MSc Finance Thesis
Dividend policy and future earnings growth in the energy
industry
Jelle Ruiter S3186148 MSc Finance
Supervisor: Dr. Kristina Czura
Abstract
This paper analyzes the relevance of dividend policy in determining future earnings growth. Previous research has analyzed the importance of dividend policy in functioning as a signaling mechanism for future earnings growth, which is captured in the dividend signaling theory. The theory suggests that dividend payout provides investors with prospects of future earnings, but studies on this issue do not all find consensus. In this study I further analyze this signaling mechanism by focusing on the energy industry. Using a system generalized method-of-moments approach, the results show that dividend payout has a positive relation with future earnings growth in the energy industry. These results are robust against mean reversion effects of earnings, share repurchasing effects and efficiency, growth, size, age, and leverage factors. Additionally, I find preliminary results that the relation between dividend payout and future earnings may be stronger in the renewable energy sector when compared to the traditional energy sector.
Keywords: dividend policy, earnings growth, dividend signaling, energy industry JEL-classification: G35
2
I. Introduction
The relevance of dividend policy has long been subject of discussion among researchers and investors. A well-known theory in this field is the dividend irrelevance theory developed by Modigliani & Miller (1961). In frictionless markets, they state, dividend policy is irrelevant for company value. This theory resulted in the concept of the so-called dividend puzzle; companies that pay dividends are given higher values by investors, although according to the dividend irrelevance theory by Modigliani & Miller (1961) there is no rational for this higher valuation. Additional studies have tried to ‘solve’ the dividend puzzle, resulting in contradicting and controversial outcomes on the relation between dividend payout and company value.
A possible explanation for the dividend puzzle can be found in dividend signaling theory, which has its foundations in the findings by Lintner (1956). Dividend signaling theory states that managers are only willing to increase dividend payout ratios when they believe they can reach and sustain high future earnings for the company. Hence, the theory implies that higher dividend payout rates tend to be a signal for higher future earnings. In this way dividend payout can be a means to decrease information asymmetry between managers and shareholders and thus create value for the company owners.
Many empirical studies have analyzed whether dividend payouts show the assumptive correlation with future earnings as according to the theory. Healy and Palepu (1988), and Arnott and Asness (2003) state that there is a positive correlation between dividend payout and future earnings, which supports the dividend signaling theory. Others, such as Benartzi et al. (1997) and DeAngelo et al. (1996) show no supporting evidence for a relation between dividends and future earnings. Fama and French (2001) noted that the proportion of dividend-paying firms has sharply declined during the years 1973 to 1999, which may also affect the outcome of studies related to dividend theory. As these studies illustrate, the existing literature shows a lack of consensus on the correlation between dividend policy and future earnings. Outcomes may be subject to different time periods, research methods or sample specifics, as will become clear in the literature section. Therefore, additional research on dividend policy can fill in gaps in existing literature and broaden perspectives.
3 (Aharony & Swary, 1980). However, if the same company is at the same time making losses, then these shareholders with long-term investments will eventually lose value.
Nevertheless, there are several reasons why dividend policy could lead to value creation for shareholders. The management of a company that has a higher payout ratio should be more committed and more critically review the projects that it will take on when compared to a similar company that has a lower payout ratio. Management will be more disciplined to invest in only the most profitable opportunities, as most of the firm’s earnings cannot be reinvested since it will be distributed to the shareholders. A high dividend payout ratio also leaves less cash available to managers to invest in empire building, expensive personal benefits or any other non-profitable investment which is not in the best interest of the shareholders. When viewing the dividend signaling theory from this perspective, a higher dividend payout ratio may cut costs, which can increase earnings growth and bring long-term value to a company. This may clarify the higher value for dividend-paying companies, which could explain the dividend puzzle.
On the other hand, a higher payout ratio also comes at a cost. A company with a high payout ratio may have to forego profitable investment opportunities in order not to lose its reputation of being a high dividend paying company. This in turn may result in low earnings growth because of a lack of investments in profitable opportunities. On the long term this would imply that a company with many available growth opportunities will lose value when it maintains a high payout ratio.
As stated, existing literature does not bring consensus on the relation between dividend policy and company earnings. Also, the interpretation of dividend policy leaves open the question whether dividend policy is related with either positive or negative consequences for earnings growth. In this study I will bring additional analyses on dividend theory by investigating the correlation between dividend payout and future company earnings growth, focusing specifically on the energy sector. The main contribution of this study is in the fact that I use a generalized method-of-moments estimation technique to account for the dynamic earnings growth variable. This makes the results of the research more efficient and unbiased, as will be explained in the data section. Furthermore, this research also contributes to existing literature by focusing explicitly on the energy industry in finding a relation between dividend payout and future earnings growth. To the best of my knowledge the energy sector has not been analyzed for the existence of a dividend signaling mechanism. The advantage of concentrating on one specific sector is that the results will not be subject to differences in macroeconomic circumstances across sectors, and thus provides robustness of the results for the specific sector.
4 made by companies in the energy sector are usually characterized by high risks because of the long term returns on the investments. The industry is rapidly innovating which can cause that today’s investments may be useless when the project is finished and would start to pay-off. A disciplined management that only invests in the most profitable investment opportunities as a result of a high-dividend payout policy can bring value to these companies, for example. The energy industry is characterized by companies that are involved in producing and supplying energy, including companies involved in the exploration and development of oil or gas refining and power utility companies. The energy industry can be divided in two categories of companies, non-renewable energy companies and renewable energy companies. Non-renewable energy companies produce energy that is extracted from oil, gas (fossil fuels) or nuclear resources, and renewable energy companies produce energy from hydropower, biofuel, and wind and solar power.
During recent years, the renewable energy sector has been attracting more and more (political) attention by various countries in the world as a result of environmental issues and climate agreements such as the United Nations Paris Agreement and the 1997 Kyoto protocol. As a consequence, the renewable energy industry shows large developments and is stimulated more and more, at the cost of traditional energy resources like gas and oil. This provides thoughts on whether there could also be a difference in the correlation between payout ratio and future earnings growth across the ‘traditional’ energy industry and the renewable energy industry. This research will address this issue and analyze whether there exist significant differences between these subindustries.
More specifically, this research will aim to answer the following research questions:
1. Is a higher dividend payout ratio a reliable signal for future earnings growth in the energy industry?
2. Is there a difference in the correlation between dividend payout policy and future earnings growth across the traditional energy industry and the renewable energy industry?
Considering the ongoing discussions on the relevance and implications of dividend policy on company value, this research can provide valuable information. Furthermore, because of the widespread attention for sustainable investments and the importance of the energy industry in this process, this research and its results can provide valuable information to investors willing to invest in (sustainable) energy companies. Lastly, the results can be valuable for dividend payout decisions of companies active in the energy industry.
5 III will explain the research method that will be used in this research, followed by section IV, which contains the data sources and descriptive statistics. Section V will show the main findings of the analysis. Finally, section VI will provide a summary of the research and state concluding comments.
II. Literature review
Determinants of a company’s future earnings are of crucial significance for the share value of a company. Financial market analysists are eager to know the profitability of their long-run investments and they therefore try to identify and analyze key factors that relate to future company profitability. This has caused that extensive research has been executed on identifying the variables that have a relation with a company’s future earnings. Also, the relevance of dividend payments on future profitability and the determinants of dividend policy have been discussed in earlier research in various papers. Therefore, this next section will provide an analysis of the existing literature on dividend policy and the related dividend signaling theory. Furthermore, this section will also provide an analysis of existing studies on the determinants of dividend payout policy, in order to analyze key factors of relevance for this study.
2.1 Dividend irrelevance and the dividend signaling theory
Dividend irrelevance theory states that a company’s dividend payout policy does not add value to a company and therefore dividend policy should not affect a company’s share price. The theory was first implied by Miller & Modigliani (1961), who argued that in a world with no market imperfections dividend distribution should not affect the company’s financial structure nor its share price. The main argument for this theory is that in perfect markets an investor should be indifferent between receiving a dividend which he then could use to reinvest in the company himself, or having these funds being reinvested in profitable projects directly by the company. The theory even argues that dividend distribution could damage the company as the money is better reinvested by the company itself, because then the company does not have to pay for the cost of dividend distribution. Hence, in a perfect market dividend policy should not affect share value from an investor’s point of view and could potentially even harm the company.
6 dividend signaling theory, which states that dividend increases signal positive future earnings prospects. However, although the theory could explain why a higher payout ratio causes short-term positive abnormal returns for example (Aharony & Swary, 1980), the empirical validation of the theory is not widely accepted. Some dividend policy studies show supporting evidence for the validation of the theory, but the theory is contested in other studies. An overview of relevant studies on the theory follows.
Dividend signaling theory states that dividend payments reduce information asymmetry on future earnings by acting as a signaling instrument between company insiders and outside company shareholders. Miller and Rock (1985) argue that well informed inside managers can provide less informed outside shareholders with a prospect of growing future company earnings through dividend payout, revealing the true nature of current company earnings to the market. Unexpected increases in dividend payout tend to signal a forecast of improved future results, causing increased expectations of future dividends which results in an increase in share value. Consistent with these findings, Healy and Palepu (1988) find evidence that abnormal stock price reactions after dividend initiations (omissions) are correlated with future earnings increases (decreases) in the year after this announcement. Healy and Palepu (1988) find that companies that started paying dividends experience earnings growth in the year of the dividend announcement and the two years following. In contrast, companies that stop paying dividends encounter earnings decline in the year before the dividend date but recover in the years after the dividend omission. In line with the dividend signaling theory, they argue that dividends payments act as a signaling mechanism for investors, who view dividend initiations and omissions as managers’ forecasts of future earnings changes.
7 The mentioned studies provide supporting evidence to the validity of the dividend signaling theory, yet the theory is not widely accepted and still used as a concept by financial market participants because not all studies show comparable results. Although some studies provide evidence for the theory, other research challenges the validity of dividend payments as being a signal for enhanced future earnings prospects. DeAngelo et al. (1996) find no evidence that a firm’s dividend decisions help picking out firms with superior future earnings, which presents doubts to the signaling theory. They argue that dividends tend not to be reliable indicators for future earnings growth, because overly optimistic behavior by managers leads these managers to overestimate prospects of the company when growth prospects vanish. The research used a sample of companies that had experienced annual earnings decline after nine or more years of earnings increases and they analyzed managers’ dividend decisions in the year of earnings decline on signaling future earnings prospects.
Another contradicting study to the signaling theory is the research by Grullon et al. (2005). They show that dividend changes are uncorrelated with future earnings changes when one controls for the non-linearities in the earnings process. They state that changes in dividends are not useful in predicting future changes in earnings, which is additional evidence that disapproves the dividend signaling theory.
Benartzi, Michaely & Thaler (1997) first analyze the reactions in the market to dividend changes. They find that firms that increase their payout ratio show positive excess returns for a 3-year period. However, they do not find unexpected future earnings growth that could explain these excess returns. Hence, they state that changes in dividends do not signal information about future earnings prospects. However, consistent with Lintner’s (1956) model, they do find that dividend-increasing firms are less likely to have earnings decreases in later years when compared to firms that do not change their payout ratio. Hence, the current earnings can be viewed as a ‘permanent’ increase, but this does not indicate anything about future earnings growth. Their findings are in line with the idea that low dividend payouts can cause inefficient principal-agent problems which lead to low earnings growth levels, and that high dividend payout ratios may lead to more critically chosen (profitable) investments.
8 Additional evidence for the argument that dividend changes have an impact on risk is provided by Dyl and Weigand (1998). They find that dividend initiations convey information that a company’s earnings and cash flows have lower volatility. However, they do not find additional evidence that company earnings will increase after a dividend increase. Nevertheless, they state that lower systematic risks may cause lower costs of capital, which may in turn lead to higher earnings.
In summary, the existing literature provides rather mixed evidence on the economic implications of dividend payout on future company earnings. The first evidence for a positive correlation between dividend payout and future earnings was found during the ‘60s of the 20th
century. However, the market environment of that time was much different from the current capital market. Fama and French (2001) found that the proportion of dividend-paying firms in the U.S. has declined significantly between 1978 and 1999, from 66.5% to only 20.8%, respectively. They argue that this is partly due to changing characteristics of publicly traded firms. However, their main finding is that firms have just become less willing to pay dividends. They argue that this may be a result of multiple explanations, but they are all linked to less perceived benefits of dividends for all kind of capital market participants.
As a result, during recent years substantial skepticism has arisen against the dividend payout theory. Since Lintner (1956), dividend preferences, the economic environment, and dividend payouts by companies have changed. This may imply that changes in dividend payouts may also affect the relation between dividend payout and future earnings growth and the validity of the dividend signaling theory. As a result of changing company behavior in terms of dividend payouts and the many contradicting studies on the existence of a positive relationship between dividend increases and future earnings, the dividend signaling theory is still not widely accepted. Additional research on this topic can bring new insights and test the validity of existing theories. Furthermore, additional research can also bring new insights on the robustness of the dividend signaling theory against the effects of changing market circumstances and differences across industries, which is where this study can bring valuable knowledge.
2.2 Determinants of dividend policy
9 In their dividend irrelevance theory Miller & Modigliani (1961) stated that the dividend policy of a company is not relevant for shareholders’ wealth in a perfect frictionless market. However, in the real-world markets are not frictionless and companies pay dividends because of varied reasons. These determinants of dividend payout policies have been discussed in numerous studies. However, existing literature only gives a rough framework of the characteristics that differentiate dividend-paying firms from non-dividend-paying firms. These characteristics will be examined in the following.
2.2.1 Profitability, growth, and company size
Fama and French (2001) noticed the fall of the proportion of firms that pay dividends between 1978 and 1999. They tried to explain the cause of this decrease in dividends and examined the characteristics of dividend payers. The three variables they study that affect the likelihood of a firm paying dividends are profitability, growth, and company size. They state that the decrease in dividend-paying firms can be explained by an increase in newly listed firms in the index they analyzed, which tends to consist of smaller firms with many growth opportunities. This could explain the decrease in dividend-paying firms as firms with many growth opportunities tend to have low, if any, dividend payout rates.
Dividend paying firms differ from non-dividend paying firms on multiple areas, an overview of which is provided in table 1. As analyzed by Fama and French (2001), the profitability for dividend-paying firms was higher than that for non-dividend-payers, in terms of earnings before interest to total assets. This gap was even larger when measuring profitability in terms of total common stock earnings to the total book value of equity.
Another aspect on which dividend paying firms differ from non-dividend paying firms is growth opportunities, which Fama and French (2001) measured by asset growth rates and R&D expenses. Asset growth rates tend to be largest for companies that have never paid dividends (16.50% per year) and is much higher than that of dividend payers (8.78%) and former dividends payers (4.67%) between 1963-1998. The same results show up when measuring growth in terms of market value of assets to book value. Furthermore, they state that firms with better growth opportunities also tend to have higher R&D expenses, which should be a rather unsurprising result.
Besides differences in profitability and growth opportunities, another characteristic on which dividend-paying firms differ from non-dividend-paying firms is that dividend-paying firms tend to be larger in size. Size of a firm is measured in terms of total assets. The total assets of dividend-paying firms tend to be more than ten times as large as non-dividend-paying firms between 1963-1998, as can be seen in table 1.
10 before interest to assets ratio, and earnings to book value of equity. Growth is measured by asset growth rates, and R&D expenses-to-assets. Firm size is measured in terms of total assets.
TABLE 1: Profitability, growth, and size differences between 1963-1998 Dividend
payers
Non-payers Profitability
Earnings before interest to assets (%) 7.82 5.73
Common stock earnings/book equity (%) 12.75 6.15
Growth
Rate of growth in assets (%) 8.78 11.62
Market equity/book equity (%) 1.39 1.42
R&D/Assets (%) 1.61 2.07
Size
Book assets ($MM) 1,389.18 110.43
*Source: Fama and French (2001)
The table reports averages of annual estimates. Fama and French (2001) did not indicate whether the differences in the variables between dividend-payers and non-payers are statistically significant.
2.2.2 Firm maturity
Taken the results by Fama and French (2001) as a basis, DeAngelo et al., (2006) show that dividend payout policy changes over the lifetime of a company. Their findings are in line with the so-called life-cycle theory of dividends which states that mature firms are more likely to pay dividends than younger firms with high investment opportunities and limited (financial) resources. Mature firms are defined as firms with a high retained earnings-to-total equity ratio, whereas younger firms are firms that have few retained earnings subject to total equity. The trade-off between paying out dividends or retaining them in the company develops during a company’s lifetime as earnings accumulate and investment opportunities decline.
In their research, DeAngelo et al. (2006) control for profitability, growth, firm size, total equity, cash balances, and dividend history. Their findings are in line with the argument for reduced agency costs when dividend payouts are larger. They argue that when mature firms had not paid dividends, then their cash balances would become so large that holding long-term debt would be meaningless for these companies, which gives rise to increased agency costs as the amount of cash available to managers becomes very large.
11 Another argument for the importance of the life-cycle factor is in share repurchasing. Firms start to use share repurchases to pay out volatile cash flows over the years and as they mature, they use regular dividends to pay out permanent cash (Jagannathan, Stephens, and Weisbach (2000) and Guay and Harford (2000)). They do so, as mature firms tend to have less profitable investment opportunities, which grows the cash on their balance sheet. Hence, repeated repurchases signal that a firm is getting at its maturing stage and that their income is stabilizing. A firm’s life-cycle stage is hence a crucial factor in determining dividend payout policy. However, their study fails to find evidence that dividend initiations are followed by greater earnings growth, higher net income, or decreased systematic risk for the firm.
As mentioned earlier, the characteristics that define a company’s dividend policy are far from fully known. Existing research only provides a rough framework. Further research should determine additional variables that may be related to dividend policy. However, determining all these variables is almost impossible considering that every company is different in nature. In this research I will follow the results of existing literature and use these findings to determine the research model, which will be discussed in the next section.
III. Methodology
As has been stated in the literature section, different studies on the topic of dividend payout theory come with different and sometimes opposite results. These differences could occur from using different variables, different methods, and different data sample specifics.
In this research I will further analyze the relationship between the dividend payout ratio and future earnings growth, by specifically focusing on the energy industry. The energy industry consists of companies involved in producing energy, which includes extraction, manufacturing, refining and distribution of resources. These energy resources can be split in non-renewable energy resources, such as fossil fuels, and renewable energy resources, such as solar and wind energy. This research will narrow down on the relationship between dividends and earnings growth, by distinguishing in these non-renewable energy companies and renewable energy companies to examine whether there exist significant differences between these two subindustries.
In this research I will follow a similar approach to Zhou and Ruland (2006) in setting up my model by measuring future earnings growth as the compound annual earnings (per share) for common shareholders.
3.1 Hypotheses
12 Hypothesis 1: The dividend payout ratio is positively correlated with future earnings growth in the energy industry.
And:
Hypothesis 2: The correlation between the dividend payout ratio and future earnings growth in the traditional energy industry does not significantly differ from that in the renewable energy industry.
This research will use analysis based on a generalized method-of-moments (GMM) estimation to test whether dividend payout ratio is positively correlated and has a significant effect on future earnings growth of companies in the (renewable) energy industry. The GMM estimation method is increasingly popular and designed for situations with ‘’small T, large N’’ panels (Roodman, 2009), which is relevant for this study as will be explained in the data section. The research methodology is using system generalized method-of-moments estimators, using future earnings growth rates of ten, five, three, and one year time-period(s).
3.2 The model
The empirical analysis in this research will test the relation between future earnings growth and payout ratio. The model will include various control variables to account for possible underlying effects incorporated in dividend payout that may affect future earnings growth. The regression model that will be used in this research is presented as follows:
𝐶𝐴𝐺𝑅(𝑡,𝑡+𝑘) = 𝛼 + 𝛽1∗ 𝑃𝑅𝑡+ 𝛾1∗ 𝜒𝑡+ 𝜀𝑡 (1)
Here 𝐶𝐴𝐺𝑅𝑡+𝑘 captures the compound annual earnings growth over a time period t to k and
𝑃𝑅𝑡 covers payout ratio in year t. All control variables are summarized by the 𝜒𝑡 component, and 𝜀𝑡 is the error term. The below table describes all variables in the model, including control
variables.
Figure 1: Definitions of the variables including control variables
CAGR(t,t+k) The earnings growth rate over period t to t+k, defined as the compound annual growth rate of earnings over the time period t to t+k in terms of earnings per share.
PRt Payout ratio in year t, defined as dividend per share in year t divided by the earnings per share in year t.
E/Pt The earnings yield in the year t, defined as earnings before interest and taxes (EBIT) divided by the (end-of-year) market capitalization.
ROAt Profitability in terms of return on assets in year t, defined as EBIT divided by total assets.
13 SALESt Firm size in terms of the natural logarithm of sales in year t.
MYt Maturity in year t, defined as retained earnings divided by total equity. CBt Cash balance in year t, defined as the natural logarithm of total cash and cash
equivalents on the balance sheet.
LEVt Leverage, defined as book value of long-term debt divided by book value of total assets.
REPt Share repurchases, defined as net common stock issued (retired) divided by EBIT in year t.
CAGR(t-5,t) Lagged earnings growth, measured as compound annual earnings growth rate between t and t-5.
CAGR(t-3,t) Lagged earnings growth, measured as compound annual earnings growth rate between t and t-3.
CAGR(t-1,t) Lagged earnings growth, measured as compound annual earnings growth rate between t and t-1.
The main variable of interest in equation (1) is the dividend payout ratio in year t (PRt). The dividend payout ratio will be regressed on future compound annual earnings growth on multiple future periods k (k = 10, 5, 3 and 1).
Earnings growth is measured as the compound annual earnings growth rate of earnings per share, which is calculated using the formula:
𝐶𝐴𝐺𝑅(𝑡,𝑡+𝑘)= (𝐸𝑃𝑆𝑡+𝑘 𝐸𝑃𝑆𝑡
)(𝑘1)− 1
Calculation errors could occur when there is a switch in the sign between the beginning value and the ending value of EPS. However, these calculation errors are no concern for this research as the Stata software that is used for the estimation techniques automatically excludes the values that suffer from this issue. Nevertheless, because of this, 4,860 observations are lost on the 3 year earnings growth variable (CAGR(0,3)), 4,378 observations on the 5 year earnings growth
variable (CAGR(0,5)), and 2,782 observations on the 10 year earnings growth variable
(CAGR(0,10)). The one-year earnings growth calculation does not suffer from this issue, as here
the absolute value for a negative beginning value of EPS can be used.
Earnings per share is the net income to common stockholders divided by the number of outstanding shares, which is calculated as:
14 The dividend payout ratio is defined as the dividends paid out divided by the net income of the company. Another method of calculating the dividend payout ratio is to divide the dividends per share (DPS) by the earnings per share (EPS), which will result in the same outcome. This is the construction for the dividend payout ratio calculation used in this research. The hypothesized relation between the dividend payout ratio and future earning growth is positive (hypothesis one) and should not differ between the two subindustries (hypothesis two).
3.2.1. Estimation technique
The estimation technique that is used in this research is a generalized method-of-moments (GMM) estimator, specifically an Arellano-Bond estimator (Arellano & Bond, 1991). The GMM estimator method is useful for situations with dynamic independent variables, which means that the independent variable is depending on its own past values and therefore can be correlated with past and current values of the error term. In these situations, the GMM estimator is preferred to a fixed effect (LSDV) model, as the results will be more efficient and unbiased. As will be stated in section 3.2.2, the lagged version of the independent earnings growth variables are used to control for the mean reversion effect. Therefore, the GMM estimating method is considered an appropriate estimating method for this study.
Moreover, the GMM method is also useful for (1) panels with few time periods and many individuals, (2) individual fixed effects in the sample and (3) heteroskedasticity and autocorrelation within individuals but not across them (Roodman, 2009). As will be explained in the data section, my data fulfills all of these requirements; the data consists of large N and small T, individual fixed effects are presents and heteroskedasticity and autocorrelation within individuals is highly likely. Therefore, the GMM estimating method is a suitable procedure for this research.
3.2.2 Co-variates
The model includes multiple control variables for profitability, size, growth opportunities, maturity, leverage, share repurchases. Moreover, the model also controls for mean-reversion of earnings using the generalized method-of-moments (GMM) method.
15 over time, meaning that dividend and earnings growth is equal, then a low payout ratio must be offset by a high earnings yield (E/P) or by high expected growth. Thus, the correlation between dividend payout and earnings can be affected by a negative relation between earnings yield and earnings growth. Therefore, the earnings yield variable is added to the model to account for this effect.
The second control variable in equation (1) is return on assets (ROAt), which is a profitability indicator used to measure the efficiency of the companies’ assets in being able to produce earnings for the company. It is measured as earnings before interest and taxes (EBIT) divided by the total assets of the respective year. A more efficiently operating company could be able to reach higher future earnings growth and on the same time be able to maintain a higher current payout ratio. However, a company that already has this prominent level of efficiency, measured as a high return on assets, should experience lower levels of future earnings growth. Therefore, a negative relation between the current return on assets and future earnings growth is expected.
The CAPEXt control variable controls for growth opportunities which is defined as the total capital expenditures divided by total assets. The more growth opportunities a company is able to take on the larger its capital expenditures will be, as growth opportunities require an upfront investment to result in profitable future payoffs. However, when a larger portion of company earnings is reinvested in growth opportunities then a smaller part would be left to distribute to the shareholders in the forms of dividends. As Fama & French (2001) analyzed, dividend paying companies generally have less growth opportunities and may therefore be able to distribute earnings as dividend payments to the shareholders instead of reinvesting them in the company which can affect the level of future earnings growth. The growth opportunities variable captures the effect of this occurrence. The expected sign for the relation between growth opportunities and future earnings growth is expected to be positive, as high current investment in growth opportunities should result in higher future earnings growth.
16 An additional control variable is added to account for maturity, the variable MYt. DeAngelo et al. (2006) show that more mature companies are more likely to pay dividends than younger firms. A developed and mature firm is in the circumstance that earnings accumulate, and investment opportunities decline. This may cause that more developed companies are more likely to have a high current dividend payout ratio, but on the same time have declining future company earnings growth. Therefore, a negative relation is expected between the dividend payout and future earnings growth. Additionally, DeAngelo et al. (2006) use the retained earnings to total equity variable as a measure for maturity. I will follow this construction and control for firm maturity with the retained earnings to total equity ratio. Additionally, DeAngelo et al. (2006) also argue that if mature firms had not paid dividends then their cash balances would have become so large that being leveraged would be meaningless for these companies, which could give rise to additional agency costs that could affect future earnings growth. For these reasons, a control variable for maturity (MYt) as well as a control variable for cash balance (CBt) is added to the model, and both are expected to be negatively related with future earnings growth.
Leverage is an additional variable added to the model, referred to as LEVt, which is defined as the debt to asset ratio. This ratio defines the extent to which a company’s assets are financed with liabilities. Leverage can create value for a firm in that it can discipline managers in companies with little growth opportunities and in this way reduce agency costs. Moreover, leverage can incentivize companies that have many growth opportunities to only invest in the most profitable opportunities. Leverage could thus create more efficiency, but it could also affect dividend payout ratio in that debt-financing has no implications for dividend payments to be distributed to the capital investor, as opposed to equity-financing. Hence, leverage can be related to payout ratio, but also to future earnings growth. This effect will be accounted for by adding a control variable for leverage. The relation between current leverage ratio and future earnings growth is expected to be positive, as a result of increased efficiency and reduced agency costs.
17 alternative for paying out dividend to the shareholders. In share repurchases companies do not distribute their earnings as cash to the shareholders, but they buy back their own shares from the existing shareholders. This affects the dividend payout ratio, as payouts are reduced when companies choose to use earnings to buy back their own shares. However, whether a firm chooses to return their profits to shareholders in the form of a cash dividend or via a share buyback, the effect is the same as in both ways the shareholders will receive the earned profits from the company. Hence, an increase in share repurchases may lower the dividend payout ratio and could understate the effect that it has on future earnings growth. Arnott and Asness (2003) do not include a direct variable for share buybacks and state that they have a lack of data to reliably test the impact of increases in buybacks. However, they state that from the initial evidence the increase in share buybacks have not understated the importance of the dividend payout ratio in their research. Fama and French (2001) account for share repurchases by treating these as an additional payout of earnings. In this research I follow a similar approach by following the repurchase payout setup of Zhou and Ruland (2006). They define a repurchase payout variable as the ratio of common stock repurchases to earnings. This variable, referred to as REPt, will be used in the model to test for the robustness of the results.
Another possible issue that could occur is that a possible positive correlation between the dividend payout ratio and future earnings growth is caused by a mean reversion of earnings. As stated by Arnott and Asness (2003): “A temporary drop in earnings could raise expected future compound earnings growth from this lower base. The temporary earnings drop would simultaneously raise the current payout ratio, D/E, because sticky dividends do not fall as much as earnings.” Hence, they state that a decrease in current earnings could cause a higher current dividend payout ratio and at the same time a higher future earnings growth, which could explain the possible correlation between these variables. To account for this issue, they added past real earnings growth to their regression of the prior ten years. If the mean reversion hypothesis would then be true, then this would be fully explained by the lagged earnings growth variables, and the dividend payout ratio would lose much of its importance.
18 In summary, the research model tries to analyze the relation between future earnings growth and the dividend payout ratio, controlling for potential efficiency, growth, size, age, leverage, share repurchase or mean reversion explanations, using a GMM approach.
IV. Data
All company classification and company financial data is retrieved from the Refinitiv Eikon financial database. Companies are classified based on their subsector in the Industry Classification Benchmark (ICB). The ICB is a globally utilized standard for the categorization and comparison of companies by industry and sector.
The ICB has four distinct levels, in which the highest level (Industry) corresponds with the broadest industry classification and the lowest level (Subsector) corresponds with the most detailed classification of the companies’ business activities. The distinct levels are industry, supersector, sector and subsector, respectively. This research will use the ‘Energy’ and ‘Utilities’ supersectors from the ICB classification to retrieve a sample of companies active in the ‘energy’ industry. From the ‘Utilities’ supersector, only the companies involved in the sector ‘Electricity’ sector are used, which excludes the ‘Gas, Water and Multi-utilities’ companies and ‘Waste and Disposal Services’ companies. From the ‘Energy’ supersector all underlying industries are included in the sample. The complete definitions of all the subsectors are included in Appendix A. Based on these specifics the companies are then retrieved from the Refinitiv Eikon database.
TABLE 2: Sample overview
Supersector Sector Subsector # of companies
Energy Oil, Gas
and Coal
Integrated Oil and Gas 71
Oil: Crude Producers 772
Offshore Drilling and Other Services 29
Oil Refining and Marketing 145
Oil Equipment and Services 320
Pipelines 76
Coal 204
Alternative Energy
Alternative Fuels 137
Renewable Energy Equipment 212
Utilities Electricity Alternative Electricity 231
Conventional Electricity 395
Total 2,592
19 two different subsectors. The final sample was then split in two sectors based on their classification for traditional energy or renewable/sustainable energy. The results of this sample split are presented in table 3. The classification is based on the subsector definition that can be found in appendix A.
TABLE 3: Energy specification sample split
Energy type classification Subsector # of companies
Traditional energy Integrated Oil and Gas 71
Oil: Crude Producers 772
Offshore Drilling and Other Services 28
Oil Refining and Marketing 145
Oil Equipment and Services 319
Pipelines 76
Coal 204
Conventional Electricity 395
Traditional Energy Companies 2,010
Renewable energy Alternative Fuels 137
Renewable Energy Equipment 212
Alternative Electricity 231
Renewable Energy Companies 580
Total 2,590
The total sample of traditional energy companies consists of 2,010 companies and the number of companies in the renewable energy sample is 580.
The financial data for these companies is retrieved via the Refinitv Eikon database, specifically the Refinitiv Fundamentals data category. This database category provides standard financial company data including items from income statements and balance sheets. The time period for data collection covers the years 1995 to 2019. This restricted time period was chosen because as stated by Fama & French (2001) dividend policy behavior tends not to be constant over a long time period. Moreover, due to data limitations for earlier years and the increased attention for environmental issues since the 2000s this time period is deemed most relevant for this particular research.
20 4.1 Overall sample statistics
Table 4 provides descriptive statistics for the total sample that includes all companies. The table consists of three parts; part (a) provides statistics on the full sample period, part (b) on the period from 1995 to 2005, and part (c) the years 2005 to 2019. The total sample is winsorized at the 1st-percentile and the 99th-percentile to remove outliers. This causes that some of the variables have the same maximum and minimum value in the time-period and subindustry subsamples.
When comparing the number of observations between the 1995-2005 and 2005-2019 period, we observe that the total sample largely consists of data from the 2005-2019 period. The average compound annual earnings growth rate (CAGR) is 4.6%, 7.2%, 15.4% and 3.4% over the ten, five, three and one year time periods, respectively. On average this earnings growth rate tends to be higher in the 1995 to 2005 period, compared to the 2005 to 2019 period. The average payout ratio is 30.9% of earnings, which tends to be roughly equal over both time-periods.
When observing the control variables, we notice that the share repurchases seem to be larger in the 2005 to 2019 period, compared to the 1995-2005 period. In 2005-2019 the average leverage ratio also tends to be slightly higher. Considering the mean share repurchase (REP) variable of -12.2 percent of EBIT in the total sample, the effect of share repurchasing and firms changing characteristics to share repurchases as mentioned by Fama and French (2001) may be present in our sample.
TABLE 4: Descriptive statistics for the total sample per period Total sample period
22 4.3 Specific industry statistics
Considering the second hypothesis, which states that the relation between payout ratio and future earnings growth does not differ between both subindustries, the total sample is split in a group of companies belonging to the traditional energy industry and a group of companies classified as active in the renewable energy industry, as described in table 3. Table 5 presents statistics for this sample split.
TABLE 5: Descriptive statistics traditional energy companies
Traditional energy companies Renewable energy companies
23 The average dividend payout ratio in the renewable energy industry sample is lower (24.9% of earnings) than that of the traditional energy industry sample (32.1% of earnings). The compound annual earnings growth rate for all considered time periods is larger in the traditional energy industry sample when compared to that of the renewable energy industry. Other control variables also seem to differ across the subindustries. The share repurchasing (REP) variable is larger in the traditional energy industry sample compared to the renewable energy industry sample, indicating that the cash amount to repurchase shares was on average higher in the traditional energy industry sample. Additionally, the earnings yield and return on assets were on average lower (or more negative) for renewable energy companies in the sample and capital expenditures as a percentage of total assets were also on average smaller. Differences between the two subindustries are thus presumable, but unfortunately the number of observations for the renewable energy industry sample is far below that of the traditional energy industry.
4.4 Normality, endogeneity and multicollinearity
All variables are tested for normality by calculating probability for skewness, kurtosis and a Skewness/Kurtosis test measure for normality. The results of these tests can be found in Appendix C. As was to be expected, all the dependent variables in the model are non-normal, suffer from kurtosis and are skewed. To transform the data to more normally distributed data, the independent variables, excluding lagged independent variables, are transformed by using log-transformation. This log-transformation does result in a loss of observations of values that are negative, which mainly affects the payout ratio (PR) and earnings yield variable (E/P) data. Therefore, I follow the assumption made by Zhou and Ruland (2006) to only include companies in the sample that have positive earnings in year t. Besides that, this makes possible the log-transformation of the data, Zhou and Ruland (2006) state the argument that there is no clear economic meaning for a negative dividend payout ratio. These negative payout ratios were mainly the result of negative earnings in year t, which made the calculation for the payout ratio negative. I follow this argumentation and only include companies with positive earnings in year t.
24 applied for the asymptotic normality of the coefficients. Therefore, it is expected that the GMM estimators approach the true distribution and that the non-perfectly normal distribution of the data is not a severe issue.
Endogeneity concerns may also be present in the dataset as it is highly likely that the 2,510 companies in our sample have unique properties that affect its earnings (growth). To cover this issue, a Hausman test is performed on the sample to analyze possible correlation between the unique errors and the dependent variables. The null hypothesis of the Hausman test is that there is no such correlation. The p-value of the Hausman test indicates a value of 0.000, meaning that the null hypothesis is rejected and that there are fixed company-level effects in the model.
To test whether the independent variables are related to each other a Pearson pairwise correlation matrix is performed, which is presented in Appendix D. Almost all independent variables are significantly correlated with each other, but most of these correlation values are low. There is no widely accepted threshold that determines when multicollinearity becomes a severe issue for analysis. Nevertheless, in the correlation matrix presented in Appendix D almost all correlation values are below 0.5, implying there is no concern for a straight linear relation between two independent variables. Therefore, the multicollinearity issue is not considered as severe for this study and the effects of multicollinearity are ignored further on.
V. Results
The main findings of the paper are presented in this section. The section will first present univariate analysis results and will proceed with the results from a multivariate analysis procedure.
5.1 Univariate analysis
Table 6 represents a Pearson correlation matrix between the (log-transformed) current dividend payout ratio and future and past earnings growth rate, where earnings growth is in terms of compound annual growth rate (CAGR) of earnings. The past and future earnings growth periods (lags and leads, respectively) are specified on 10, 5, 3, and 1 year time periods.
25
TABLE 6: Pearson correlation of dependent variable and payout ratio for all companies
PR CAGR (-10,0) CAGR (-5,0) CAGR (-3,0) CAGR (-1,0) CAGR (0,1) CAGR (0,3) CAGR (0,5) CAGR (0,10) PR 1 CAGR (-10,0) -.326*** 1 CAGR (-5,0) -.306*** .542*** 1 CAGR (-3,0) -.217*** .370*** .457*** 1 CAGR (-1,0) -.031*** .023** -.004 -.044*** 1 CAGR (0,1) .048*** .017 .024*** .023*** -.000 1 CAGR (0,3) .173*** -.226*** -.202*** -.155*** -.001 -.073*** 1 CAGR (0,5) .152*** -.203*** -.175*** -.147*** .011 -.059*** .742*** 1 CAGR (0,10) .105** -.041 -.158*** -.096*** .036*** -.019* .555*** .676*** 1
*** significant at 1%-level, ** significant at 5%-level, * significant at 10%-level
Table 7: Pearson correlation of dependent variable and payout ratio for traditional energy companies
PR CAGR (-10,0) CAGR (-5,0) CAGR (-3,0) CAGR (-1,0) CAGR (0,1) CAGR (0,3) CAGR (0,5) CAGR (0,10) PR 1 CAGR (-10,0) -.326*** 1 CAGR (-5,0) -.308*** .54*** 1 CAGR (-3,0) -.215*** .379*** .445*** 1 CAGR (-1,0) -.035*** .022* -.006 -.018** 1 CAGR (0,1] .041*** .032** .020** .025*** -.001 1 CAGR (0,3) .164*** -.235*** -.209*** -.179*** -.003 -.058*** 1 CAGR (0,5) .142*** -.197*** -.177*** -.153*** .007 -.037*** .752*** 1 CAGR (0,10) .091** -.048 -.163*** -.089*** .039*** -.020* .560*** .678*** 1
26
Table 8: Pearson correlation of dependent variable and payout ratio for renewable energy companies
PR CAGR (-10,0) CAGR (-5,0) CAGR (-3,0) CAGR (-1,0) CAGR (0,1) CAGR (0,3) CAGR (0,5) CAGR (0,10) PR 1 CAGR (-10,0) -.329*** 1 CAGR (-5,0) -.297*** .553*** 1 CAGR (-3,0) -.235*** .318*** .519*** 1 CAGR (-1,0) -.013 .028 .016 -.190*** 1 CAGR (0,1) .194*** -.046 .048** .020 .007 1 CAGR (0,3) .260*** -.175*** -.175*** -.079*** .002 -.134*** 1 CAGR (0,5) .242*** -.252*** -.170*** -.138*** .045* -.149*** .723*** 1 CAGR (0,10) .248*** -.015 -.143*** -.153*** .009 -.000 .524*** .659*** 1
27 can also be seen in the analysis, as (almost) all of the long-period future earnings growth variables are negatively related with the long-term past earnings growth variables.
Table 7 represents a Pearson correlation matrix for the traditional energy companies’ sample. This table represents comparable results to the complete sample correlation matrix; however, the larger proportion of the total sample is represented by traditional energy companies which should be noted. Table 8 indicates the results for the renewable energy companies’ sample. Comparable results are applicable to this subsample. The mean reversion effect seems to be present in this subsample, the correlation between past earnings growth and dividend payout is negative and the correlation between future earnings growth and payout ratio is positive. The correlation between the current payout ratio and the future earnings growth leads seems to be even stronger in the renewable energy company sample, compared to the traditional energy company sample.
The univariate analysis results show comparable results to that of Zhou and Ruland (2006). As they indicate, the positive relation between dividend payout and future earnings growth rate may be the result of low past earnings growth of dividend paying companies due to mean reversion of earnings. This is what is controlled for in the multivariate analysis.
5.2 Multivariate analysis
For the multivariate analyses, the generalized method-of-moments (GMM) model is used. In the analyses fixed effects are controlled for using a twostep model, which results in more efficient and robust outcomes. To preserve sample size in panel data with gaps, as applicable to the sample in this study, forward orthogonal-deviations are used instead of first differencing, which is recommended by Roodman (2009). Lastly, heteroskedasticity and autocorrelation is accounted for by using robust standard errors.
28
TABLE 9: System GMM for all energy companies without controls
Dependent variable CAGR (0,10) (1) CAGR (0,5) (2) CAGR (0,3) (3) CAGR (0,1) (4) Intercept .059*** (.008) .061*** (.009) .106*** (.010) .617*** (.089) PR .022*** (.008) .028*** (.009) .059*** (.011) .486*** (.073) Wald-proba .007 .002 .000 .000 Hansen-probb .000 .000 .000 .385 AR(1)c .000 .000 0.000 .046 AR(2)d .078 .000 0.000 .706 Observations 4122 7087 8442 10953 # of companies 666 934 1001 1164 # of instruments 17 22 24 26
*** significant at 1%-level, ** significant at 5%-level, *** significant at 10%-level
a – Probability result of the Wald-test to find out whether explanatory variables are significant
b - Test for over-identifying restrictions in GMM dynamic model robust estimation
c – Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (no autocorrelation)
d – Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (no autocorrelation)
Table 9 also indicates the validity of the models.The Hansen-probability test is an indication for the validity of the instruments in the models. The Hansen-probability is low for the ten, five, and three year lead models, indicating that the number of instruments is appropriate for the number of groups. This means that the instruments in the three models are valid, and the number of instruments does not outnumber the number of groups in the models. The one-year lead model is not an appropriate model as the Hansen test can not reject the null hypothesis that all instruments in the model as a group are exogenous.
The Wald-statistic is significant for all models indicating that the regressors are jointly significant in explaining the dependent variable. Furthermore, all of the four equations show statistics for first- and second-order autocorrelation. First-order autocorrelation (AR(1)) is present in all four equations at a 5%-significance level. Second-order autocorrelation is present in the five-year and three-year lead model indicated by the AR(2) probability indicator. Due to the presence of autocorrelation and non-validity of the one-year lead model, these models are not the most efficient models.
29 When comparing the results in table 9 with that of table 10, we see that the coefficients for the payout ratio are quite robust for the ten, five and three year leads in both tables. This means that even after adding control variables to the model, the relation between payout ratio and future earnings growth does not change so much.
TABLE 10: System GMM for all energy companies with controls
Dependent variable CAGR (0,10) (1) CAGR (0,5) (2) CAGR (0,3) (3) CAGR (0,1) (4) Intercept -.050 (.073) -.153* (.090) -.314*** (.111) -.426 (.597) PR .024*** (.007) .035*** (.011) .051*** (.015) .217*** (.081) E/P .005 (.006) .002 (.009) -.005 (.010) -.103* (.061) ROA -.043*** (.010) -.095*** (.016) -.123*** (.022) -.002 (.134) CAPEX -.001 (.005) .009 (.007) .018** (.009) .018 (.031) SALES .000 (.004) .008* (.005) .011* (.007) .023 (.033) MY -.025*** (.008) -.037*** (.011) -.033** (.013) -.073 (.051) CASH -.002 (.003) -.011*** (.004) -.009 (.006) -.009 (.028) LEV -.005 (.005) -.004 (.007) -.016** (.006) -.025 (.023) SHREP -.006 (.005) -.008 (.011) -.008 (.015) -.059 (.096) CAGR(-5,0) -.008 (.053) .013 (.025) -.071 (.058) -.296 (.213) CAGR(-3,0) -.009 (.024) -.015 (.018) -.027 (.029) .141 (.104) CAGR(-1,0) -.000 (.003) .001 (.003) -.003 (.004) .037* (.020) Wald-proba .000 .000 .000 .000 Hansen-probb .065 .002 .000 .027 AR(1)c .100 .001 .056 .007 AR(2)d .863 .165 .349 .683 Observations 1192 2300 2840 3633 # of companies 309 443 525 614 # of instruments 46 61 67 73
*** significant at 1%-level, ** significant at 5%-level, *** significant at 10%-level
a – Probability result of the Wald-test to find out whether explanatory variables are significant
b - Test for over-identifying restrictions in GMM dynamic model robust estimation
c – Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (no autocorrelation)
d – Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (no autocorrelation)
Wald-30 probability is significant for all models indicating that the regressors are jointly significant in explaining the dependent variable. Furthermore, all of the four equations do not suffer from second-order autocorrelation, indicating there is no critical concern for autocorrelation in the models.
In the presented models, payout ratio is positive and strongly significant at all different earnings growth levels, indicating that the dividend payout is positively related to future earnings growth in the energy sector. At the one-year lead level (column 4 in table 10) a 1% increase in the dividend payout increases next year’s earnings growth with 0.00217, or 0.22 percentage points. The relation is less strong at longer time periods; a current 1% increase in payout increases the future ten year earnings growth with 0.00024, or 0.02 percentage points. These results lead us to accept the first hypothesis. Current dividend payout tends to be positively correlated with future earnings growth, the effect is strongest at the shortest time period and it diminishes on longer time periods.
Notably, most of the past earnings growth control variables in the models are not significant. Only the one year future earnings growth tends to be mildly significant at the ten percent level with the previous year’s earnings growth. The mean reversion effect is thus not visible in our analysis.
From the other control variables only the current return on assets (ROA) and maturity (MY) variable are consistently correlated with future earnings growth at all future earnings growth levels, except for the one year future earnings growth. The current return on assets is negatively correlated with future earnings growth at ten, five and three year leads, indicating that a higher return on assets now decreases the future earnings growth. This matches my expectation because a company that already has a prominent level of efficiency, measured as a high return on assets, should experience lower levels of future earnings growth.
31 5.2.1. Traditional energy & renewable energy sample comparison
As indicated in the data section (section IV), the total sample largely comprises companies active in the traditional energy industry. A plausible explanation for this is that the renewable energy industry is a relatively new and developing industry and therefore comprising of few(er) companies, whereas the traditional (gas, oil etc.) energy industry is a mature industry in which companies have had a longer time period to enter the industry. Additionally, the industry specifications of the Industry Benchmark Classification (ICB) changes frequently and the alternative energy subsector has been classified only in December 2019, which could also be a factor causing the relatively few numbers of renewable energy companies. This skewed distribution of the total sample is important to consider the final results.
Table 11 presents the results of the estimation model for the traditional energy sample, and table 12 provides the results for the renewable energy industry.
The traditional energy subsample shows comparable results to the total sample. The payout ratio is positively correlated with future earnings growth at all future earnings growth levels, and the effect is strongest on the shortest future time period. The return on assets variable and maturity variable are the only controls that are consistently significant over all models and both variables are negatively related with future earnings growth. The validity of the variables in the model seems to be adequate and there is no indication for second-order autocorrelation.
32
TABLE 11: System GMM for traditional energy companies
Dependent variable CAGR (0,10) (1) CAGR (0,5) (2) CAGR (0,3) (3) CAGR (0,1) (4) Intercept -.094 (.075) -.138 (.088) -.259** (.113) -.456 (.425) PR .023*** (.007) .032*** (.011) .047*** (.015) .165** (.084) E/P .004 (.006) .001 (.008) -.007 (.010) -.091* (.049) ROA -.043*** (.012) -.095*** (.016) -.123*** (.023) .007 (.077) CAPEX -.006 (.005) .012 (.008) .023** (.011) .031 (.035) SALES .003 (.004) .007 (.005) .006 (.007) .029 (.031) MY -.022** (.009) -.034*** (.011) -.030** (.015) -.067 (.049) CASH -.004 (.003) -.010** (.004) -.006 (.006) -.012 (.028) LEV -.006 (.005) -.005 (.006) -.014** (.007) -.027 (.024) SHREP .001 (.008) -.017 (.013) -.009 (.018) -.007 (.087) CAGR(-5,0) -.007 (.025) .013 (.025) -.056 (.061) -.246 (.221) CAGR(-3,0) .011 (.026) -.019 (.017) -.037 (.028) .134 (.111) CAGR(-1,0) .001 (.004) -.001 (.004) -.002 (.005) (.053)** (.024) Wald-proba .000 .000 .000 .001 Hansen-probb .104 .016 .000 .011 AR(1)c .117 .003 .075 .012 AR(2)d .908 .127 .398 .893 Observations 1096 2071 2521 3195 # of companies 280 386 446 510 # of instruments 46 61 67 73
*** significant at 1%-level, ** significant at 5%-level, *** significant at 10%-level
a – Probability result of the Wald-test to find out whether explanatory variables are significant
b - Test for over-identifying restrictions in GMM dynamic model robust estimation
c – Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (no autocorrelation)
33
TABLE 22: System GMM for renewable energy companies
Dependent variable CAGR (0,10) (1) CAGR (0,5) (2) CAGR (0,3) (3) CAGR (0,1) (4) Intercept .375 (.343) -.405 (.272) -.685** (.346) -7.099* (4.269) PR .017 (.020) .056** (.025) .076** (.034) 1.069*** (.399) E/P .009 (.010) -.031 (.020) -.040 (.027) -.654** (.274) ROA -.040* (.022) -.104*** (.032) -.113* (.058) -1.081 (.769) CAPEX .012 (.009) .011 (.016) .013 (.019) .194 (.156) SALES -.031** (.015) .030 (.022) .076*** (.025) .630** (.285) MY -.060*** (.021) -.046 (.044) -.037 (.037) .343 (.250) CASH .012 (.010) -.024 (.020) -.062*** (.023) -.438** (.216) LEV -.003 (.011) .029 (.027) -.015 (.023) .132 (.189) SHREP -.012*** (.004) .001 (.013) -.006 (.012) -.384*** (.136) CAGR(-5,0) -.124 (.110) -.178** (.079) -.170** (.086) -.930 (.698) CAGR(-3,0) -.006 (.022) .003 (.043) -.019 (.090) .059 (.558) CAGR(-1,0) -.002 (.001) .004 (.003) -.011* (.006) -.055* (.030) Wald-proba .000 .000 .000 .002 Hansen-probb .984 .713 .620 .284 AR(1)c .423 .060 .209 .352 AR(2)d .317 .282 .497 .492 Observations 96 229 319 438 # of companies 29 57 79 104 # of instruments 43 61 67 72
*** significant at 1%-level, ** significant at 5%-level, *** significant at 10%-level
a – Probability result of the Wald-test to find out whether explanatory variables are significant
b - Test for over-identifying restrictions in GMM dynamic model robust estimation
c – Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (no autocorrelation)
d – Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (no autocorrelation)
5.3 Robustness
34 on the ten-, five-, and three-year level. However, at the one-year level the relation is insignificant. Furthermore, in the fixed effects model the mean reversion effect is visible, which is not present in the GMM method. Additionally, when using this model, the control variables for current sales and cash balances tend to be negatively correlated with future earnings growth. The results of the fixed effects model are considered additional evidence to support the first hypothesis, as current dividend payout ratio also tends to be positively correlated with future earnings growth in this model.
In the estimation model I have controlled for the effect of share repurchases on future earnings growth, as share repurchases may weaken the relation between current dividend payout and future earnings growth which is indicated in the methodology section. The control variable for share repurchases does not show a significant relation with future earnings growth, implying that the correlation between current dividend payout and future earnings growth is consistent when controlling for share repurchases.
In summary, the multivariate results show that current dividend payout rate is positively related to future earnings growth at the ten-, five-, three- and one year level in the energy sector. This means that hypothesis one can be accepted as there is a positive correlation between dividend payout rate and future earnings growth in the energy industry. This result is robust to mean-reversion and share repurchasing effects.
The second hypothesis can not be reliably accepted due to a lack of observations for the renewable energy industry sample, which makes that the model for this sample can not be properly defined. Future studies can address this issue when there will be more data available as time passes.
VI. Conclusion
35 to provide investors with prospects of (positive) future earnings. Others, such as DeAngelo et al. (1996) and Benartzi et al. (1997) find no positive relation between dividend payout and future earnings growth. The study by Fama and French (2001) shows that time-varying components may also affect the relation between dividend payout and future earnings growth.
In this research I provide supporting results for the validity of the dividend signaling theory by specifically analyzing the energy sector. I find a positive relation between current dividend payout and future earnings growth on ten-, five-, three- and one year levels. Using a system generalized method-of-moments methodology these results are robust for mean reversion effects of earnings, share repurchasing effects and efficiency, growth, size, age, leverage factors.
Additionally, when splitting the sample in traditional energy companies and renewable energy companies, I find initial evidence that the relation between current dividend payout and future earnings growth is even stronger in the renewable energy industry when compared to the traditional energy industry. These results are considered initial results, as due to the limited number of observations on renewable energy companies the econometrical model is not properly defined. Future research can thus use my findings and further analyze the industry specific effects by adding data and validity to the model, as more observations will appear over time. Moreover, additional research on the relation between dividend payout and future earnings growth can also test for robustness of the relation in other specific industries.
