The information content of dividends:
empirical evidence in the short-,
intermediate- and long-run.
Master thesis
July 2015
University of Amsterdam, Amsterdam Business School
MSc Business Economics, Finance track
Author:
Jenk Lemmink
Student number:
10189823
2
Statement of Originality
This document is written by Jenk Lemmink 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 the contents.
3
Abstract
This paper shines new light on the ongoing debate about the information content of dividends by combining the signaling and the life cycle theory. The relation between changes in a firm’s dividend policy and future profitability in the short, intermediate and long term is tested. A new methodology is used, next to the traditional linear regression model, in order to correct for possible reversed causality and omitted variable bias. In the short term, we find that changes in a firm’s dividend policy are positively related to profitability, supporting the signaling theory. When the estimated time period is increased however, this relation becomes negative. These findings in the intermediate and long term provide empirical evidence for the life-cycle theory.
4 Content 1 Introduction……… 5 2 Literature……… 7 2.1 Theoretical framework………..7 2.2 Empirical evidence………...9 3 Methodology……… 14 3.1 Hypotheses.………... 14 3.2 Methodology..……… 15
3.3 Data and Descriptive Statistics………..19
4 Results………. 23 4.1 Initial Analysis………..23 4.2 Alternative Specification………...24 4.3 Robustness……….28 4.4 Limitations………. 30 5 Conclusion……….. 31 6 Bibliography……… 33 Appendix……… ……... 36
5
1 Introduction
The information content of dividends is one of the most important topics in corporate finance. It is well documented that dividend increases in general result in higher abnormal stock returns in the days following the dividend change announcement. This indicates that the stock markets expect earnings to increase after dividends have been increased, which is based on the signaling theory. The signaling theory hypothesizes that managers of firms with good prospects are expected to increase their dividends, while firms with an uncertain future are
more likely to decrease their dividend payouts. However,does future profitability indeed
increase after dividends have been increased, as stock markets and the signaling theory seem to predict?
Healy & Palepu (1988) and Nissim & Ziv (2001) show that there is a clear pattern of
earnings increase in the years after dividends have been increased. Benartzi, Michaely, & Thaler (1997) and Grullon, Michaely, Benartzi, & Thaler (2005), on the other hand, conclude that no relation exists between dividend changes and future profitability. Hence, there is an ongoing debate about the information content of dividends and the actual realization in future earnings.
A recent development related to this topic is the increased interest in understanding the effect of firm life cycles on dividends. This new life-cycle theory expects dividends to be paid by mature and established firms, reflecting a life cycle in which young firms have relatively
low cash and manyattractive investment opportunities, while mature firms in general have
generated more cash internally and have less attractive investment opportunities available.
These attractive investment opportunities, however, will drive long term profitability and are preferred to be financed internally. So, an increase in dividends could therefore also signal that a firm reaches their mature growth phase, where investments are reduced and long term profitability will be flattened.
The main objective of this paper is to combine the life-cycle and signaling theory in
order to shed new light on the debate about the information content of dividends. A standard linear regression model and a new two-stage-least-square instrumental variable approach will be used to test how changes in a firm’s dividend policy affect future profitability in the short (1-2 year period), intermediate (3-6 year period) and long run (7-10 year period).
Several papers, as mentioned above, have already investigated whether dividend
6
the signaling theory and only test the impact in the short and intermediate term. Another shortcoming of those papers is that the models used to estimate a potential causal effect, are more likely to suffer from reversed causality and omitted variable bias. A new two-stage-least-square instrumental variable approach is therefore used, next to the standard linear regression model, in order to correct for both biases. The instrument which is used is the introduction of the “Jobs and Growth Tax Relief Reconciliation Act of 2003” in the United States. The most important regulatory change of JGTRRA, related to this paper, is the
reduction of both the long-term capital gain rates as well as the tax rates on dividend income. JGTRRA is a dummy variable that has a value of 1 when the fiscal year ends after the
implementation of JGTRRA on 31 December, 2002. If the fiscal year ends before 31
December, 2002, the instrumental variable has a value of 0. The instrument seems to satisfy both conditions for being a valid instrument. It is relevant and exogenous, although
exogeneity of the instrument is more controversial due to the time-dimension in a before-after dummy variable.
The results are surprising compared to the findings in prior papers and provide
empirical evidence for both the signaling and the life-cycle theory. After correcting for the reversed causality and omitted variable bias, we find that profitability increases significantly in the year following a dividend increase. For dividend decreases, we observe a different pattern and conclude that profitability is significantly lower in the year after a dividend decrease. Both findings support the signaling theory and provide empirical evidence for the information content of dividends in the short run. The intermediate and long-term findings, on the other hand, indicate that profitability declines significantly in years 3 till 9 after a dividend increase. Next to this, we observe that profitability increases significantly in years 3 till 8 following a dividend decrease. This is in line with the life cycle theory, which hypothesizes that a negative relation exists between dividends and long-term profitability.
The rest of this paper is structured as follows. Section 2 gives an overview of the
present theoretical and empirical literature related to this topic. Section 3 describes the methodology and data. Section 4 presents the results of our analyses and robustness checks. This section also describes some limitations of the research. Finally, a conclusion is drawn in section 5.
7
2 Literature Review
This section gives an overview of the present literature on dividend policy. First, the theoretical framework related to dividend policy will be described. This is followed by an overview of the existing empirical literature that actually tested the relation between changes in a firm’s dividend policy and future profitability.
2.1 Theoretical Framework
Since the pioneering work by Miller & Modigliani (1961) there has been an ongoing debate about dividend policy. Prior to this paper, most economists believed that a firm becomes more valuable if it starts paying out higher dividends. This view is derived from the dividend discount model developed by Gordon (1959), which states that the value of the firm is equal to the present value of the dividends. Gordon (1959) assumes that investors are risk averse and applies the “bird-in-the-hand” argument to payout policy. Investors prefer certain cash dividends in their hands over the hope that risky reinvestments pay off in the form of capital gains. Miller & Modigliani (1961), on the other hand, argue that dividend policy is irrelevant under perfect and complete capital markets and therefore neither creates nor destroys value for the shareholders. If investment opportunities are kept constant over time, higher dividends result in lower investments and therefore in lower capital gains, keeping the total wealth for the shareholders the same. The value of a firm depends, in their opinion, only on the
productivity of the firm’s assets and not on the payout policy. Nevertheless, dividend irrelevance becomes more debatable considering market imperfections such as taxes,
information asymmetry, conflicts of interest between managers and shareholders, transaction costs and irrational investor behaviour.
Black (1976) introduces payout taxes into the frictionless M&M model where payout
policy is irrelevant for firm value. Investors are always better off under a policy of cash retention, which generates relative nontaxable capital gains rather than receiving a taxable cash dividend. This paper therefore finds no strong explanation as to why firms pay dividends and concludes that they should permanently avoid all payouts under infinite-horizon models. As a result, Black (1976) introduces the “dividend puzzle”. Authors have produced extensive and sometimes conflicting research through the years. However, there is still no clear answer to the question why firms pay dividends and the “dividend puzzle” is not completely solved
8
yet.
A possible explanation to the “dividend puzzle” could be the information content of
dividends, since there is an asymmetric information problem in firm valuation. Insiders have better information about a firm’s fundamental value compared to the outside investors. Dividends could therefore be used as a signal taken by the better informed insiders that provides credible information to the less informed investors. In his classical study, Lintner (1956), shows that a dividend-smoothing behaviour can be found in payout policy.
Furthermore, he reports that earnings are the key determinant of a firm’s payout policy and managers are strongly reluctant to cut regular dividends. Dividends are therefore only increased when management believes that earnings will permanently increase to a higher level. Those factors combined make dividends a credible signal provided by the informed party. Based on this, Bhattacharya (1979), John & Williams (1985) Miller & Rock (1985) conclude that mangers, as insiders, choose payout levels to signal their private information about earnings prospects to the uninformed investors. An increase in dividends typically signals that a firm will do better in the future, while a decrease normally suggests that it will do worse.
Most empirical studies provide evidence for the signaling theory mentioned above.
Those studies show that dividend increases result in higher abnormal stock returns in the days surrounding the dividend change announcement. Lang & Litzenberger (1989), on the other hand, show that companies with a market-to-book ratio larger than 1 do not experience such a significant increase in stock price in the days following a dividend increase announcements. This could imply that a dividend increase for high-growth companies is a signal for reduced investments.
A recent development in payout policy literature is therefore the increased interest in
understanding the effect of firm life cycles on dividends. This life cycle theory of dividends is based on the notion that as a firm matures, its ability to generate cash overtakes its ability to find profitable investment opportunities, which makes it optimal for the firm to distribute the cash to their shareholders in the form of dividend. The funding for this new payout theory is the pecking order theory, developed by Myers (1983). This widely accepted theory argues that firms prefer internal over external funding due to the asymmetric information problem.
Attractive investment opportunities are ideally financed with retained earnings, followed by debt and in the worst case scenario by the issuance of new equity. A young firm has relatively many attractive investment opportunities available, but is generally not profitable enough to
9
be able to meet all its financing needs through internally generated cash. The firm will
therefore minimize the dividend payments to their shareholders. Once the firm has reached its mature growth phase in its life cycle, attractive investment opportunities available are reduced and the firm has probably generated more cash internally than it could invest profitably. At this moment in time, the firm would increase its dividends. Cash and investment
opportunities available are the main determinants of a firms dividend policy according to the life-cycle theory. So, an increase in dividends could therefore also signal to the uninformed investors that a firm’s growth rate and profitability are expected to decline in the future. The reason for this is the reduced level of investments in attractive investment opportunities that will drive long term profitability growth.
2.2 Empirical Evidence
In the previous section we have seen that various dividend theories advocate the relation between dividends changes and future earnings. In addition, several empirical studied have observed that stock prices in general tend to go up when dividends are increased and fall when dividends are cut. However, less is known about the actual realization of future earnings. Do future earnings indeed increase after a dividend increase, as markets seem to expect?
Healy & Palepu (1988) are among the first researchers that have actually examined
this potential relationship between changes in a firm’s dividend policy and future earnings. The study takes only the most extreme changes in dividend policy into account. Their sample consists of 131 firms that start paying dividends for the first time and 172 firms that omit their dividend payments. First, they notice that firms experience significant earnings growth in the five years prior to a dividend initiation announcement . Similarly, they observe a significant earnings decrease in the two years prior to dividend omission announcements. Next to this, Healy & Palepu (1988) discover that dividend initiators experience a significant earnings increase in the two years after the dividend initiation, like the signaling theory predicts. On the other hand, firms that omit their dividends payments suffer an earnings decrease in the year in which the omission announcement takes place. However, earnings improve
significantly in the next several years. These results are the opposite of what the signaling theory predicts, but provide statistical evidence for M&M’s dividend irrelevance proposition and the life-cycle theory.
10
Brickley (1983) also shows that there is a positive relation between both
special/regular dividend increases and profitability in the year following the announcement of the dividend increase. The relation was more significant for regular dividend increases than for special dividend increases. Next to this, Aharony & Amihud (1994) find that quarterly dividend changes are positively related with future earnings for at least four quarters after the dividend change. Both papers are consistent with the signaling theory and provide evidence for the information content of dividends in the short run.
Unlike the previous studies, Benartzi, Michaely, & Thaler (1997) use a larger sample
and control for many other factors in order to increase accuracy of estimation. They have selected all US companies that were traded on the New York Stock Exchange and American Stock Exchange for at least two years during the sample period from 1979-1991. Dividend initiation and omission events are excluded from the sample, which is also in contrast to most prior research that only studied the extreme changes in payout policy. The resulting sample consists of 1025 firms and 7186 firm-year observations. Benartzi et al. (1997) test the influence of dividend changes on future earnings in two ways. The first is by doing a categorical analysis, in which they compare the earnings of firms that change dividends in a given year to those that did not. Industry trends are taken into account and they control for a possible earnings drift by correcting for previous earnings growth. The underlying assumption here is that earnings follow a random walk, so that the change in earnings measures
unexpected profitability. The second way of testing the potential causal effect, is by doing a regression analysis where future earnings within 2 years after the dividend change are regressed on a dividend change variable for year 0. They control for 27 firm characteristics that Ou & Penman (1989) find helpful in predicting earnings changes. The results show that there is a very strong lagged and contemporaneous correlation between dividend changes and earnings. However, in the two year period following a dividend increase, they were unable to find any statistical relation between changes in earnings and the sign and magnitude of the dividend change. For dividend decreases the results are stronger. Similar to Healy & Palepu (1988) they conclude that earnings significantly increase in the two years after a dividend cut.
Nissim & Ziv (2001) have reexamined this relation between dividend changes and future earnings using a different methodology and an alternative measure for future profitability. They provide strong empirical evidence that dividend changes are positively related to future earnings changes, future earnings and future abnormal earnings in the short run, like the signaling theory predicts. The sample selection criteria were similar to Benartzi
11
et al. (1997), while the sample was extended to the a period from 1963 till 1997. This results in a sample consisting of 100,666 observations: 811 dividend decreases, 13,221 dividend increases and 86,634 no-change observations. Most prior studies assumed that earnings follow a random walk with drift and measure unexpected profitability as the observed change in earnings minus the estimated drift. Using a similar approach, Nissim & Ziv (2001) find like most previous studies that dividend changes are in general not positively related to future profitability. However, they argue that those results are likely to be biased because of issues in
the estimation model. In their opinion, important control variables are omitted. Next to this,
there is a measurement error in the dependent variable that correlates with the dividend change variable. Prior research has also suggested that the relation between dividend changes and earnings is not symmetric for dividend increases and dividend decreases. Nissim & Ziv (2001) therefore measure the effect of dividend increases and dividend decreases separately using dummy variables. After modifying the regression model, they show that dividend changes are positively associated with earnings changes in each of the two years following a dividend change. Next to this, they have re-examined the relation between dividend changes and the level of future earnings and “abnormal earnings”. Those abnormal earnings are
defined as the difference between total earnings and the required earnings based on the cost of equity. The results show that dividend increases are positively related to both measures of earnings in each of the four subsequent years and provide evidence for the signaling theory. Dividend decreases, on the other hand, are not related to future profits under both measures. The insignificance of the coefficient for dividend decreases could be explained by the fact that dividend decreases represent only a small part of the sample.
Grullon et al., (2005) criticize the results obtained by Nissim & Ziv (2001). The
assumption of linear mean reversion in earnings is in their opinion inappropriate since they argue that the actual mean reversion process of earnings is more likely to be nonlinear. The reasoning behind this is that large changes in earnings revert faster than small changes, while negative changes revert faster than positive changes. Assuming linearity when the true functional form of unexpected earnings is nonlinear has the same implications as leaving out relevant independent variables. They show that after controlling for the nonlinear pattern in earnings, the significant relation between dividend increases and future earnings growth, which is found by Nissim & Ziv (2001), disappears.
Zhou & Ruland (2006) have done similar research and test the 1 year, 3 year and 5
12
listed on the New York Stock Exchange, American Stock Exchange and the NASDAQ in the period from 1950 through 2003. They furthermore require that companies have positive earnings in year 0 and a book value of equity greater than $250,000. Their sample includes 40,968 observations for the 1 year growth regression and decreases to 27,925 observations in the five-year growth regression. Similar to Nissim & Ziv (2001) they make use of a
multivariate linear regression. The final results also support the signaling theory and show that a strong, positive association exists between current dividend payouts and future earnings growth in each of the time intervals.
The only paper that investigates the relationship between changes in payout policy and
long-term future earnings growth is written by Arnott & Asness (2003). They focus on the growth of the aggregate market for the U.S. equity market portfolio however, defined as the S&P 500 Index. Unlike the life-cycle theory, the results strongly suggest that low payout ratios historically precede low future earnings growth. But does this relationship also exists at company level? Furthermore, we have to take into account that the model did not control for other factors, like the mean-reversion process in the distribution of earnings.
Grullon & Michaely (2004) have estimated the relation between share repurchase
programs, unexpected change in earnings and the level of investments within 3 years after the share repurchase. They were unable to find any evidence that share repurchase firms
experience unexpected profitability growth within the 3 years after the event. A decline in profitability is observed, instead of the earnings growth which is expected from theory. This implies that share repurchase programs are not a credible signal for future profitability. The results further demonstrate that repurchasing firms decrease their investments as hypothesized by the life-cycle theory. Free cash flows available therefore seem to be an important
determinant for the level of investment and earnings growth in the long-run. Share
repurchases are a one-time action and in that sense differ from dividends, but the results of this study provide some useful insides. Especially, since this paper tries to estimate the relation between changes in the dividend policy and earnings in the long-term.
13
3 Methodology
This section provides an overview of the methodology used in order to answer our research question. First, the testable hypotheses are stated. Second, the methodology is presented. Finally, the sample selection process and dataset are described.
3.1 Hypotheses.
The main objective of this paper is to combine the life-cycle and signaling theory and test how changes in a firm’s dividend policy affect future profitability in the short (1-2 year period), intermediate (3-6 years period) and long-run (7-10 years period), using a standard linear regression model and a new two-stage-least-square instrumental variable approach. The aim of this section is to discuss the hypotheses that have been tested.
Based on the signaling theory we hypothesize there is a positive (negative)
relationship between dividend increases (decreases) and profitability in the short term after a dividend policy change. An increase in dividends typically signals that a firm will do better in the near future, while a decrease suggests that it will do worse since managers are strongly reluctant to cut regular dividends. There is enough empirical evidence that supports this signaling theory. Healy & Palepu (1988), for example, observe that dividend initiators experience significant earnings growth in the two years after the initiation. Nissim & Ziv (2001) and Zhou & Ruland (2006) find similar results and both show that there is a strong, positive association between current dividend payouts and future earnings in the short and intermediate time interval.
On the other hand we hypothesize, that in the long run there is a negative (positive)
relationship between dividend increases (decreases) and profitability growth. If dividends are decreased, there is more cash available to invest in attractive investment opportunities. Since firms have a strong preference to finance attractive investment opportunities with internally generated cash, we expect firms with more cash available to have a higher level of
investments. Those investments will not pay off immediate, but it takes a longer period for the company to get the full benefits from it. So, an increase in dividends could therefore also signal that a firm reaches their mature growth phase, where investments are reduced and long term profitability will be flattened. This expectation is based on the life-cycle theory. There is lack of empirical studies that have actually tested the effect of dividend changes on the long
14
term profitability, at company level. However, Grullon & Michaely (2004) show that repurchasing firms significantly decrease their investments within 3 years after the share repurchase program. Free cash flows available therefore seem to be an important determinant for the level of investments and long-term profitability.
Hypothesis 1: There is a positive (negative) relationship between dividend increases
(decreases) and profitability in the short term after a dividend policy change.
Hypothesis 2: There is a negative (positive) relationship between dividend increases
(decreases) and profitability in the long term after a dividend policy change.
3.2 Methodology.
The multiple OLS linear regression model in equation (1) is the starting point of our analysis and will be used to test how changes in a firm’s dividend policy affect future profitability in the short, intermediate and long term. The approach in our initial analysis is similar to the regression analyses performed in prior papers, mentioned in section 2.2.
The change in profitability over the tested time period is the dependent variable in the
model. Return on equity is used here as our measure for profitability. The main independent
variable, ∆𝐷𝐼𝑉0 , reflects the dollar change in annual cash dividends per share paid to the
common shareholders. Cash dividends in the US are generally quarterly declared. We have therefore calculated the annual cash dividends per share by accumulating the quarterly dividends. Prior research shows that the relation between dividend changes and future earnings is not symmetric for dividend increases and dividend decreases. Because of this, we
have chosen to add two dummy variables into the model and interact them with ∆𝐷𝐼𝑉0 in
order to measure the effects of dividend increases and decreases separately. In this way, 𝛽1
measures the effect of dividend increases, while 𝛽2 estimates the effect of dividend decreases.
This approach of separating the effects for dividend increases and dividend decreases is similar to Nissim & Ziv (2001).
The model controls for several other factors that likely be related to the change in
profitability in the chosen time period. First of all, we control for the return on equity in the year prior to the dividend change announcement. Ou & Penman (1989) and Nissim & Ziv (2001) demonstrate that return on equity, which measures the ratio of earnings over the book
15
value of equity, is an important predictor for future earnings changes. Especially because of the fact that earnings tend to be mean reverting. When profitability is already high, companies will find difficulty having strong earnings growth in the future. We therefore expect this coefficient to be negative. Moreover, we control for past earnings growth. Healy & Palepu (1988) and Benartzi et al., (1997) present in their papers that future earnings growth and dividend changes are highly correlated with earnings changes in the years prior to the dividend change announcement. We control for this effect by including a variable that measures the change in return on equity in the studied time interval prior to the dividend change. Furthermore, we control for market to book ratio. High earnings growth is more likely for companies with high market to book ratio’s, since the market already expects high growth for those companies. We will also control for size, because the life cycle theory predicts that more established and mature firms are less likely to exhibit strong future
earnings growth. Book value of assets is used as the measurement for firm size. Finally,we
control forleverage and financial flexibility by adding the debt to equity ratio into the
regression. In each regression we correct for potential heteroskedasticity by using heteroskedastic-robust standard errors.
∆𝑅𝑂𝐸0,𝑡 = 𝛽0+ 𝛽1∗ 𝐷𝑃𝐶 ∗ ∆𝐷𝐼𝑉0+ 𝛽2∗ 𝐷𝑁𝐶 ∗ ∆𝐷𝐼𝑉0+ 𝛽3∗ 𝑅𝑂𝐸−1+ 𝛽4∗ ∆𝑅𝑂𝐸−𝑡,0+ 𝛽5∗𝑀 𝐵 0+ 𝛽6∗ 𝑇𝐴0+ 𝛽6∗ 𝐷 𝐸0+ 𝜖 (1) Where:
∆𝑅𝑂𝐸0,𝑡 = the change in return on equity over 1, 2, 5 and 10-year time periods. Return on
equity is measured as income before extraordinary items divided by the book value of equity. 𝐷𝑃𝐶 = a dummy variable that equals 1 if dividends are increased and 0 if not.
𝐷𝑁𝐶 = a dummy variable that equals 1 if dividends are decreased and 0 if not.
∆𝐷𝐼𝑉0= the dollar change in annual cash dividends per share paid to common shareholders at
time 0.
𝑅𝑂𝐸−1= the return on equity in the year prior to the dividend change. Return on equity is
measured as income before extraordinary items divided by the book value of equity.
∆𝑅𝑂𝐸−𝑡,0 = the change in return on equity in the 1, 2, 5 and 10-year timespan prior to the
dividend change. Return on equity is measured as income before extraordinary items divided by the book value of equity.
16
𝑀
𝐵 0= the firms market to book ratio in the year of the dividend change. The market to book
ratio is measured as the market value of equity (share price × common shares outstanding) divided by the book value of equity.
𝑇𝐴0 = the book value of total assets in the year of the dividend change.
𝐷
𝐸0= the debt to equity ratio in the year of the dividend change. The debt to equity ratio is
measured as long term debt divided by the book value of equity.
The models used in prior studies and in our initial analysis, however, are apparently suffering from reversed causality and omitted variable bias. Omitted variable bias implies that several other factors which influence future profitability are excluded from the model. This omitted variable bias can be addressed directly by including the omitted variables into the regression. Nevertheless, it is impossible to include all of them into the model. Reversed causality, on the other hand, indicates that causality runs both from dividend changes to future earnings, as well as from future earnings to dividend changes. Prior studies have shown empirically that this is likely to be the case. Healy & Palepu (1988) and Benartzi et al. (1997), for example, show that a very strong lagged and contemporaneous correlation exists between dividend changes and earnings. Both biases make the error term correlated with the repressors, which limits accuracy and reliability of the estimated results.
A two-stage-least square instrumental variable (IV) regression is a general way to obtain a consistent estimator when the regressor is correlated with the error term. This is therefore the solution to addressing the potential omitted variable and the reverse causality bias. A valid instrumental however, is difficult to obtain and must satisfy two conditions, known as the instrumental relevance and instrument exogeneity conditions.
The instrument used in this paper is a before-after dummy variable for the introduction
of the “Jobs and Growth Tax Relief Reconciliation Act of 2003” , that passed the United States Congress and was signed into law by president George W. Bush on May 26, 2003. The most important regulatory change of JGTRRA, related to this paper, is the reduction of both the long term capital gain tax rates as well as the tax rates on dividend income. For the first time in the history of the United States, long-run capital gains and dividend income were taxed at the same level. Individuals in the 25% or higher income tax bracket pay 15%, where those in the lower brackets do not have to pay taxes at all on dividend income and long term capital gains after JGTRRA. Prior to JGTRRA dividends were taxed heavier than capital
17
gains. The regulation applies in retrospect to the taxable years beginning after December 31, 2002. JGTRRA was designed to expire in 2012 however, president Obama extended the program. No expiration date is set yet. (Tax Act 2003 summary: jobs and growth Tax Relief
Reconciliation Act of 2003., 2003).
JGTRRA is a reasonable instrument since it seems to satisfy the two conditions for
being a valid instrument. The instrument is relevant because JGTRRA appears to be highly correlated with the change in annual cash dividends per share paid to common shareholders, which is the main independent variable in the original “biased” model. Dividend paying stocks become more attractive to investors since dividend taxes are reduced by the
introduction of JGTRRA. After the tax rate reduction on dividend income, it becomes cheaper for the firms to return money directly to the investors. Moreover, executives generally own a large stake in their own company. With the decreased tax rate they also have an incentive to increase the dividends in order to return part of the firm’s profits quickly to themselves. Those two arguments support a positive correlation between JGTRRA and a change in annual cash dividends per share paid to common shareholders. JGTRRA also seems to the be exogenous, although this is more controversial due to the time-dimension in a before-after dummy variable. Instrument exogeneity implies that the instrument is uncorrelated with the error term, so that there is no direct effect of the instrument on the dependent variable running through omitted variables. JGTRRA only introduced new regulation on personal income taxes. Corporate tax rate were not affected. This paper therefore assumes that there is no direct relationship between JGTRRA and the change in firm profitability, measured as return on equity.
In order to address the reversed causality and omitted variable bias, we will
re-estimate regression equation (1) using a two-stage-least-square instrumental variable
approach. In the first stage regression the endogenous variable ∆𝐷𝐼𝑉0 will be regressed on the
new JGTRRA-variable and similar control variables as in the initial analysis. ∆𝐷𝐼𝑉0 measures
the change in annual cash dividends per share paid to common shareholders. JGTRRA is a dummy variable that has a value of 1 when the fiscal year ends after the implementation of JGTRRA on 31 December, 2002. If the fiscal year ends before 31 December, 2002, the instrumental variable has a value of 0. The computed predicted values of the endogenous variable and the control variables from the first stage regression are then used in the second
stage regression. In the second stage regression is ∆𝑅𝑂𝐸0,𝑡 regressed on those predicted
two-stage-least-18
square IV-approach can be seen as reliable estimates, under the assumption that JGTRRA satisfies the exogeneity condition. They will be analyzed and compared to prior research and the findings from our initial analysis. Nevertheless, if it turns out that the instrument does not satisfy the exogeneity condition these results are likely to be biased. In both stages of the IV-regression is an intercept included and are the sum of squared residuals minimized using OLS. Furthermore, we have corrected for potential heteroskedasticity by using heteroskedastic-robust standard errors.
3.3 Data and Descriptive Statistics
A panel dataset consisting of accounting and market value data for listed firms on the American Stock Exchange, New York Stock Exchange and the Nasdaq is requested from Compustat. The sample consist of observations in the sample period starting at fiscal year 1980 and ending at fiscal year 2013. This large sample-period is chosen in order to increase accuracy of estimation. Firms with less than 3 observations in the sample period are excluded. Cash dividends in the US are generally quarterly declared. Quarterly data is therefore
requested and we have calculated the annual cash dividends per share by accumulating those quarterly dividends. Both active, as well as inactive firms are included in order to limit “survivorship bias”. An important addition of this paper in comparison to the studies performed by Benartzi et al. (1997), Nissim & Ziv (2001) and Grullon et al., (2005), is the fact that this paper includes both dividend initiations and omissions into the sample. This limits calculating the percentage change for the dollar change in dividends per share paid. However, results will be biased if the most extreme changes in dividend policy are not included in the sample because they seem to have a significant influence on future earnings growth. This is supported by the empirical results obtained by Healy & Palepu (1988).
Table 1 presents the firm-year observations for each year and for the total sample. The
sample selection criteria, mentioned above, resulted in a sample of 119,077 dividend observations for 12,424 firms listed on the American Stock Exchange, New York Stock Exchange and Nasdaq. The sample consists of 35,435 dividend increases, 13,586 dividend decreases and 70,056 no-change observations. Dividends are far more frequently increased or not changed than they are decreased. This is in line with the existence of dividend-smoothing behaviour in payout policy as explained by Lintner (1956). He reports that earnings are the
19
key determinant of a firm’s payout policy and managers are strongly reluctant to cut regular dividends.
Table 2 provides an overview of the descriptive statistics on dividend changes, market to book ratio, return on equity, debt-to-equity ratio and total assets for the total sample and the subsamples of dividend increases, dividend decreases and no changes in the dividend policy. We observe that dividend increases are generally smaller in magnitude than dividend
decreases. The mean and median values for dividends increases are equal to respectively $0.146 and $0.070, while those values for dividend decreases are -$0.325 and -$0.187. Furthermore, the rest of Table 2 shows the pattern like we have expected based on the life-cycle theory. Firms that increase their dividend payouts are in general larger and more profitable than firms who cut their dividend payouts or remain them unchanged. The mean and median values for return on equity and total assets are higher for the firms who increase their dividends in comparison to the firms who don’t. Finally, we observe that firms who cut their dividends generally have higher debt-to-equity ratio’s. Financial flexibility therefore seems to have a key impact on a firm’s dividend policy. This could be due to credit risk. Another reason for this could be the pecking order theory developed by Myers (1983). This theory hypothesizes that firms prefer internal over external funding due to the asymmetric information problem. Attractive investment opportunities are preferred to be financed with retained earnings, followed by debt and in the worst case scenario by the issuance of new equity.
20
Table 1: Sample
This table reports the number of observations for dividend increases, dividend decreases, no change in dividends per share and the total observations for each year and the complete sample. Our total sample consist 119,939 dividend observations of 12,424 firms listed on the American Stock Exchange, New York Stock Exchange and Nasdaq in the period 1980-2013.
Fiscal Year Div. Inc. Div Dec. No change Total Fiscal Year Div. Inc. Div Dec. No change Total 1981 1,069 530 780 2,377 1999 1,338 384 2,587 4,309 1982 1,149 375 953 2,477 2000 1,095 339 2,899 4,333 1983 805 598 1,097 2,500 2001 962 393 2,997 4,352 1984 1,021 376 1,260 2,657 2002 940 400 2,917 4,257 1985 920 419 1,334 2,673 2003 1,185 305 2,818 4,308 1986 752 513 1,383 2,648 2004 1,218 306 2,646 4,170 1987 861 482 1,441 2,784 2005 1,274 282 2,566 4,122 1988 1,075 279 1,475 2,829 2006 1,245 295 2,507 4,047 1989 1,061 335 1,428 2,824 2007 1,080 374 2,438 3,892 1990 1,040 331 1,503 2,874 2008 1,147 327 2,349 3,823 1991 858 443 1,659 2,960 2009 727 738 2,461 3,926 1992 828 516 1,849 3,193 2010 810 499 2,516 3,825 1993 932 442 2,108 3,482 2011 1,059 293 2,467 3,819 1994 1,311 399 2,435 4,145 2012 1,261 279 2,207 3,747 1995 1,475 336 2,532 4,343 2013 1,072 517 2,104 3,693 1996 1,361 417 2,746 4,524 1997 1,202 517 2,931 4,650 Total 35,435 13,586 70,056 119,077 1998 1,302 547 2,663 4,512
21
Table 2 : Descriptive Statistics.
This table reports the descriptive statistics of dividend changes and other firm characteristics that are used as control variables. The descriptive statistics are reported for the total sample and for the subsamples of dividend increases, dividend decreases and no changes in dividend policy. Dividend change is measured as the dollar change in annual cash dividends per share. M/B ratio is the market value of equity scaled by the book value of equity. Return on equity is measured as the amount of income before extraordinary items returned as a percentage of total equity. D/E ratio the book value of long term debt scaled by the book value of equity. Finally, Total Assets report the book value of assets measured in millions of dollars.
Mean SD 10% 25% 50% 75% 90% A. Dividend Increases (N= 35,435) Dividend change 0.146 0.258 0.015 0.037 0.070 0.149 0.300 M/B ratio 2.125 1.650 0.848 1.164 1.661 2.492 3.856 Return on Equity 12.752% 11.008% 4.074% 8.502% 12.504% 16.889% 23.104% D/E ratio 0.685 1.215 0 0.120 0.411 0.877 1.464 Total Assets (millions of $) 4877.827 10524.55 88.124 298.033 1069.121 3871.44 12956.24 B. Dividend Decreases (N=13,586) Dividend change -0.325 0.392 -0.825 -0.425 -0.187 -0.07 -0.025 M/B ratio 2.080 1.983 0.581 0.934 1.529 2.515 4.051 Return on Equity 5.710% 23.386% -13.118% 2.263% 10.554% 16.231% 22.464% D/E ratio 0.830 2.815 0 0.113 0.415 0.899 1.741 Total Assets (millions of $) 4075.579 9324.619 56.399 186.205 749.128 3094.976 10958.9 C. No Change (N= 70,056) Dividend change 0 0 0 0 0 0 0 M/B ratio 2.469 2.381 0.699 1.067 1.723 2.956 5.027 Return on Equity -1.344% 29.415% -33.023% -4.479% 6.572% 13.154% 20.173% D/E ratio 0.608 2.148 0 0.003 0.201 0.679 1.429 Total Assets (millions of $) 1325.962 4565.132 21.015 53.192 183.319 733.791 2566.347 D. All Events (N= 119,077) Dividend change 0.006 0.236 -0.030 0 0 0.020 0.109 M/B ratio 2.395 2.267 0.724 1.089 1.701 2.833 4.766 Return on Equity 3.625% 25.610% -19.343% 1.494% 9.441% 15.079% 21.691% D/E ratio 0.647 1.972 0 0.018 0.279 0.766 1.467 Total Assets (millions of $) 2337.291 7374.365 26.79 79.107 330.767 1468.989 5536.03
22
4 Results
This section provides an overview of the results from our statistical analyses. First, the results from our initial analysis are presented. Second, we will illustrate the results obtained by re-estimating the model using an instrumental variable approach, in order to correct for possible reversed causality and omitted variable bias. This is followed by a section that covers several robustness checks. Finally, some limitations of this research will be discussed.
4.1 Initial Analysis.
Using a standard linear regression model for testing the relation between dividend changes and future profitability, we find unexpected results in comparison to prior empirical studies that have tested the relation using similar models. The results from our initial analysis are
presented in table 3and indicate that profitability will be significantly higher in each of the
two subsequent years after a dividend increase. On the other hand, profitability will be significantly lower in each of the four years following a dividend decrease. These findings strongly support hypothesis 1 and are consistent with the signaling theory. They provide evidence for the information content of dividends in the short-run. Prior empirical studies, however, have found contradicting results that partly support this signaling theory. Nissim & Ziv (2001), for example, argue that dividend increases are positively related to profits in each of the four subsequent years, while dividend decreases were not related to future profits at all. Benartzi et al. (1997), on the other hand, find that earnings changes are essentially unrelated to the sign and magnitude of the dividend change in the two years following a dividend increase. For dividend decreases, their results are stronger and show a clear pattern of earnings increases in the two years following a dividend cut.
The results from the initial analysis further demonstrate a negative and weakly
significant relation between dividend increases and profitability from year 5 up to and
including year 8, after a dividend increase announcement. For dividend decreases we observe
a similar pattern. The coefficient DNC * ∆DIV0 is also negative and weakly significant in
years 6, 7 and 8 after dividend payouts have been decreased. These results partly support hypothesis 2, which is based on the life-cycle theory. Profitability in the intermediate/long-term seems to decrease after an increase in dividend payouts as the life-cycle theory predicts. However, profitability is also decreasing when dividends are decreased which is the opposite
23
of what the life-cycle theory hypothesizes. The only paper that investigates the relationship between changes in payout policy and long-term future earnings growth is performed by Arnott & Asness (2003). Their results are comparable to our results for dividend decreases and strongly suggest that low payout ratio’s historically proceed low future earnings growth. Our results from the initial analysis indicate that this relationship also exists at company level.
4.2 Alternative Specification.
Nonetheless, the models used in prior studies and in our initial analysis seem to be suffering from reversed causality and omitted variable bias. Reversed causality implies that causality runs both from dividend changes to future earnings growth, as well as from future earnings growth to dividend changes. Prior studies have shown empirically that this is likely to be the case. Healy & Palepu (1988) and Benartzi et al., (1997), for example, show in their papers that future earnings growth and dividend changes are highly correlated with earnings changes in the years prior to the dividend change announcement. Omitted variable bias implies that several other factors that influence future profitability growth are excluded from the model. Again, this is expected to be the case. Both biases make the error term correlated with the repressors, which limits accuracy and reliability of the estimated results. We use an instrumental variable approach to address these issues. After correcting for the reversed causality and omitted variable bias by running a two-stage-least-stage IV regression with the introduction of JGTRRA as our instrument, we observe some surprising results. These results are presented in table 4.
First of all, we find that profitability will be significantly higher in the year following a
dividend increase. For a dividend decreases, we find the opposite and conclude that
profitability will be significantly lower in the year after the dividend decrease. In year 2 we observe a similar negative relation for dividend decreases, although this coefficient is not statistically significant at a 10% confidence level. Both findings support hypothesis 1 and are consistent with the signaling theory. They provide evidence for the information content of dividends in the short-run. In our initial analysis, we also find significant coefficients that are in line with the signaling theory. So, in the short run both estimation methods leads to the same conclusion. The coefficients obtained, using the IV-approach, are nonetheless only significant for year 1.
24
The outcomes of estimating the model using the IV-approach are starting to differ
from our initial analysis when the estimated time period is increased. The intermediate and long term results indicate that profitability significantly declines in the 3 till 9-year time interval after a dividend increase. On the other hand, profitability will be significantly higher in years 3 till 8 following a dividend decrease. Both findings support hypothesis 2, which is based on the life-cycle theory. Profitability growth is increasing in the intermediate and long-term after a dividend decrease, while it is decreasing after a dividend increase. Prior studies failed to uncover this relation. Arnott & Asness (2003) is the only empirical study that has investigated the relationship between changes in payout policy and earnings growth in the long-run. They argue that low payout ratios historically proceed in low future earnings growth at a market level. Our results, using the IV-approach, are indicating the opposite. We therefore conclude that this relationship does not exists at company level. Benartzi et al. (1997), Nissim & Ziv (2001), Brickley (1983) and Aharony & Amihud (1994), on the other hand, do not provide empirical evidence for the life-cycle theory in the intermediate term. The only paper that find similar empirical results in the intermediate term is presented by Healy & Palepu (1988). They show that earnings decline in the first year and improve significantly in the next several years after a dividend omission. The sample selection criteria could be a reason for this difference in results. Healy & Palepu (1988) were only interested in dividend omissions and initiations, while the other papers mentioned above excluded those extreme changes in payout policy in their dataset. This paper studies all dividend changes and therefore also includes dividend omissions and imitations. Hence, the data selection criteria and difference in methodology could be a reason for the different results obtained compared to most prior research in the intermediate and long-term.
Additionally, we find similar to Ou & Penman (1989) and Nissim & Ziv (2001) that
return on equity and past change in return on equity are important predictors for future
profitability growth. The coefficients ROE-1 and ∆ROEt,0 are negative and strongly
significant. This indicates that earnings tend to be mean reverting. When profitability is already high, companies will find difficulty having strong earnings growth in the future.
Furthermore, we observe that the introduction of JGTRRA in 2003 is a strong
instrument. When the first stage F-statistic is smaller than 10, it indicates that the instrument is weak. This would imply that the TSLS estimator is biased, which make the results
unreliable. This first-stage F statistic is decreasing when the time interval in increased. However, in all cases it is much above the critical value of 10.
interval. Those time intervals are 1 till 10 years after the change in dividend policy. The change in return on equity over the chosen time period is our measurement for profitability change. It is regressed on variables for dividend increases and dividend decreases plus several control
variables. Robust t-statistics are reported in the parentheses.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) ∆ROE0,1 ∆ROE0,2 ∆ROE0,3 ∆ROE0,4 ∆ROE0,5 ∆ROE0,6 ∆ROE0,7 ∆ROE0,8 ∆ROE0,9 ∆ROE0,10
Constant -1.398*** -1.556*** -1.661*** -1.208*** -1.332*** -1.405*** -1.585*** -1.606*** -1.345*** -1.596*** (-10.61) (-8.67) (-8.29) (-5.59) (-5.53) (-5.27) (-5.42) (-5.35) (-3.76) (-4.14) DPC * ∆DIV0 5.110*** 1.427** -0.666 -1.332 -1.710** -2.502** -2.751** -2.353* -1.846 -0.794 (14.55) (2.41) (-1.01) (-1.54) (-2.03) (-2.48) (-2.47) (-1.70) (-1.53) (-0.55) DNC * ∆DIV0 -3.704*** -1.846*** -1.055** -1.809*** -0.574 -1.514** -2.213** -1.907** -1.792 -1.338 (-10.16) (-4.03) (-2.10) (-3.53) (-0.90) (-2.13) (-2.56) (-2.11) (-1.50) (-1.02) M/B0 0.763*** 0.392*** 0.274*** 0.164** 0.249*** 0.143 0.207** 0.153 -0.145 0.00723 (17.12) (6.48) (3.94) (2.10) (2.94) (1.48) (1.99) (1.45) (-1.06) (0.06) ROE-1 -0.391*** -0.178*** -0.141*** -0.137*** -0.136*** -0.128*** -0.132*** -0.114*** -0.0884*** -0.106*** (-55.15) (-25.90) (-18.69) (-15.63) (-13.00) (-10.83) (-10.12) (-7.32) (-5.22) (-5.49) ∆ROEt,0 -0.553*** -0.434*** -0.464*** -0.454*** -0.465*** -0.438*** -0.450*** -0.477*** -0.471*** -0.449*** (-71.35) (-56.37) (-53.49) (-49.34) (-45.11) (-37.71) (-34.15) (-31.42) (-26.51) (-22.21) TA0 0.00009*** 0.00002** 0.00001 0.000002 0.000009 0.00002 0.00003* 0.00002 0.00004* 0.00005* (13.95) (2.56) (1.21) (0.19) (0.71) (1.14) (1.83) (0.82) (1.88) (1.92) D/E0 0.203** 0.784*** 0.983*** 1.083*** 1.088*** 1.077*** 0.988*** 0.964*** 1.379*** 1.465*** (2.00) (4.92) (5.93) (7.06) (6.64) (6.24) (5.28) (5.61) (6.89) (5.72) Adjusted-R2 0.230 0.200 0.224 0.226 0.235 0.212 0.220 0.234 0.228 0.218 Robust Std. Errors X X X X X X X X X X N 101691 80785 65418 53032 43185 35031 27851 21872 16817 12831 Time Interval (Years) 1 Y 2 Y 3 Y 4 Y 5 Y 6 Y 7 Y 8 Y 9 Y 10 Y Standard errors in parentheses *p<0.10, ** p<0.05, *** p<0.01 ***
1-10 year time interval . The change in return on equity over the chosen time period is our measure for profitability change. It is regressed on variables for dividend increases and dividend decreases plus several control variables. The instrument used is a dummy variable that has a value of 1 after the introduction of JGTRRA in 2003 and value of 0 before. The Wald F-statistic in the lower column indicates the relevance of this instrument from the first-stage regression. Robust t-statistics are reported in the parentheses.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) ∆ROE0,1 ∆ROE0,2 ∆ROE0,3 ∆ROE0,4 ∆ROE0,5 ∆ROE0,6 ∆ROE0,7 ∆ROE0,8 ∆ROE0,9 ∆ROE0,10
Constant -1.968*** -1.385*** -0.994*** -0.202 0.121 0.229 -0.345 -0.394 -0.648 -2.011*** (-13.33) (-6.93) (-4.37) (-0.81) (0.42) (0.70) (-0.98) (-1.01) (-1.41) (-2.96) DPC * ∆DIV0 15.82*** -3.337* -19.18*** -30.03*** -38.00*** -41.92*** -33.38*** -27.52*** -18.85*** 3.177 (13.38) (-1.69) (-7.58) (-9.31) (-9.85) (-9.26) (-6.79) (-4.80) (-2.87) (0.25) DNC * ∆DIV0 -8.729*** -0.830 3.656** 4.998*** 13.39*** 17.60*** 11.73*** 16.45*** 7.740 -12.27 (-7.26) (-0.55) (2.08) (2.59) (4.59) (5.42) (3.71) (3.70) (1.46) (-1.20) M/B0 0.799*** 0.385*** 0.252*** 0.129 0.195** 0.0975 0.163 0.118 -0.158 0.0144 (17.88) (6.36) (3.60) (1.63) (2.28) (0.99) (1.55) (1.10) (-1.16) (0.11) ROE-1 -0.401*** -0.175*** -0.130*** -0.121*** -0.114*** -0.103*** -0.113*** -0.0976*** -0.0784*** -0.108*** (-55.74) (-25.15) (-17.00) (-13.51) (-10.59) (-8.53) (-8.47) (-6.14) (-4.56) (-5.39) ∆ROEt,0 -0.558*** -0.433*** -0.463*** -0.451*** -0.463*** -0.436*** -0.449*** -0.478*** -0.471*** -0.447*** (-71.95) (-56.21) (-53.31) (-49.12) (-45.07) (-37.64) (-34.17) (-31.51) (-26.54) (-22.09) TA0 0.00007*** 0.00003*** 0.00005*** 0.00007*** 0.00009*** 0.0001*** 0.0001*** 0.00009*** 0.00008*** 0.00003 (9.49) (3.38) (4.47) (4.50) (5.71) (5.92) (4.86) (3.02) (2.96) (0.63) D/E0 0.175* 0.794*** 1.022*** 1.142*** 1.135*** 1.151*** 1.079*** 1.038*** 1.423*** 1.432*** (1.82) (4.96) (6.06) (7.36) (6.95) (6.69) (5.71) (5.99) (7.00) (5.48) Adjusted-R2 0.224 0.199 0.217 0.209 0.206 0.175 0.198 0.214 0.221 0.214 IV Regression X X X X X X X X X X F-Test Instrument 3315.402 2436.396 1823.603 1385.605 900.300 653.094 474.557 324.206 176.604 57.818 N 101691 80785 65418 53032 43185 35031 27851 21872 16817 12831 Time Interval (Years) 1 Y 2 Y 3 Y 4 Y 5 Y 6 Y 7 Y 8 Y 9 Y 10 Y Standard errors in parentheses *p<0.10, ** p<0.05, *** p<0.01 ***
27
4.3 Robustness
To evaluate the robustness of our results, we have repeated the two-stage-least-square instrumental variable analysis using alternative estimation procedures.
The residual in our initial model might be correlated over time within each entity. To
correct for this potential autocorrelation, we have re-estimated model (1) using clustered standard errors by firm identifier. The results of this first robustness check can be found in table 5 of the appendix. The coefficients are the same compared to our initial analysis, as expected. There were also no large changes in significance for the coefficients of interest
DPC * ∆DIV0 and DNC * ∆DIV0, indicating that the autocorrelation is likely to be weak.
A second robustness check has been done by including industry and year fixed effects
into the standard linear regression model (1) from our initial analysis. In our alternative specification, we have already corrected for omitted variable bias by re-estimation model (1) using an instrumental variable approach. However, running a fixed effect regression is another method to control for these omitted variables. We assume here that the omitted variables vary across entities but do not change over time or industry. The results from including industry and year fixed effect into the original model are presented in table 6 of the appendix. They are slightly different from the results reported in the previous section.
Dividend increases do no longer imply a significant increase in profitability in year 2 after the
dividends have been increased. In the long run the coefficient DPC * ∆DIV0 is similar in sign
and magnitude, but not significant anymore in years 5 and 8. These results therefore provide weaker empirical evidence for the life-cycle and signaling theory compared to our initial analysis. For dividend decreases the results do not differ from our initial analysis and again do not provide empirical evidence for the life-cycle theory. Next to this, we observe that the adjusted R2 which indicates how well the model fits the data, is in all time intervals slightly higher when industry and year fixed effects are included into the model. Finally, we notice that the results from including fixed effects to the original model differ strongly from the results obtained in our instrumental variable regression. Although both methods propose to control for omitted variables they lead to different outcomes.
In the main part of the thesis, we have used the change in return on equity as our
dependent variable and measure for profitability. The third robustness check is done by redefining this measurement for profitability. The dollar change in net income before extraordinary items is used as the new dependent variable and regressed on the variables for
28
dividend increases and decreases plus several control variables, using the instrumental variable approach. The results are shown in table 7 of the appendix. We observe a similar
pattern for the coefficients DPC * ∆DIV0 and DNC * ∆DIV0, although they are larger and
there are some slightly changes in significance. The profitability growth in year 1 after the dividend increase, for example, is not significant anymore. Similarly, we find a positive and
significant coefficient for DPC * ∆DIV0 in the 9 year time period. This coefficient is also
insignificant in years 3 and 4, while it was significant when return on equity was used as the
measure for profitability. Furthermore, we observe that the adjusted R2 is much lower than it
was in the initial model. This indicates that the previous model, with the change in return on equity as the dependent variable, fits the data better.
In addition, we have split the sample into “two states of the world”. The first one is the
group whichincludes all firms with a market to book ratio above the median market to book
ratio for the total sample. The other group includes all firms with a market to book ratio below the median market to book ratio for the total sample. Finally, we have repeated the two-stage-least-square instrumental variable analysis for both groups. This is done because the proven intermediate and long term profitability growth after a dividend decrease is driven by an increased level of investments, according to the life cycle theory. The size of the profitability growth therefore also strongly depends on the amount of attractive investment opportunities available. Market to book ratio gives an indication of this. For this reason, we hypothesize that the long-term reaction to dividend changes is stronger for firms in the “high market to book value” group. The results obtained from those analyses can be found in tables 8 and 9. For both groups, we observe results that are in line with the life-cycle theory in the
intermediate and long term. Similarly, we found significant evidence that supports the signaling theory in the short run for firms in the above median market-to-book ratio group. When we compare the results of both groups, we furthermore observe that the negative reaction after a dividend increase in the intermediate and long term is stronger for firms in the below median market-to-book ratio group. A reason for this could be, that those firms do not generate enough cash internally and therefore cannot afford it to increase their dividend payouts. The results for dividend decreases, on the other hand, are similar for both groups. The reaction is stronger and more significant in the intermediate term for firms in the above
median group, while in the long run the coefficient DNC * ∆DIV0 is larger for the below
median group. This is not whatwe have hypothesized and the results of these analyses
29
4.4 Limitations.
There are some limitations to this research. First of all, the nature of the data limits a number of methodology procedures and analyses. It is, for example, impossible to calculate the relative change in cash dividends per share for all data points because some firms in the sample are paying out dividends in a given year, while they did not do that in the year before. A relative change is a better indicator for the change in dividends, compared to the dollar change used in this paper. Excluding dividend initiations and omissions from the sample, like Benartzi et al. (1997) and Nissim & Ziv (2001) did, is a potential solution to overcome this problem. However, this leads to a sample selection bias which gives unrealistic estimates. The
nature of the data alsomakes it impossible to calculate the compounded profitability growth
for a given firm in the estimated time period. This problem exists due tothe fact that the
indicator for profitability couldbe a negative value, which is the case when the company
suffered a loss in that given year in the sample period.
Other limitations of this paper are related to the methodology. We use an instrumental variable approach to correct for possible reversed causality and omitted variable bias in our initial model. The instrument used is a before-after dummy variable, that has a value of 1 when the fiscal year ends after the implementation of the Jobs and Growth Tax Relief
Reconciliation Act on 31 December, 2002. If the fiscal year ends before 31 December, 2002, the instrumental variable has a value of 0. The instrument seems to satisfy the instrument
relevance and exogeneity conditions for being a valid instrument. It is relevant sincea
positive correlation exists between JGTRRA and the change in annual cash dividends per share paid to common shareholders, which is our main independent variable. Next to this, we assume that JGTRRA is exogenous which implies that there is no direct effect of the
instrument on the dependent variable running through omitted variables. This assumption however is more controversial. Theoretically there is no direct relation between the
introduction of JGTRRA and the change in firm profitability, since JGTRRA only introduced
new regulation on personal income taxes. However, itcould also be argued that the
instrument does not satisfy the exogeneity condition because it’s a before-after dummy variable. The time dimension in such a dummy variable is the reason for this. It could influences the change in return on equity through changing market condition over time. However, is this a direct or indirect effect of the instrumental variable measuring the exogenous shock of the introduction of JGTRRA? Nevertheless, when it turns out that the
30
instrument does not satisfy the exogeneity condition we have to consider that it limits accuracy and reliability of the estimates from our IV-analysis.
This paper furthermore assumes that a linear mean reversion process exists in earnings. When profitability is already high, companies will find difficulty having strong earnings growth in the future. We control for this phenomenon by including variables into the model that measure a firm’s return on equity in the previous year and past changes in return on equity. Fama & French (2000) and Grullon et al. (2005), however, argue that this process of mean reversion in earnings is most likely to be highly nonlinear. The reasoning behind this is that large changes in earnings revert faster than small changes, while negative changes revert faster than positive changes. Assuming linearity when the true functional form of unexpected earnings has a nonlinear pattern has in their opinion the same implications as leaving out relevant independent variables. This is a second limitation of this study. A suggestion for future research is therefore to re-estimate the two-stage-least-square IV-regression model and in addition control for this non-linear mean reversion pattern in earnings.
31
5 Conclusion
This paper combines the life-cycle and signaling theory and analyzes how changes in a firm’s dividend policy affect future profitability in the short, intermediate and long term. It has been tested using two different methodologies.
In our initial analysis we have tested the relation using a standard linear regression model, comparable to the models used in prior studies like for example Benartzi et al. (1997) and Nissim & Ziv (2001) . We find that profitability increases significantly in each of the two subsequent years after a dividend increase. On the other hand, profitability decreases in each of the four years following the a dividend decrease. These findings are in line with the signaling theory and provide evidence for the information content of dividends in the short run. In the intermediate and long run, however, we don’t find strong evidence for life-cycle theory as we have hypothesized.
The results from our initial analysis and in prior studies are likely to suffer from
omitted variable and reversed causality bias. We therefore have re-estimated the model from our initial analysis, using a new two-stage-least-stage instrumental variable approach in order to correct both biases. The introduction of the “Jobs and Growth Tax Relief Reconciliation Act of 2003” in the United States is used as a relevant and exogenous instrument. The results provide empirical evidence for both the signaling and the life-cycle theory. This is surprising, since prior empirical studies failed to uncover this relation empirically.
We document that profitability is significantly higher in the year following a dividend
increase. For dividend decreases, we observe the opposite and conclude that profitability is significantly lower in the first year the after the dividend change. Although only significant for year 1, these findings clearly support the signaling theory and provide evidence for the information content of dividends in the short run.
The intermediate and long-run results, however, indicate that profitability declines
significantly in years 3 till 9 after a dividend increase. On the other hand, profitability
increases significantly in years 3 till 8 following a dividend decreases. This is in line with the
life- cycle theory, which hypothesizes that an increase in dividends signals that a firm reaches
their mature growth phase and intermediate/long term profitability will be flattened. The reason for this is the reduced level of investments in attractive investment opportunities that will drive long term profitability.
32
Although there a some limitations of this research, we still believe that our results shed
new light on the ongoing debate about dividend policy and it could be the starting point for further research.
33
Bibliography
Aharony, J., & Amihud, D. (1994). Regular dividend announcements and future unexpected earnings: An empirical analysis. Financial Review, (29), 125–151.
Arnott, R. D., & Asness, C. S. (2003). Surprise! Higher dividends =Higher earnings growth.
Financial Analysts Journal, 59(1961), 70–87.
Baker, H. K., Powell, G. E., & Veit, E. T. (2002). Revisiting the dividend puzzle. Do all of the pieces now fit? Review of Financial Economics, 11, 241–261.
Benartzi, S., Michaely, R., & Thaler, R. H. (1997). Do Changes in Dividends Signal the Future or the Past? The Journal of Finance, 52(3), 1007–1034.
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. Journal of Portfolio Management, (2), 5–8. Borges, M. (2008). Is the Dividend Puzzle Solved?
Brickley, J. (1983). Shareholders wealth, information signaling, and the specially designated dividend: An empirical study. Journal of Financial Economics, 12, 187–209.
DeAngelo, H., DeAngelo, L., & Skinner, D. J. (2007). Corporate Payout Policy. Foundations
and Trends® in Finance, 3(2008), 95–287.
DeAngelo, H., DeAngelo, L., & Stulz, R. M. (2006). Dividend policy and the
earned/contributed capital mix: a test of the life-cycle theory. Journal of Financial
Economics, 81, 227–254.
Easterbrook, F. H. (1984). Two Agency-Cost Explanations of Dividends. American Economic
Review, 74(4), 650–659.
Fama, E. F., & French, K. R. . (2000). Forecasting profitability and earnings. Journal of
