average market-to-book ratios of dividend payers relative to non-dividend payers. The higher this premium, the more non-paying firms will initiate dividends. If the premium is negative, it implies that investor demand for dividends is very low and that paying out additional dividends should negatively influence stock returns. Catering theory suggests that this demand for dividends might influence future returns in either a negative or a positive way, depending on the demand for dividends of investors during the estimation period.
9 Hypotheses set 1: testing for significant effects of dividends on returns:
H1: The dividend asset ratio has a significant positive effect on stock returns in the period surrounding the financial crisis.
Hypotheses set 2: testing for significant effects of dividends on returns during the financial crisis:
H2: The dividend asset ratio has a significant positive effect on stock returns during the financial crisis.
The way in which these hypotheses will be tested will be further explained in the next data and methodology section.
3.
Data and methodology
This part of the thesis will explain the choice in data, its accessibility and origination, the research assumptions and finally the variables and methodology which are used to test the two hypotheses.
3.1 Data choice and accessibility
10 the period of 2004-2014 are unavailable have been removed from the sample. This leaves a sample of 281 firms divided over 12 countries in Europe. All relevant yearly data for these companies have been taken from the Datastream database (Thomson Reuters) and was constructed as panel data.
3.2 Research assumptions
Economists disagree about which forms of dividends should be included when testing for correlations. Dividends are a part of the profit of a company that is paid to shareholders. There are two different forms of firm payouts: dividends and stock repurchases. Studies by Gordon (1959) and Miller & Modigliani (1961) only considered dividends and not share repurchases for dividend policy. Nowadays another important aspect to firm payouts lies in share repurchases. Grullon & Michaely (2002) found that share repurchases by listed firms have increased explosively relative to cash dividends: in 2000 a total of 41.8 per cent of total earnings were used for share repurchase programs by U.S. corporations. In 1980 only 4.8 per cent of total earnings were used. A similar study by Von Eije & Megginson (2007) later also found the same trend for European stock markets. Studies on the “substitution hypothesis”, derived by Grullon & Michaely (2002) suggest that there should be little to no difference for a firm between paying dividends and repurchasing shares. Due to higher taxes on cash dividends John & Williams (1985) find that share repurchases are economically better for a firm’s shareholders than dividends. Research by DeAngelo, DeAngelo & Skinner (2000) and Jagannathan, Stephens & Weisbach (2000) further shows that while dividends are paid out of stable earnings, share repurchases are derived from more or less transitory earnings. Based on all these studies, Grullon & Michaely (2002) conclude that if all else is constant, share repurchases are a substitute for dividends. Fama & French (2001) also show that share repurchase programs have become a secondary method of payout for many firms over the last twenty years.
11 likely to be canceled as a company in need of cash will prefer cancelling repurchase programs than stop paying out dividends. However, as omitting dividends signals as a bad sign to the financial markets, omitting dividends generally brings about strong decreases in share prices, which is why management generally doesn’t cut back on dividends (Lintner, 1956; Kalay, 1980). When dividends have been announced these are paid by the firm at the announced date, giving cash and stock dividends more stability. Furthermore, despite the fact that repurchases are more relevant now than they were in the past, only looking at dividends is unlikely to bias the results (Fuller & Goldstein, 2011). It is for these reasons that this thesis will only focus on dividends and not on share repurchases. As dividends in Europe are most often paid out yearly and total assets are based on the yearly balance sheets, the choice has been made to look at yearly data.
Testing for dividend payout and its effects of returns can be done in various ways. While some studies focus on dividend yield (Blume, 1980; Fama & French, 1988; Bali, 2003 and Boudoukh et al. 2007), this variable does not focus on dividends relative to the book value of assets, but rather on the relative yield of dividend relative to market capitalization. While this does show the relative height of dividends, dividend payout can also be reflected by dividing dividends over total assets. This dividend asset ratio shows the payment of dividends relative to assets and is focused on book value of assets and not on the value that is reflected by the markets. The main independent variable will therefore be the dividend asset ratio, equal to the amount of dividends paid in year t divided by the average amount of assets of the firm in year t and year t-1.
Another assumption that has to be made lies in the determination of the crisis years. The financial markets peaked in October 2007 and hit their lowest point in February 2009. As the choice has been made to look at yearly data, the only real crisis year is 2008, as in 2007 the Stoxx 600 index still went up by about three per cent and in 2009 the index went up by about 25 per cent. A dummy variable has been created in order to account for this crisis year, where a value of zero is given in a non-crisis year and a value of one in the crisis year. In order to test for effects of dividends during the crisis, an independent variable is created where the dummy variable is multiplied by the dividend asset ratio for each year. In order to test for effects during the period surrounding the crisis, the period of 2004-2014 has been chosen, as the end of the crisis was in the beginning of 2009.
12 variable. The market capitalization ratio and leverage ratio were both used in Boudoukh et al. (2007) in order to control the test variables. They further used the logarithm of market capitalization, but that was because they were looking for changes in dividend yield, which is based on market capitalization, and not on total assets. As the dividend asset ratio is based on total assets using the logarithm of total assets as a control variable is more logical. The last control variable for the regression analysis is the catering variable. Adding this ratio will show time varying effects of investor dividend demand. The idea behind adding this variable is to test whether the main significance of the amount of dividends comes from catering theory or from signaling theory.
Finally Christie (1990) found evidence of u-shaped pattern in returns due to the nature of non-dividend paying stocks, a quadratic function of the non-dividend asset ratio is included in order to check for the form of the model.
3.3 Variables
In order to test whether the dividend asset ratio has a positive effect on returns, dependent, independent and control variables have to be created in order to do an OLS-regression analysis. This regression will show the correlation between the dependent variable and the various independent variables. In this part of the thesis these variables will be more discussed in detail.
3.3.1 Dependent variable: returns ( )
The dependent variable, returns, is the variable that is being tested for the regression, as the goal of this thesis is to show that returns are dependent on dividends. The returns are based on the Total Return Index for each firm. A return index shows a theoretical growth in value of a shareholding over a specified period, assuming that dividends are re-invested to purchase additional units of an equity at the closing price applicable on the ex-dividend date. For testing purposes, the dependent variable Ri,t for year t is
equal to the natural logarithm value of the total return index at year t divided by the total return index value at year t-1 for firm i. This variable is shown in equation 1.
13 3.3.2 Independent variables
Four different independent variables are used. While the main independent variable is the dividend asset ratio, there have also been other independent variables that need to be described: the square of the dividend asset ratio, the crisis dummy and the crisis dummy multiplied by the dividend asset ratio.
The dividend asset ratio is the amount of dividends paid by a firm i in year t over the average total assets of firm i in year t and t-1. The reason for dividing dividends over the average of total assets is because dividends are paid out over the course of the year and is different for each firm. For this thesis, the assumption is made that yearly dividends are paid in the middle of the year and that quarterly dividends are paid linearly over the year. The dividend asset ratio is the main independent variable of this thesis and will always be included into the regression models. Its expression is written in equation 2.
(2)
The squared dividend asset ratio is the square of the amount of dividends paid by a firm i in year t over the square of the average total assets of firm i in year t and t-1. It is included into the regression models, because u-shaped patterns have been found in other articles. This variable is shown in equation 3.
(3)
The dummy variable (DC) for the crisis is included in order to account for the years in which there was a crisis. In the years 2004-2007 and 2009-2014 this variable is equal to zero. For the crisis year 2008, the value of this variable is equal to one.
The dummy multiplied by the dividend asset ratio (DCDA) is used to test whether or not the amount of dividends has a significant effect on returns during the crisis. This variable simply shows the value of dividends over total assets for crisis years and zero for non-crisis years. Its expression is written in equation 4.
14 3.3.3 Control variables
Four different control variables have been included into the model in order to test whether or not the relation between the dependent and independent variables is valid.
The control variable MCBV is equal to the average market capitalization of a firm i in year t divided by the average book value of equity of a firm i in year t. Equation 5 depicts this variable.
(5)
The control variable TA is equal to the logarithm of the average of total assets of a firm i in year t. This expression is shown in equation 6.
(6)
The leverage ratio LEV is equal to the average level of debt of a firm i at year t divided by the average of total assets of firm i at year t as shown in equation 7.
(7)
15 These are all the variables that will be included into the model. The next methodology section will show how these variables are used to test the hypotheses and the summary statistics of these variables.
3.4 Methodology
In order to test for the influence of the dividend asset ratio on returns during and in the period surrounding the financial crisis, ordinary least squares (OLS) regression with t-tests will be used. In this section, the regression formulas, its summary statistics and the robustness tests will be explained. The basic and robustness models all have their own summary statistics that will be presented in this section.
3.4.1 The dividend asset ratio regression model
Regression is a way of testing whether there is a correlation between one dependent and one or multiple independent variables. In this case the regression analysis is done in three steps. The dependent variable, returns, and the control variables remain unchanged over these steps and are always used in the model. However, in every new step an additional independent variable is introduced. The first model includes only DA and DC as independent variables, in order to test directly whether the dividend asset ratio has an effect on returns. Its equation is shown in equation 9.
(9)
Where α is the intercept or constant of the regression line and is the standard error included for modeling purposes. The other variables are listed in section 3.3 of this thesis. The results of this equation will be presented in the first column of the result tables in the results section of this thesis.
In the second step, the squared dividend asset ratio is added to the model. This gives equation 10. These results will be presented in the second column of the result tables.
(10)
16 (11) Where α is the intercept or constant of the regression line and is the standard error included for modeling purposes. The other variables are listed in section 3.3 of this thesis. The summary statistics of all these variables for the dividend asset ratio regression model are listed in table 1.
Table 1: summary statistics of the dividend asset ratio regression model.
Table 1 shows the summary statistics for the dividend asset ratio regression model. The dependent variable of the model is firm returns per year (R). The independent variables are the dividend asset ratio (DA), the squared dividend asset ratio (DA2), the dummy variable for the crisis (DC) and the dummy variable for the crisis multiplied by the dividend asset ratio (DCDA). The control variables for the model are the market-to-book ratio (MCBV), the logarithm of total assets (TA), the amount of debt over
assets or leverage ratio (LEV) and the catering variable (CAT) equal to the ratio of the market-to-book ratio of dividend-paying firms and the market-to-book ratio of non-dividend-paying firms per year. The number of observations, the mean, median,
maximum, minimum, skewness and kurtosis are presented for each variable.
R DA DA2 DC DCDA MCBV TA LEV CAT
Observations 3029 3029 3029 3029 3029 3029 3029 3029 3029 Mean 0.113 0.027 0.001 0.087 0.003 7.467 15.910 0.243 1.364 Median 0.145 0.020 0.000 0.000 0.000 2.053 15.870 0.232 1.217 Maximum 1.157 0.214 0.046 1.000 0.200 3803.550 20.646 0.909 3.305 Minimum -1.208 0.000 0.000 0.000 0.000 -1872.963 10.835 0.000 0.718 Skewness -0.735 2.548 6.251 2.936 7.431 25.574 -0.027 0.473 2.026 Kurtosis 4.498 12.308 53.181 9.620 81.502 943.798 2.652 3.040 6.455
17 maximum, it seems likely that these maxima are outliers. The MCBV variable also has an interesting maximum and minimum, especially when compared to the mean. Both these low and high values are due to write-offs during the crisis, where the book value of equity decreased rapidly towards zero at the end of the year or even went below zero (insolvency). These outliers will also be taken out with winsorisation in the robustness checks.
3.4.2 Robustness test: winsorisation of the data
As can be seen from the summary statistics for the basic regression model in table 1, the maxima and minima of R, DA, DA2, DCDA, MCBV, TA and LEV are rather far apart. In order to check for the robustness of the model, it is possible to take away (some of) these outliers by winsorising the data at a one per cent level. Basically, for the variables R, MCBV and TA, data below the 0.5th percentile is set equal to the 0.5th percentile value and data above the 99.5th percentile is set equal to the 99.5th percentile value. This effectively removes one percent of the outliers that could have biased the results. As dividends and total assets are never negative, the minimum for the DA, DA2, DCDA and LEV variables will always be equal to zero. This is why, for those variables, data above the 99th percentile is set equal to the 99th percentile value, which results in the same one per cent of the outliers being removed. While the regression equations remain unchanged, the summary statistics change and are listed in table 2. The largest differences are in the maxima and minima. The results for these tests will be listed in a separate table in the results section.
18 Table 2: summary statistics after winsorisation for the dividend asset ratio model.
Table 2 shows the summary statistics for the winsorised dividend asset ratio regression model. The dependent variable of the model is firm returns per year (R). The independent variables are the dividend asset ratio (DA), the squared dividend asset ratio
(DA2), the dummy variable for the crisis (DC) and the dummy variable for the crisis multiplied by the dividend asset ratio (DCDA). The control variables for the model are the market-to-book ratio (MCBV), the logarithm of total assets (TA), the amount of debt over assets or leverage ratio (LEV) and the catering variable (CAT) equal to the ratio of the market-to-book ratio
of dividend-paying firms and the market-to-book ratio of non-dividend-paying firms per year.The number of observations, the mean, median, maximum, minimum, skewness and kurtosis are presented for each variable.
As regression fails when variables are multicollinear, correlation matrices for each of the three regression models are presented. Table 3, presented here, contains the correlation matrix for the dividend asset ratio model. Table 4 for the winsorised dividend asset ratio model is presented in the Appendix.
There are high correlations between variables that can be shown immediately. First of all, it can be seen that that the DA and DA2 variables are very highly correlated, with a correlation coefficient of 0.899. This is the reason why the three-step model has been set up. By testing independently for dividend effects with and without the squared dividend effects, this issue of multicollinearity is independently dealt with. A second high correlation is the 0.72 correlation between the DC and DCDA variable. This is however very logical, as the DCDA variable has nine out of ten values equal to the DC variable by its nature. These high correlations cannot be avoided by winsorising at a one per cent level.
R DA DA2 DC DCDA MCBV TA LEV CAT
19 Table 3: correlation matrix for the dividend asset ratio model
Table 3 shows the correlation matrix for the dividend asset ratio model. The dependent variable of the model is firm returns per year (R). The independent variables are the dividend asset ratio (DA), the squared dividend asset ratio (DA2), the dummy variable
for the crisis (DC) and the dummy variable for the crisis multiplied by the dividend asset ratio (DCDA). The control variables for the model are the market-to-book ratio (MCBV), the logarithm of total assets (TA), the amount of debt over assets or leverage
ratio (LEV) and the catering variable (CAT) equal to the ratio of the market-to-book ratio of dividend-paying firms and the market-to-book ratio of non-dividend-paying firms per year.
Dividend Asset Ratio R DA DA2 DC DCDA MCBV TA LEV CAT
R 1.000 DA 0.048 1.000 DA2 0.031 0.899 1.000 DC -0.021 0.028 0.009 1.000 DCDA -0.012 0.220 0.193 0.720 1.000 MCBV -0.027 -0.007 -0.004 0.034 0.011 1.000 TA -0.019 -0.055 -0.037 -0.001 -0.017 0.100 1.000 LEV -0.089 -0.083 -0.044 0.005 -0.020 -0.008 0.160 1.000 CAT -0.022 0.006 0.010 -0.008 -0.009 0.004 0.092 -0.009 1.000
The next section of this thesis will present the results of the OLS-regression equations and the other results that are found through empirical testing.
4.
Results
This section of the thesis shows the results of the regression equations presented in section three. The first part of section four deals with the dividend asset ratio model, whereas the second part deals with the winsorised dividend asset ratio model and other robustness tests. All OLS-regressions tests have been using Newey-West standard errors, in order to correct for heteroskedasticity and autocorrelation in the residuals of the models. Panel data has been used for testing purposes. No fixed or random time and cross-sectional effects were found in the sample. Additional robustness tests have been done by pooling or clustering the panel data.
20 results including DA2 but excluding the DCDA variable in the model and follows equation 10. The last step, depicted by (3) and (6), shows the full regression line from equation 11 and contains all variables. These steps are used in order to check the results independently. Table 5 presents all coefficients of the regression between independent and control variables to the independent variable. Beneath these coefficients, the values of the regression equations are presented in brackets. One star (*) indicates the t-test is significant at the five per cent significance level, two stars (**) indicate the t-t-test is significant at the one per cent significance level. For each of the three models inferences regarding the coefficients and the p-values will be made.
4.1 Dividend asset ratio model
The results for the dividend asset ratio model can be found in Table 5. First of all, it can be seen that the main independent variable DA has a positive coefficient with returns of 0.525 for the first step, 1.131 for the second step and 1.164 for the third and final step. With the DA variable being statistically significant in the whole model, it can be inferred that the dividend asset ratio has a positive and significant relationship with returns in the period surrounding the financial crisis. This is in line with theories of agency and signaling theory by Jensen (1986), and Bhattacharya (1979) respectively. The DA2 variable has negative coefficients with returns, but is statistically insignificant in all steps. It can therefore be inferred that the linear model is most correct and that the squared dividend asset ratio has no statistically significant effect on returns. The DC variable has a negative coefficient with returns, which implies that the returns in crisis years are negative. It is statistically insignificant however. The DCDA has a negative and statistically insignificant coefficient of minus 0.476 with returns. This implies that while during the crisis years the dividend asset ratio had a negative effect on returns, no inferences can be made regarding its statistical significance. It is interesting to find a negative sign however, as this would imply that a higher dividend asset ratio correlates with lower expected returns during a crisis, which goes against the stability signaling theory by Bhattacharya (1979) and the agency theory of Jensen (1986). The fact that the dividend asset ratio and returns are negatively correlated in the crisis years can be due to the fact that firms keep their dividends constant during the crisis, while returns are negative. Furthermore during the crisis firms had to write-off on financial instruments due to their fair value decreasing from liquidity issues, which would decrease both total assets thereby increasing the dividend asset ratio. This process would also decrease market value, which is the basis for the returns.
21 based on the book value which gives higher risks. This is also in line with the HML variable in the famous three factor model by Fama & French (1993). The (very) low correlation coefficient is due to the size issue with percentage of the ratio compared to returns. Based on the summary statistics, the market-to-book ratio is on average 70 times larger than the returns. The total assets control variable (TA) is statistically insignificant and has the same size issue as the market-to-book ratio. Its positive correlation coefficient implies that the higher the total amount of assets a firm has, the higher its returns are. The leverage ratio (LEV) has a highly significant negative effect on returns. This implies that, statistically, the higher the levels of debt the lower the returns of the firm. It can be seen that the catering variable (CAT) has a negative but statistically insignificant coefficient. This would imply that over the period of 2004-2014, catering demand or dividend demand did not affect returns. The adjusted R-squared or coefficient of determination of this model is low at 1.0 per cent. This implies that one per cent of the observed outcomes are replicated by the model.
4.2 Robustness check : winsorised dividend asset ratio model
22 Table 5: results for the (winsorised) dividend asset ratio model
Table 5 shows the results for the (winsorised) dividend asset ratio model. The dependent variable of the model is firm returns per year (R). The independent variables are the dividend asset ratio (DA), the squared dividend asset ratio (DA2), the dummy variable for the crisis (DC) and the dummy variable for the crisis multiplied by the dividend asset ratio (DCDA). The control variables for the model are the market-to-book ratio (MCBV), the logarithm of total assets (TA), the amount of debt over assets or leverage ratio (LEV) and the catering variable (CAT) equal to the ratio of the market-to-book ratio of dividend-paying firms and the market-to-book ratio of non-dividend-paying firms per year. The coefficients of the correlation analysis are without parentheses in the table. The t-values are shown within the parentheses. Columns (1), (2) & (3) show the results for the dividend asset ratio model whereas columns (4), (5) & (6) show the results for the winsorised dividend asset ratio model. Columns (1) and (4) use equation 9, columns
(2) and (5) use equation 10 and columns (3) and (6) use equation 11.One star (*) indicates the t-test is significant at
the five per cent significance level, two stars (**) indicate the t-test is significant at the one per cent significance
level.Table 5 presents all coefficients of the regression between independent and control variables to the
independent variable. Beneath these coefficients, the t-values of the regression equations are presented in brackets.
23 2004-2014, catering demand or dividend demand did not affect returns. The adjusted R-squared or coefficient of determination of this model is low at 1.0 per cent in step (3). This implies that only 1.0 per cent of the observed outcomes are replicated by the model.
There are some other checks that have been done on both regression models. As some of the results from the previously listed regression models differ –a lot– from findings in the literature with regards to dividend asset ratios and dividend yields having positive effects on future returns, multiple other robustness checks have been performed. None of these changes in the variables have resulted to higher statistical significances or –large– differences in the coefficients. The first choice was in the definition of the crisis years. For the regression results listed in table 5, there was only one crisis year, namely 2008. As the crisis started in 2007 and ended in 2009, all the tests were redone with the dummy crisis (DC) variable having values of one for 2007 and 2009 separately as well. This changed virtually nothing with regards to the sign or the height of the coefficients. The statistical significance of the main independent variables dropped below five per cent. The reason for this probably lies in the fact that on average the index went up by three per cent during 2007 and 25 per cent over 2009, while dividends remained somewhat stable year-on-year.
A second check has been done in the methodology. As panel data is used, there are multiple ways of structuring the data in order to test with OLS-regression. The method used in this thesis is with longitudinal or panel data for which no fixed or random effects for time-sectional or cross-sectional data have been found. Another test has been done by pooling the data. The assumption has to be made that all observations are from different samples of individual firms, which goes against the logic of the data. The results are however the same as with the panel data regression.
5.
Conclusions and final remarks
24 results of the control variables, it can be deduced that firm returns are negatively correlated to market-to-book ratios and to higher leverage ratios. These results are robust to one per cent winsorisation. The results suggest that while dividends over assets have a positive effect on returns in the period surrounding the crisis, the dividend asset ratio has a negative effect on returns during the crisis itself. Based on these results, it can be concluded that the first hypothesis of this thesis has been answered positively: there is enough evidence to infer that the dividend asset ratio has a positive correlation with returns in the period surrounding the financial crisis. There is however no statistical evidence to infer that during the crisis, the dividend asset ratio has the same effect on returns. Quite the opposite, as the data suggest that the dividend asset ratio has a negative impact on returns during the crisis.
It is concluded that investors are not indifferent to dividend policies. Instead, they value dividends most in positive periods of time and favor them less during crisis years, which goes against the findings of Fuller & Goldstein (2011), which show increasing demand for dividends in downward markets. While results over the whole period are consistent with theories of dividend signaling and agency theory, results during the crisis years are less consistent with these theories. An explanation for the negative coefficient of the dividend asset ratio with respect to returns during the crisis lies in possible write-offs on assets and subsequent decreases in market capitalization. With the assumption of constant dividends, this fact would increase the dividend asset ratio while decreasing the market returns, thereby giving a negative correlation.
25
6.
References
Allen, F., Bernardo, E., Welch, I. 2000. A Theory of Dividends Based on Tax Clienteles. The Journal of Finance 55 (6), 2499-2536.
Baker, M., Wurgler, J. 2004. A Catering Theory of Dividends. The Journal of Finance 59 (3), 1125-1165.
Bali, R. 2003. An empirical analysis of stock returns around dividend changes. Applied Economics 35, 51-61.
Bhattacharya, S. 1979. Imperfect Information, Dividend Policy, and "The Bird in the Hand" Fallacy. The Bell Journal of Economics 10 (1), 259-270.
Black, F. 1976. The Dividend Puzzle. The Journal of Portfolio Management 2 (2), 5-8.
Black, F., Scholes, M. 1974. The effects of dividend yield and dividend policy on common stock prices and returns. The Journal of Financial Economics 1, 1-22.
Blume, M.E. 1980. Stock returns and dividend yields: some more evidence. The Review of Economics and Statistics 62 (4), 567-577.
Boudoukh, J., Michaely, R., Richardson, M., Roberts, M.R. 2007. On the importance of measuring payout yield: implications of empirical asset pricing. The Journal of Finance 62 (2), 877-915.
Brennan, M., Thakor, A. 1990. Shareholder Preferences and Dividend Policy. The Journal of Finance, 45 (4), 993-1018.
Christie, W.G. 1990. Dividend yields and expected returns: The zero dividend puzzle. The Journal of Financial Economics 28, 95-125.
Cook, D.O., Krigman, L., & Leach, J.C. 2004. On the timing and execution of open market repurchases. The Review of Financial Studies 17 (2), 463-498.
DeAngelo, H., DeAngelo, L., Skinner, D. 2000. Special dividends and the evolution of dividend signaling. The Journal of Financial Economics 57, 309-354.
26 Easterbrook, F. 1984. Two agency-cost explanations of dividends. The American Economic Review 74 (4), 650-659.
Fama, E., French, K. 1988. Dividend yields and stock returns. The Journal of Financial Economics 22, 3-25.
Fama, E., French, K. 1993. Common risk factors in the returns on stocks and bonds. The Journal of Financial Economics 33, 3-56.
Fama, E., French, K. 2001. Disappearing dividends: changing firm characteristics or lower propensity to pay? The Journal of Financial Economics 60, 3-43.
Fuller, K., Goldstein, M. 2011. Do Dividends Matter more in Declining Markets? The Journal of Corporate Finance 17 (3), 457-473.
Gombola, M.J., Liu, F. 1993. Considering dividend stability in the relation between dividend yields and stock returns. The Journal of Financial Research 16 (2), 139-150.
Gordon, M. 1959. Dividends, Earnings, and Stock Prices. The Review of Economics and Statistics 41, (2), 99-105.
Grullon, G., Michaely, R. 2002. Dividends, Share Repurchases, and the Substitution Hypothesis. The Journal of Finance 57 (4), 1649-1684.
Hakansson, N. 1982. To pay or not to pay dividend. The Journal of Finance 37 (2), 415-428.
Jagannathan, M., Stephens, C.P., Weisbach, M.S. 2000. Financial flexibility and the choice between dividends and stock repurchases. The Journal of Financial Economics 57, 355-384.
Jensen, M.C., Meckling, W.H. 1976. Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. The Journal of Financial Economics 3 (4), 305-360.
Jensen, M.C.. 1986. Agency costs of free cash flow, Corporate Finance, and Takeovers. The American economic review 76 (2), 323-329.
Kahneman, D., Tversky, A. 1982. The psychology of preferences. Scientific American 246, 160-173.
27 Lintner, J. 1956. Distribution of Incomes of Corporations Among Dividends, Retained Earnings, and Taxes. The American Economic Review 46 (2), 97-113.
Miller, M.H., Modigliani, F. 1961. Dividend Policy, Growth and the Valuation of Shares. The Journal of Business 34 (4), 411-433.
Shefrin, H., Statman, M. 1984. Explaining investor preference for cash dividends. The Journal of Financial Economics 13, 253-282.
Vermaelen, T. 2005. Share repurchases. Foundations and trends in Finance (1), 3: 171-268.
28 APPENDIX:
Table 4: correlation matrix for the winsorised dividend asset ratio model
Table 4 shows the correlation matrix for the winsorised dividend asset ratio model. The dependent variable of the model are firm returns per year (R). The independent variables are the dividend asset ratio (DA), the squared dividend asset ratio (DA2), the dummy variable for the crisis (DC) and the dummy variable for the crisis multiplied
by the dividend asset ratio (DCDA). The control variables for the model are the market-to-book ratio (MCBV), the logarithm of total assets (TA), the amount of debt over assets or leverage ratio (LEV) and the catering variable (CAT) equal to the ratio of the market-to-book ratio of dividend-paying firms and the market-to-book ratio of
non-dividend-paying firms per year.
Wins. Dividend Asset R DA DA2 DC DCDA MCBV TA LEV CAT
