An Examination of the Relationship between Dividend Payout and Future Earnings Growth within Emerging and Frontier Markets.

(1)

An Examination of the Relationship between Dividend

Payout and Future Earnings Growth within Emerging and

Frontier Markets.

Idan Isvoranu

i.isvoranu@student.rug.nl

(2)

1

ABSTRACT

Special thanks to Professor Auke Plantinga and Wytze Riemersma for making this collaboration between the University of Groningen and Trustus Capital Management possible.

Additional gratitude is dedicated to Marco Balk and Johan Smith whom supported and assisted me throughout the entirety of this program, helping me overcome issues which were vital towards the creation of this paper.

(3)

2

1. Introduction

Market investors continuously attempt to analyse firms’ behaviour and financial characteristics to draw conclusions beneficial for predicting future growth. One of the most commonly examined correlations across financial literature is that between dividend policy and profitability. Since Miller and Modigliani (1961) creation of the Dividend Irrelevance Theorem the correlation between dividend payout and future earnings growth was accepted as a negative one across financial literature. Their paper suggests, that shareholders will be indifferent towards increased wealth based on dividend payments, or price valuation: and thus, reinvestment should be maximised to assist the firm to grow. An interpretation of this theory therefore implies, that increasing dividend payout would stagger a firm’s growth due to its limitation of cash flows available for further investment. A similar idea was raised by Myer’s (1984) Pecking Order theory, as he noted indifference towards sources of funding, and thus the one with the lowest cost of capital should always be used first- retained earnings.

However, when Arnott and Asness (2003) examined this correlation in practice in the context of the US equity market, they found contradicting proof- providing a new point of view to what was until that point accepted as financial fact. This surprise discovery of a positive relationship between the payout ratio and future earnings growth sparked a wave of further studies on the matter. These studies examined the relationship using both economy aggregate methodologies, as well as firm level data- however all were exclusively conducted in the context of Developed markets.

The nature of the relation between dividend payout and future earnings growth has important implications for investing decisions, in particular for those investors involved in investment styles such as value and growth. While the results examined by nearly all previous literature provide strong evidence in favour of a positive correlation in the context of Developed markets, relatively little is known about Emerging and Frontier markets.

In contrast to Developed markets, Emerging and Frontier markets are characterised by rapid economic development, smaller market capitalisation, higher illiquidity, and reduced market accessibility for international investors (MSCI, 2020). The financial climate in these countries, comprised of factors such as investment, credit, and tax regulations lead to differentiated characteristics of the firms found within them. On aggregate, the financial characteristics of the companies found within them varies from those commonly found in Developed markets: characteristics such as level of preference to leverage, tax treatment of dividends, target dividend payout (Benavides et al, 2016), and increased volatility, all brings to question whether the positive correlation found by Arnott and Asness (2003) can also be detected across these markets.

Using firm level data for the examination period of 1999-2019, this thesis sets out to test the hypothesis that there does exist a positive correlation between the dividend payout ratio and future earnings growth across both Emerging and Frontier markets; as was observed in Developed markets.

The conclusions drawn from these findings will offer an additional tool for international investors to assess firms’ growth prospects across these two market types, and provide them with a superior advantage towards making better investment decisions.

(5)

4 Additional robustness test will be conducted and will examine the strength of the correlation across an alternative measure of earning: EBITDA.

Furthermore, two potential explanations to the phenomena will be explored. First, the Dividend Signalling Hypothesis will be tested. Following this theory, managers use dividends – or in the case of this examination, dividend payout – as a way to reduce informational asymmetry. Thus, managers will intentionally manipulate dividends behaviour in order to signal to the market of positive future prospects.

Then, the Free Cash Flow Hypothesis will be tested. Using Tobin’s Q as a proxy for positive growth potential of the firm: an interaction variable will be added to the regression in order to assess whether increased dividends with the goal of reducing agency costs, can explain the positive correlation between earnings growth and payout.

(6)

5

2. Literary Review

A firm’s dividend behaviour is often seen as one of the strongest channels of communication between managers and shareholders. It can be used as a way to signal confidence to investors regarding the future of the company, or on the contrary, it may also hold information seen as early signs of a future decline. As such, gaining a deeper understanding of the relationship between dividend structure and future earnings growth has been at the centre of a large body of financial literature.

Since 1961, with Miller and Modigliani’s publication of the Dividend Irrelevance Theorem, the relationship between payout and earnings growth has been generally accepted as an inverse one. In their study they concluded that investors are indifferent between dividend or capital gains, and therefore a firm’s dividend policy would have no effect over its share price or capital structure. Based on this interpretation: decreasing payout, and thus increasing retained earnings would allow for more funds to be invested in future profitable projects, ultimately yielding higher earnings growth. This theory however, heavily relies on a set of assumptions. Assumptions such as the lack of taxes, symmetrical information across capital markets, and that all investments yield a positive NPV. As will be later explored, these assumptions are unrealistic- and when relaxed, heavily interfere with the consistency of the theory.

Despite its limitations, Miller and Modigliani’s (1961) theorem continued to influence financial literature and was later used to develop Myer’s (1984) Pecking Order Theory. This theory notes that firms are indifferent towards sources of funding, and thus, would always choose the one with the lowest cost of capital: hence retained earnings. Once again, this idea contradicts reality, assuming that investors will naturally interpret undertaking of additional debt as a negative sign towards the future of the firm. An assumption that has been proven inaccurate across multiple studies (Brander and Lewis, 1986), and will be further elaborated on later in this paper.

Lastly, the negative relationship between dividend payout and future earnings growth was noted by Gordon (1962) in the constant-growth valuation model. In his model, Gordon (1962) suggests that expected return is a simple function of dividend yield – denoted as Dividend/Price (D/P) – plus a constant growth variable (G). Dividend yield can further be broken down into the product of: Dividend Payout – Dividend/Earnings (D/E) – multiplied by Earnings yield – Earnings/Price (E/P) –. Following the (unrealistic) assumption that dividends and growth are constant through time, an implication of this model suggest that a low payout ratio (D/E) would be naturally offset by increased earnings yield (E/P). Thus, implying a negative correlation between payout and earnings.

(7)

6 markets using a 10-year time horizon, with the added control of inflationary changes. Their findings agreed with Arnott and Asness (2003) and concluded that the correlation does not only appear in US equity markets as previously examined, but is an international phenomenon that remains robust across all Developed markets tested. These findings confirm that the correlation noted is robust to international differences such as managerial culture or tax regimes. Vivian (2006) used a more narrowed approach- where he investigated the correlation’s robustness across aggregate portfolios of firms based on industry indices in the UK market. From his findings, he concluded that the positive relationship between aggregate payout and aggregate earnings growth is robust across all industries tested, and therefore industry membership does not influence the correlation. He further noted that the findings exert statistically stronger significance across the five-year time horizon versus the ten-year horizon. Finally, Parker (2005) used index aggregate financial parameters to test the hypothesis in the US, UK, and Australian markets, with particular focus on its robustness over time. His findings suggest that while correlation remains positive across time periods, its strength and statistical significance varies across decades.

Arnott and Asness (2003) paper, along with its noted successors, followed one similar characteristic which limits an investor’s ability to apply these finding in an investment strategy- they all reached their conclusions based on aggregate level data of market or industry indices. In nearly all noted examinations, the indices used were capitalisation weighted, and thus the financial characteristics of a few firms hold the risk of significantly overshadowing those of others. This risk is particularly highlighted when comparing and contrasting the papers of Fama and French (2001) and DeAngelo and Skinner (2004). While both papers investigated dividend payout patterns across an identical time frame; firm level data reported a decrease in payout ratios across the examined period, while aggregate data reported an increase, skewed upwards due to the weighting of the index leaders.

In order to overcome these limitations and provide results applicable for investment decisions, this exploration will follow methodologies inspired by the works of Zhou and Ruland (2006), and Flint et al. (2010), who examined the correlation between dividend payout and earnings growth using firm level data in the US and the Australian markets.

Based on the findings of the previously noted literature in the field, a series of potential explanations arise.

In their examination of dividend policy decisions in a setting with asymmetric information, Miller and Rock (1985) discuss the manner in which managers can use dividend increases to project their confidence in stability and future growth. An interpretation of this cash flow signalling hypothesis would suggest, that a positive correlation between dividend payments and future earnings growth would likewise further translate to an identical correlation between dividend payout and future earnings growth. Flint et al. (2010) and Vivian (2006) both attempt to examine the reliability of this explanation to the phenomena. While Flint et al (2010) did not find significance in their sample, Vivian (2006) did, however only across the shorter time horizon examined.

(8)

7 by Flint et al (2010) using the similar methodology of employing Tobin’s Q as a measure of profitable investment opportunities. While Zhou and Ruland (2006) found significant evidence for this explanation in the US market, Flit et al (2010) were unable to statistically confirm the theory, however did not deny it either.

Lastly, the most commonly investigated explanation of the correlation examined in this paper, is attributed to mean reversion in earnings, combined with the sticky downwards functionality of dividends. As Lintner (1956) denotes, earnings exhibit strong mean reversion, however dividends are less volatile. Therefore, temporarily low earnings with a relatively static dividend, would lead to a high payout ratio which will confirm the future increase in earnings as they return to the mean over time (Fama & French, 2000). This association has been noted and accounted for by nearly all previous studies mentioned through the addition of lagged earnings as a control variable in the model.

This paper’s contribution to the aggregate body of financial literature concentrates around the examination of the reliability of the correlation between Dividend Payout and Future Earnings Growth within the context of Emerging and Frontier markets.

All financial markets are categorised into one of three classes: Developed, Emerging, or Frontier. The classification criteria vary across bodies; however, the most commonly accepted one, and the one used in this examination is that of the Morgan Stanley Capital International group (MSCI). When classifying the different capital markets, MSCI emphasises three main pillars: Sustainability of economic development, size and liquidity of capital markets, and international investor market accessibility criteria. The table below, extracted from the MSCI Market Classification Framework, illustrates the criteria benchmarks which must be satisfied per classification group:

Illustration #1: MSCI Market criteria May 2020

(9)

8 payout structure, however, their degrees of influence varies. This variation is not consistent based on a particular industry, but rather country specific trends play a larger role. Mitton (2004) and Lin et al (2011) further explore the drivers for this variance in characteristics, and found corporate governance, ownership structure, and taxation to be the largest determinants for a firms’ financial characteristics. In line with the conclusions later found in this paper, Porta et al (2000) notes that these drivers vary greatly across countries, however the differences found between Frontier and Emerging market countries are larger than those amongst Emerging and Developed market countries- explaining the similarities and differences in the trends our exploration will later derive.

Inspired by the methodologies of the previously noted literature in the field, particularly Zhou and Roland (2006) firm level examination of the correlation, the following parameters were used as control variables in this paper’s primary regression:

Firm Size: The results yielded from Nawaz and Azhar (2019) found that firm size in Emerging markets holds a negative effect on earnings growth across different methods of measurements, noting that larger firms generally see a slower growth rate than smaller firms with similar characteristics. This can be attributed to Chan, Karceski, and Lakonishok (2003) explanation of the phenomena, noting that as a firm grows and its innovation spreads, the level of competition in the market rises, and thus it becomes harder to obtain abnormal returns. An additional argument for this view made in Gibart’s Law, which states that firm growth is independent of its size, and thus a larger firm will experience marginally slower growth the larger it is. Therefore, this variable need to be controlled for, and is expected to express a negative coefficient.

Return on Asset (ROA): As is noted by Mussalam (2018), over the long run, profitability measures such as return on assets tend to exhibit mean reverting characteristics. The further away the return on asset ratio is from its mean, the slower earnings growth will be (French & French, 2000). Freeman, Ohlson and Penman (1982) attributed these findings to competitive market environments, as they state that market competition prevents consistent abnormal returns from being achieved, and while reducing ratios such as ROA in the short term when are above their mean, in the long-term lead to further market growth. Therefore, ROA is expected to be inversely correlated with future earnings growth, and thus the coefficient of this variable is expected to be negative.

Earnings Yield: Following the theory outlined in Flint et al (2010), provided the market is efficient and no mispricing of securities occur, a positive relationship should hold between P/E and future earnings growth. This statement is in line with the findings of Allen et al (1998), whom found through their study that in the Australian markets, firms at the upper quartile of earnings yield had below median earnings growth. As earnings yield is calculated as the inverse of P/E, this variable is expected to have a negative coefficient, indicating a higher earnings yield will be associated with lower future earnings growth. Financial Leverage: As Fama and French note in their 2002 paper, given strong firm characteristics are present, highly leveraged companies will, on average, have higher levels of investment in non-current assets, which in turn yield higher earnings growth. This is further in line with the findings of Brander and Lewis (1986) noting higher leverage allows firms to behave in a predatory manner against weaker competition, leading to higher earnings potential in the short run. Therefore, the variable of leverage is hypothesized to be positively correlated with earnings growth in the context of our model.

(10)

9 (2006) identified that asset growth does not exhibit non-ununiform mean reversion characteristics statistically significant enough to interfere with the robustness of the model. Sensitivity tests were conducted in the case of our analysis and similar conclusions were found. Therefore, Asset Growth will be identified as a single variable in the regression, expressed on the same time horizon as earnings growth.

(11)

10

3. Methodology

3.1. Data

The data employed in this exploration was extracted from Bloomberg L.P. financial data vendor software. As the data required is derived from securities traded in Emerging and Frontier market exchanges, the exploration holds the risk of facing inaccurate or incomplete data sets (Rosenberg & Goodwin, 2016). To control for this potential issue, data points were randomly selected, and cross examined with other data providers such as Wharton Research Data Services and Thomson Reuters Eikon. No discrepancies were found for any of the observations tested. However, the examined data contains many incomplete panels, thus, conducting country dependent examinations was impossible to achieve.

In order to ensure a sufficient amount of data, while maintaining the upmost relevance of the trends found for potential investors, the time period used for this examination was 1999-2019.

The universes observed includes 12 markets, 6 Frontier (Pakistan, Kenya, Kuwait, Morocco, Bahrain, Vietnam) and 6 Emerging (Brazil, Hong Kong, India, South Africa, South Korea, Taiwan). These markets were selected following the MSCI Frontier Market Index and the MSCI Emerging ex-China Market Index, where the sampled countries make up more than roughly 70% of the respective indices, and thus can be trusted as accurate proxies for the trends apparent in “Emerging” and “Frontier” markets (MSCI, MSCI Frontier Market Index (USD), 2020). A total of 15,996 firms are observed- making this the largest examination of its kind to explore this phenomenon. Table 1 illustrates in further detail the composition of the universes observed.

To avoid the issue of survivorship bias, each country sample includes all listed and delisted firms on its exchange for the entire observation period. Unlike previous literature in the field, no additional criteria were imposed on the sample, except the requirement that the firm paid dividend for at least one year throughout its listing.

(12)

11

Table #1: Universe Data by Country

Country Firms Observed (N = 15,996) Weight in Corresponding Universe I. Emerging Markets (12,938) Brazil 467 4% Hong Kong 2,608 20% India 1,909 15% South Africa 1,488 12% South Korea 3,355 26% Taiwan 3,111 24%

II. Frontier Markets (3,058)

Pakistan 760 25% Kenya 79 3% Kuwait 227 7% Morocco 161 5% Bahrain 73 2% Vietnam 1,758 57%

(13)

12 The primary multivariant regression for this analysis is as follows:

𝐸𝐺𝑖𝑡_ℎ= 𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+ 𝛽2𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝐸𝑃𝑖𝑡+ 𝛽5𝐿𝐸𝑉𝑖𝑡+ 𝛽6𝐴𝐺𝑖𝑡_ℎ+ 𝛽7𝑃𝐸𝐺𝑖𝑡_−1,3,5+ 𝜖𝑖𝑡 (1)

Using the extracted parameters, the following variables were derived:

𝐸𝐺𝑖𝑡ℎ= Earnings Growth, the dependent variable in the examined regression. Measured as the compounded annual earnings growth rate from 𝑡 to 𝑡ℎ (with h denoting the three

observation horizons of one- three- and five- years).

𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡 = Dividend Payout, the key independent variable in the regression. Measured as year 𝑡

annual reported dividend, measure as total cash outflow, divided by year 𝑡 Bottom Line Net Earnings.

𝑆𝑖𝑧𝑒𝑖𝑡 = Firm Size. Measured as the natural log of Market Value of Equity at year 𝑡. Market

Value of Equity will be calculated as Closing Share Price at year end multiplied by Total Outstanding shares. This computation follows that seen in Fama and French (2002), and is common across financial literature.

𝑅𝑂𝐴𝑖𝑡 = Return on assets. Measured as end of year 𝑡 Bottom Line Net Earnings divided by end

of year 𝑡 Book Value of Total Assets.

𝐸/𝑃𝑖𝑡 = Earnings Yield. This variable is calculated as the firm’s Bottom Line Net Earnings for

year 𝑡 divided by the firm’s year end Market Value of Equity.

𝐿𝐸𝑉𝑖𝑡 = Financial Leverage. For the purposes of this exploration this variable will be computed

as firm’s Total Book Value of Short- and Long-Term Debt at end of year 𝑡, divided by the firms Total Assets at the end of year 𝑡.

𝐴𝐺𝑖𝑡ℎ= Future Growth of Total Assets. Measured as compounded annual growth in Total Assets from year 𝑡 to year 𝑡ℎ.

𝑃𝐸𝐺𝑖𝑡−1,3,5 = Lagged Earnings Growth. Represented as three separate variables in the regression.

(14)

13

3.2. Descriptive statistics

Table 2 and Table 3 report the descriptive statistics for the two data sets. In order to control for the effects of outliers in the data, the top and bottom 1% of observations were removed using the winsoring approach. This is in line with other similar literature in the field. Each examined universe is constructed of firms from multiple countries; each with its own restrictive legislation, investor expectations, and firm behaviours. Furthermore, similar to other studies, the groups do not differentiate observed firms by parameters such as industry membership, and thus, median capital structures, growth characteristics, and reinvestment choices may differ. This variation however does not interfere with the results as each firm in the panel is only regressed across itself and its own characteristics. Therefore, using descriptive statistics as reference measures to similar examinations supplements the interpretation of the yielded results, however is limited in its ability to accurately describe the samples.

The Payout ratio observed, with the mean of 0.4459 in Emerging markets, is comparable to that found in Flint et al (2010) study of 0.4113. Similar payout ratios between developed and Emerging markets is in line with the conclusions of Aivzan and Booth (2003), whom examined aggregate payout ratios across 9 Emerging markets in comparison to the United States and noted similar results. However, the mean payout ratio of 0.5965 in Frontier markets is slightly higher than those seen in both developed and Emerging markets. This idea is in line with the observations noted in Trang (2016) study, which hypothesises that higher payout ratios in Frontier market firms might be a consequence of favourable tax treatment of dividends over capital gains, combined with the variable costs of capital.

Furthermore, in both samples the variation in payout ratios is quite large, with an interquartile range of 0.1177 to 0.6062 in Emerging markets, and 0.2279 to 0.7281 in Frontier markets. This further illustrates the variation which is observed driven both by country standards, and industry. It is important to note however, that such variation will not interfere with the reliability of the results as each observation is only regressed across its corresponding firm panel.

The mean and median Return On Asset variable appears to be relatively similar in the case of both Emerging and Frontier markets, with a median value of 0.0302 and 0.0383 respectively. In both cases, the values are significantly lower than that noted in Zhou and Ruland (2006), of 0.074, for the US market. Singh (2004) elaborates on this observation, noting that developed markets exhibit higher average returns on assets and equity associated with increased efficiency of utilisation of non-current assets and human capital. However, in line with the findings of Trang (2016) in Frontier market firms, this suboptimal return on investment may be offset by the lower cost of labour and capital, thus making it more efficient than that found in Emerging markets.

The mean leverage observed in the study notes comparatively low debt to asset ratios of 0.2498 and 0.2671 for the Emerging and Frontier samples respectively. This is significantly lower than those noted by Zhou and Ruland (2006) of 0.465. In Bem et al (2017) examination of capital structure of construction firms in Emerging and Frontier markets, similar findings were concluded. These findings were noted to be correlated with the higher cost of capital often found in less developed markets, leading to lower gearing of firms.

(15)

14

Table #2: Descriptive Statistics- Emerging Markets Universe

Variables Standard Deviation Mean Minimum 25th Percentile Median 75th Percentile Maximum A. Dependent Variables

One Year Earnings Growth 2.3502 0.0614 -23.7608 -0.6349 0.0307 0.4396 20.6956

Three Years Earnings Growth 0.4245 0.1537 -0.7012 -0.1191 0.0830 0.3161 2.7279

Five Years Earnings Growth 0.2625 0.1078 -0.5227 -0.0695 0.0774 0.2406 1.3838

B. Independent Variables

Dividend Payout Ratio 0.7314 0.4459 -2.4894 0.1177 0.2960 0.6062 7.1481

Firm Size 2.5492 22.4458 16.4625 20.5095 22.1737 24.4381 29.1246

Return on Assets 0.1385 0.0119 -1.1203 -0.0032 0.0302 0.0732 0.3258

Earnings Yield 0.3607 0.0341 -3.5566 -0.0068 0.0451 0.1145 1.0629

Financial Leverage 0.1844 0.2498 0.0003 0.0964 0.2259 0.3682 0.9413

One Year Asset Growth 0.3519 0.1345 -0.5771 -0.0255 0.0651 0.1938 3.1697

Three Years Asset Growth 0.2169 0.1144 -0.4056 -0.0015 0.0738 0.1813 1.4408

Five Years Asset Growth 0.1716 0.1061 -0.3233 0.008 0.0776 0.1702 1.0143

Past One Year Earnings Growth 2.2680 0.0654 -18.7330 -0.6775 0.0442 0.2066 19.3995

Past Three Years Earnings Growth 0.4215 0.1496 -0.7319 -0.2402 0.1124 0.1353 2.3463

Past Five Years Earnings Growth 0.2688 0.1063 -0.5805 -0.1939 0.0785 0.0747 1.0951

(16)

15

Table #3: Descriptive Statistics- Frontier Market Universe

Variable Standard Deviation Mean Minimum 25th Percentile Median 75th Percentile Maximum A. Dependent Variables

One Year Earnings Growth 2.8925 0.0119 -25.1976 -0.4237 0.0309 0.3849 19.2626

Three Years Earnings Growth 0.4634 0.1322 -0.7278 -0.1099 0.0755 0.2805 2.7653

Five Years Earnings Growth 0.2788 0.0894 -0.5676 -0.074 0.0665 0.2152 1.4067

B. Independent Variables

Dividend Payout Ratio 0.8234 0.5965 -1.3077 0.2279 0.4656 0.7281 10.0568

Firm Size 2.7187 24.0756 16.3867 22.7045 24.4373 25.8693 30.3928

Return on Assets 0.0743 0.0470 -0.2723 0.0101 0.0383 0.0843 0.2984

Earnings Yield 0.5616 0.1573 -2.8571 0.0002 0.0884 0.2001 5.5995

Financial Leverage 0.1911 0.2671 0.0008 0.1063 0.2389 0.3986 0.8544

One Year Asset Growth 0.2594 0.1139 -0.4078 -0.0284 0.0601 0.1912 1.7848

Three Years Asset Growth 0.1676 0.1009 -0.2781 -0.0052 0.0725 0.1744 0.9211

Five Years Asset Growth 0.1349 0.0963 -0.2261 0.0073 0.0787 0.1656 0.6657

Past One Year Earnings Growth 2.6194 0.0144 -18.9801 -0.4885 0.0449 0.1687 22.6189

Past Three Years Earnings Growth 0.4517 0.139 -0.7345 -0.2192 0.0703 0.1235 2.6666

Past Five Years Earnings Growth 0.2761 0.0189 -0.5845 -0.1771 0.0624 0.0798 1.3083

(17)

16

3.3. Diagnostics

Due to the cross-sectional time series nature of the data, and the significant number of incomplete panel sets, using the appropriate regression approach plays a key role in the reliability of the results. Therefore, a few diagnostic tests must be executed to determine the econometric characteristics of the data and the most appropriate way to treat them.

The analysis attempts to examine the relationship between the predictor variable, Dividend Payout, and the outcome variable of Future Earnings Growth as it varies over time across different entities. As each firm is unique, its individual unobserved characteristics may unevenly influence the relationship examined. In order to control for the effects of these time-invariant characteristics, intuition dictates the Fixed Effects model should be utilised. To further verify this approach both the Breusch Pegan LM test, and the Hausman test are conducted with the null hypothesis that the unique errors are correlated with the regressed variables, and thus the Random Effects model is appropriate. Based on our results, we reject the null hypothesis with a statistically significant p-value of 0.000 for both tests, concluding the differences in coefficients are systematic and the Fixed Effects model should indeed be used.

Next, the skewness and kurtosis of the data is further examined. In order to assess the normality characteristics of data’s error term distribution, the Jarque-Bera asymptotic test for normality is executed. As is expected from our data, we reject the null hypothesis with a significant p-value at the 1% level, and conclude the errors are asymptotic.

Attending to potential issues of multicollinearity is a fundamental step for every examination due to the its risk of inflating the error terms and R-squared. To ensure no such issues were present in the data, a Pearson pairwise cross correlation matrix was used to examine that no two variables in the regression have a correlation higher than the ‘rule-of-thumb’ amongst financial scholars of 0.3-0.5. Furthermore, the variance inflation test was used. Both leading to the same conclusion that the data does not exhibit any risk for potential multicollinearity.

When analysing macro panels with large time series, the risk of serial correlation is abnormally high. This holds the potential of an inflated R-squared and deflated standard errors. After running the Wooldridge test for autocorrelation in panel data with the null hypothesis of no autocorrelation present, we reject the null hypothesis with a statistically significant p-value at the 1% level.

Furthermore, one of the greatest risk for any data analysis is the effect of a non-constant distribution of the error terms on the reliability of its results. In order to detect such groupwise heteroskedasticity of the residuals the modified Wald test was used as the appropriate approach for a Fixed Effects model. Due to a statistically significant p-value of 0.000, we reject the test’s null hypothesis of homoskedasticity, and therefore conclude the data is heteroskedastic.

(18)

17

4. Results

4.1. Univariate Regression

Fama and French (2000) noted strong evidence in favour of mean reversion in earnings and profitability. In order to account for that possibility in this paper, a univariate regression was executed. Using a Pearson’s pairwise product-moment correlation matrix, the correlations between the variables of Payout, one- three- and five- year Earnings Growth, and one- three- and five- year Past Earnings Growth are examined and illustrated in Table 4 and Table 5.

Table 4 and Table 5 present positive correlation between Payout, and three- and five- year ahead Earnings Growth- significant at the 1% level for both samples. Furthermore, an additional, stronger negative correlation is exhibited between Payout and Past Earnings Growth on the three- and five-year lag variables, further supporting these findings. Flint et al (2010), suggests these observations can be associated with mean reversion and sticky dividend behaviour. While a high dividend payout ratio can be seen as a sign of increased future earnings growth, as earnings rise, dividends might have difficulties increasing at similar pace, and thus reducing the payout amount in comparison to net earnings over time. Arnott and Asness (2003) associated these findings with the volatility of earnings compared to the relatively static, “sticky”, attribution of dividend payments which due to adverse effect of their reduction, cannot move parallel with earnings- thus expressed in the changing Payout ratio.

(19)

18

Table #4: Correlation Matrix- Emerging Markets Universe

Payout PEG (-1,0) PEG (-3,0) PEG (-5,0) EG (0,1) EG (0,3) EG (0,5)

Payout 1.000 PEG (-1,0) 0.010 1.000 PEG (-3,0) -0.229* 0.138* 1.000 PEG (-5,0) -0.334* 0.128* 0.505* 1.000 EG (0,1) 0.080* 0.014* -0.031* -0.039* 1.000 EG (0,3) 0.217* -0.071* -0.229* -0.273* 0.263* 1.000 EG (0,5) 0.197* -0.068* -0.225* -0.265* 0.186* 0.652* 1.000

Table 4 presents Pearson’s Pairwise Correlation Matrix to the Emerging market Universe. Illustrating correlation across the Payout ratio, past earnings growth across the three observation horizons- denoted as PEG(-h,0), and future earnings growth across the three observations horizons- denoted as EG(0,h). * shows significance at the 0.01 level.

Table #5: Correlation Matrix- Frontier Market Universe

Payout PEG (-1,0) PEG (-3,0) PEG (-5,0) EG (0,1) EG (0,3) EG (0,5)

Payout 1.000 PEG (-1,0) -0.042* 1.000 PEG (-3,0) -0.309* 0.181* 1.000 PEG (-5,0) -0.297* 0.150* 0.524* 1.000 EG (0,1) 0.005 0.027* -0.009 -0.024 1.000 EG (0,3) 0.212* -0.056* -0.199* -0.268* 0.297* 1.000 EG (0,5) 0.221* -0.053* -0.242* -0.267* 0.181* 0.673* 1.000

(20)

19

4.2. Multivariant regression

Following the diagnostic characteristics of the data, three potential regression approaches are viable: Feasible Generalised Least Squares (GLS) approach, the Driscoll-Kraay standard errors Fixed Effects model, or the Fama MacBeth 1973 procedure, as seen in Zhou and Ruland (2006). All three of these methodologies allow for the presence of heteroscedastic cross-sectional standard errors with a degree of serial correlation. However, due to large number of panels in the examination, many of which are incomplete, the Fama MacBeth 1973 procedure is limited in its analysis and tends to over-reject the significance of observations. Therefore, the investigations primary results will be presented using the Driscoll-Kraay standard errors Fixed Effects model and are presented in Table 6. Further robustness confirmations of the result’s statistical significance are presented in Table 7 using the GLS model. Upon analysis of the results, we fail to reject the examined hypothesis of this paper, and conclude that the correlation between earnings growth and dividend payout found in previous studies of developed markets, does indeed also exist in both Emerging and Frontier markets alike.

By examining the adjusted R-squared for the three estimations across both universes, a good proxy of the reliability of the models can be deduced. In both cases, despite having highly significant independent variable coefficients, the results exhibit moderately weak power explaining variation in earnings growth based on the current model for the one-year time horizon. This is indicated by the R-squared value of 6.95% and 4.80% as seen in columns (1) and (4) in Table 6. However, when the time horizon is extended to three- and five- year period, the model outperforms those found in other literature of developed markets. Based on the results presented in columns (2) and (3), in Emerging markets 39.69% and 45.60% of the variation of the three- and five-year earnings growth can be explained by the model. The reliability of the regression in capturing the dependent variable movements presents similar reliability for Frontier markets as shown in columns (5) and (6), where 42.27% and 42.51% of the variation can be explained for the three- and five- year time horizons.

(21)

20 An analysis of the remaining control variables finds that in all cases the expected sign and patterns were demonstrated across all three time horizons. From this it can be concluded that the effects these control variables account for in predicting future earnings growth are similar across all market types.

The variables of Size, Earnings Yield, and Return on Assets are all negative and significant across all three horizons for both markets. These results illustrate opposite relationship between size and earnings growth, showing that larger firms would have slower relative earnings growth in comparison to smaller firms, assuming all other variables remain constant. The negative size coefficients are larger in our regression than in those of other studies of developed markets (-0.138, and -0.0884 for Emerging markets, and -0.123 and -0.0813 for Frontier markets, in comparison to -0.012 and -0.007 in US markets), suggesting that the trend is stronger in Emerging and Frontier markets, and the size of the firm will have a more significant adverse effect on earnings growth in our examination than in developed markets. Furthermore, negative return on asset and earnings yield coefficients are in line with the findings of Fama and French (2000), which underlined the inverse correlation between current profitability and long-term earnings growth as was further elaborated on in Section 2 of this paper. The Return On Asset coefficients noted in columns (2) (3) and (5) (6) are higher than the corresponding values in the US market investigations of -0.974 and -0.646 for the three- and five-year horizons respectively. This can be further associated with the stronger level of mean reversion of earnings which was previously noted across the Emerging and Frontier market samples. The coefficients noted for Earnings Yield across developed and Emerging universes vary insignificantly, by less than 0.05. However, the strength of the correlation between earnings yield and future earnings growth is significantly diminished by more than 85% across the three- and five- year horizon in the Frontier markets. This can be associated with increased volatility of prices found in Frontier markets, thus influencing the coefficient of this ratio (Seth & Singhnia, 2019).

Similar to the trends noted in Zhou and Ruland (2006) and Flit et al. (2010), the variables of Leverage and Asset Growth are both positive and significant at the 1% level. These findings indicate that firms with higher debt to assets ratios would yield higher earnings growth in the three- and five-year horizon. Table 6 further illustrates that in the context of Emerging and Frontier markets, higher leverage ratio would have a more significant positive effect on earnings growth than in developed markets. This is denoted by the values of 0.175, and 0.136 for Emerging markets in columns (2)-(3), and 0.151, and 0.123 for Frontier markets in columns (5)-(6), in comparison to those seen in US markets of 0.065 and 0.058 (Zhou & Ruland, 2006). This behaviour is further discussed by Alfaro et al (2017), in their examination of the advantages of debt in Emerging and Frontier markets. They found that firm size plays an important role in the efficient utilisation of debt into future earnings- noting that due to the overall smaller size of firms found in developing markets, leverage appears to be more beneficial. The mean reversion noted in Table 4 and Table 5 is seen to further materialise in the multivariant regressions through a negative coefficient across the three- and five-year horizons, for both universes. Similar to the results noted in Vivian (2006), and is further supported by the low correlation found in columns (1) and (4), one-year earnings growth and past one-year earnings growth are only correlated with one another, and do not have a significant statistical correlation across other horizon measures. Furthermore, in Table 4 and Table 5 the strength of the mean reversion coefficients, identified by a the negative Pearson’s pairwise correlation across earnings growth and past earnings growth, appeared of similar magnitude across both samples for the three- and five- year horizons. However, as is seen in columns (5) and (6) in comparison to (2) and (3), the significance of the past earnings growth variables are larger by nearly the factor of two in Frontier markets than those found in Emerging markets. From these results we can deduce, that while mean reversion occurs, and is equally statistically significant across both universes, it is more responsible for future earnings growth movements in Frontier markets than in Emerging markets.

(22)

21 Furthermore, the results remain true and robust when examined across a GLS regression approach Table 7, which found similar levels of significance to nearly all coefficients.

The findings of this paper are further in line with other research in the field, both of that has been conducted using firm level analysis such as Zhou and Ruland (2006), or Flint et al. (2010), but also aggregate examinations as seen by Arnott and Assness (2003) or Parker (2005). We find strong, statistically significant correlation between Payout and Future Earnings Growth across all time horizons, and eliminated the possibility of mean reversion being fully responsible for this pattern.

Table #6: Future Earnings Growth as a Function of Dividend Payout using

Driscoll-Kraay Fixed Effect model for Emerging and Frontier Markets

Emerging Markets Universe Frontier Markets Universe

(1) (2) (3) (4) (5) (6)

INDEPENDENT VARIABLES One Year

Earnings Growth Three Year Earnings Growth Five Year Earnings Growth One Year Earnings Growth Three Year Earnings Growth Five Year Earnings Growth Dividend Payout 0.331*** 0.152*** 0.0764*** -0.0986 0.0711*** 0.0354*** (0.0594) (0.0101) (0.00453) (0.108) (0.00886) (0.00724) Firm Size -0.216*** -0.138*** -0.0884*** -0.120** -0.123*** -0.0813*** (0.0501) (0.00985) (0.00372) (0.0442) (0.0154) (0.00666) Return on Assets -3.293*** -1.916*** -1.224*** -5.391*** -2.219*** -1.333*** (0.933) (0.172) (0.102) (0.608) (0.136) (0.163) Earnings Yield -1.206*** -0.701*** -0.402*** -0.249* -0.0891*** -0.0626*** (0.215) (0.0372) (0.0252) (0.120) (0.0160) (0.00603) Leverage Ratio -0.0898 0.175*** 0.136*** -1.229* 0.151*** 0.123** (0.202) (0.0339) (0.0102) (0.318) (0.0455) (0.0421) Asset Growth 1.430*** 0.701*** 0.643*** 1.298*** 0.846*** 0.586*** (0.109) (0.0475) (0.0354) (0.179) (0.0627) (0.0551)

Past One Year Earnings Growth -0.0632*** -0.000753 0.000947 -0.101*** -0.000791 -0.00201

(0.0102) (0.00154) (0.000876) (0.0229) (0.00713) (0.00527)

Past Three Years Earnings Growth -0.0123 -0.0586*** -0.0384** -0.0230 -0.121*** -0.0615***

(0.0410) (0.0123) (0.0135) (0.0679) (0.0247) (0.0191)

Past Five Years Earnings Growth -0.0699 -0.118*** -0.0928*** -0.0938 -0.240*** -0.137***

(0.0516) (0.0202) (0.00670) (0.245) (0.0519) (0.0179) Constant 5.136*** 3.347*** 2.144*** 3.621*** 3.138*** 2.050*** (1.193) (0.226) (0.0886) (1.205) (0.369) (0.150) Driscoll-Kraay adjusted R2 _0.0695 _0.3969 _0.4560 _0.0480 _0.4227 _0.4251 Observations 20,759 14,191 10,174 4,219 2,897 1,949 Number of Firms 4,133 3,176 2,589 938 765 633

Table 6 presents the regression results for the equation: 𝐸𝐺𝑖𝑡ℎ= 𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+ 𝛽2𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝐸𝑃𝑖𝑡+

𝛽5𝐿𝐸𝑉𝑖𝑡+ 𝛽6𝐴𝐺𝑖𝑡_ℎ+ 𝛽7𝑃𝐸𝐺𝑖𝑡_−1,3,5+ 𝜖𝑖𝑡. The model used is the Fixed Effects Model with Driscoll-Kraay standard errors-

(23)

22

Table #7: Future Earnings Growth as a Function of Dividend Payout using the

Generalised Least Squares model for Emerging and Frontier Markets

(1) (2) (3) (4) (5) (6)

Earnings Growth Three Year Earnings Growth Five Year Earnings Growth One Year Earnings Growth Three Year Earnings Growth Five Year Earnings Growth Dividend Payout 0.254*** 0.114*** 0.0722*** -0.169 0.0534*** 0.0204*** (0.0223) (0.00482) (0.00360) (0.0374) (0.00766) (0.00679) Firm Size 0.00955* -0.00307*** -0.00281*** 0.0252** -0.00144 -0.00150 (0.00555) (0.00110) (0.000793) (0.0116) (0.00227) (0.00176) Return on Assets -1.278*** -1.115*** -0.835*** -1.059* -0.841*** -0.478*** (0.300) (0.0581) (0.0415) (0.581) (0.113) (0.0880) Earnings Yield 0.00990 -0.142*** -0.101*** 0.0583 0.0181* 0.0109 (0.101) (0.0201) (0.0140) (0.0523) (0.00963) (0.00675) Leverage Ratio -0.192** 0.0576*** 0.0387*** -0.318* 0.0748** 0.0765*** (0.0830) (0.0166) (0.0120) (0.170) (0.0338) (0.0266) Asset Growth 1.479*** 0.943*** 0.912*** 1.448*** 1.002*** 0.940*** (0.0586) (0.0200) (0.0181) (0.144) (0.0451) (0.0441)

Past One Year Earnings Growth -0.00147 -0.00127 0.000313 -0.0162 -0.00581 -0.00647*

(0.00764) (0.00159) (0.00112) (0.0221) (0.00460) (0.00344)

Past Three Years Earnings Growth -0.0222 -0.0844*** -0.0540*** -0.0307 -0.0705*** -0.0702***

(0.0404) (0.00816) (0.00592) (0.0913) (0.0173) (0.0153)

Past Five Years Earnings Growth -0.273*** -0.176*** -0.0999*** -0.313** -0.232*** -0.155***

(0.0611) (0.0122) (0.00862) (0.133) (0.0278) (0.0220)

Constant -0.301** 0.120*** 0.0943*** -0.169*** 0.0534*** 0.0204***

(0.137) (0.0273) (0.0197) (0.0374) (0.00766) (0.00679)

Observations 20,759 14,191 10,174 4,219 2,897 1,949

Number of Firms 4,133 3,176 2,589 938 765 633

Table 7 presents the results of the robustness check of Table 6 using the GLS model for the regression equation: 𝐸𝐺𝑖𝑡ℎ=

𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+ 𝛽2𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝐸𝑃𝑖𝑡+ 𝛽5𝐿𝐸𝑉𝑖𝑡+ 𝛽6𝐴𝐺𝑖𝑡ℎ+ 𝛽7𝑃𝐸𝐺𝑖𝑡−1,3,5+ 𝜖𝑖𝑡. Columns (1)-(3) illustrate

(24)

23

5. Discussion

5.1. Robustness- Alternative Measure of Earnings

Equation (1) presented in Section 3 of this paper, finds a positive relationship between dividend payout and net earnings growth across different time horizons. However, these results may be due to the examination’s choice of earnings measure. In the original mode, Net Earnings also often referred to as the “Bottom Line Earnings” are used. This measure is defined as the net value remaining for distribution, or reinvestment after interest, depreciation, taxes, operating expenses, and all other recurring or one-time expenses have all been deducted from income. The amount remaining is the net amount available to the firm for distribution of dividend, or reinvestment. In order to further test the sensitivity of the trend uncovered, equation (1) is modified, and will now implement an alternative measure of earnings: EBITDA.

The revised model is now presented as:

𝐸𝑎𝑙𝑡.𝐺𝑖𝑡ℎ = 𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+ 𝛽2𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝐸𝑃𝑖𝑡+ 𝛽5𝐿𝐸𝑉𝑖𝑡+ 𝛽6𝐴𝐺𝑖𝑡ℎ+ 𝛽7𝑃𝐸𝑎𝑙𝑡.𝐺𝑖𝑡−1,3,5+ 𝜖𝑖𝑡 (2)

As Table 8 demonstrates, when substituting Earnings Before Interest Tax Depreciation and Amortisation (EBITDA) into our model instead of the previous measure of Net Earnings, the pattern remains robust across both universes. These results are in line with those found in the developed markets by Zhou and Ruland’s (2006) examination, which notes no significant changes between the two measures are found. While the polarity and trend of all the variables in the regression remained the same across both measures, the use of the alternative measure of earnings influenced the strength of some of the coefficients noted. Particularly, the one of interest- Payout, and the lagged earnings growth. As can be seen in columns (1)-(3), the Emerging market sample behaves more similar to Zhou and Ruland (2006) findings, as while minor changes can be seen in the coefficients (no greater than 0.02 in variation) their magnitude and significance remains constant across both measures of earning. However, as observed in columns (4)-(6), the use of EBITDA significantly reduces the predictive power of the Dividend Payout- which is now significant at the 10% level instead of the 1% level across the one- and three-year horizon.

(25)

(26)

25

Table #8: Alternative Measure of Earning- Future EBITDA Growth as a Function of

Dividend Payout using Driscoll-Kraay Fixed Effect model for Emerging and Frontier

Markets

(1) (2) (3) (4) (5) (6) INDEPENDENT VARIABLES One-Year EBITDA Growth Three-Years EBITDA Growth Five-Years EBITDA Growth One-Year EBITDA Growth Three-Years EBITDA Growth Five-Years EBITDA Growth Dividend Payout 0.432*** 0.175*** 0.0998*** 0.173* 0.0956* 0.0174 (0.0373) (0.0106) (0.00605) (0.0700) (0.0462) (0.0209) Firm Size -0.0805*** -0.0529*** -0.0316*** -0.0939*** -0.0487*** -0.0320*** (0.0108) (0.00630) (0.00497) (0.0161) (0.00608) (0.00737) Return on Assets -1.265*** -0.616*** -0.423*** 0.182 0.215*** 0.112** (0.304) (0.145) (0.0977) (0.175) (0.0650) (0.0452) Earnings Yield -0.0913 -0.119*** -0.0494*** -1.085** -0.794*** -0.531*** (0.0662) (0.0162) (0.0134) (0.428) (0.0767) (0.130) Leverage Ratio 0.358*** 0.268*** 0.165*** -0.0658* -0.00600 -0.0297*** (0.0706) (0.0324) (0.0174) (0.0333) (0.00724) (0.00765) Asset Growth 0.821*** 0.839*** 0.844*** 0.586*** 0.674*** 0.536*** (0.0478) (0.0331) (0.0425) (0.0431) (0.0334) (0.0712) Past One-year EBITDA Growth -0.0528*** -0.0129*** -0.0114*** -0.0315 -0.0279*** -0.0155** (0.0135) (0.00148) (0.00183) (0.0394) (0.00585) (0.00599) Past Three-Years EBITDA Growth -0.184*** -0.131*** -0.0769*** -0.154 -0.106*** -0.0646** (0.0391) (0.0121) (0.0155) (0.0920) (0.0253) (0.0266) Past Five-Years EBITDA Growth -0.300*** -0.192*** -0.155*** -0.485*** -0.253*** -0.132*** (0.0821) (0.0371) (0.0118) (0.130) (0.0557) (0.0406) Constant 1.816*** 1.203*** 0.716*** 2.384*** 1.211*** 0.827*** (0.256) (0.144) (0.113) (0.402) (0.145) (0.164) Driscoll-Kraay 𝑅2 _0.0954 _0.3443 _0.3960 _0.0928 _0.3498 _0.3302 Observations 22,653 16,363 11,811 3,416 2,303 1,418 Number of Firms 4,201 3,396 2,768 825 680 486

Table 8 illustrates the results for equation (2) with the alternative measure of earnings as EBITDA: using the Fixed Effects model with Driscoll-Kraay standard errors presented in the parentheses. The revised model is therefore: 𝐸𝑎𝑙𝑡.𝐺𝑖𝑡ℎ = 𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+

𝛽₂𝑆𝑖𝑧𝑒_𝑖𝑡+ 𝛽₃𝑅𝑂𝐴_𝑖𝑡+ 𝛽₄𝐸𝑃_𝑖𝑡+ 𝛽₅𝐿𝐸𝑉_𝑖𝑡+ 𝛽₆𝐴𝐺_𝑖𝑡_ℎ+ 𝛽₇𝑃𝐸_{𝑎𝑙𝑡.}𝐺_𝑖𝑡_−1,3,5+ 𝜖_𝑖𝑡. Columns (1)-(3) illustrate the results of the model for the

(27)

26

5.2. Potential Explanations

5.2.1. Cash Flow Signalling Hypothesis

When similar primary results as those identified in this paper were found by Arnott and Asness (2003), they appeared as a surprise in academia due to their contradiction to conventional wisdom. Arnott and Asness (2003), Gwilyn et al (2006), and Zhou and Ruland (2006), offer a series of potential explanation of the phenomena. While some have proven to be correct when explored later through firm level examinations of international markets (Aivazian & Booth, 2001), other had difficulty to hold true (Flint et al., 2010). This section will test these potential explanations across our data samples to see whether they can provide a deeper understanding of the trends discovered.

Based on the findings of the sensitivity tests conducted, primary evidence for Cash Flow Signalling Hypothesis emerge. The Cash Flow Signalling Hypothesis, noted by Lintner (1956), suggests that due to informational asymmetry between managers and shareholders, managers can use dividends to signal their assurance regarding the firm’s future cash flows.

Increase in dividends would signal to investors of the managers’ confidence in being able to consistently provide these payments in perpetuity, and thus indicate manager’s insider knowledge of future earnings growth. This signal is valued by investors as it is a “costly signal” which cannot be mimicked by weaker firms, thus, the message it sends to the market can be trusted. Share prices tend to react more to a dividend decrease than they do to dividend stationarity or increases, hence dividends are known to be “sticky downwards” (Lintner 1956). An increase in dividend signals to the market of the managers long term positive outlook and willingness to commit to the added payments in perpetuity, as the manager knows that once dividend payments are increased it will be difficult to bring them back down without shareholder’s pullback. Furthermore, dividends are a costly signal also due to reasons such as the increased costs of issuing additional shares in the future, and the opportunity cost undertaken by paying the dividends instead of further reinvestment, which may later need to be financed through debt (Miller Rock 1985). Thus, increase in dividends is often time seen as the best way management can assure shareholders of future growth, and reduce the informational asymmetry gap.

To test this hypothesis the following regression was utilised:

𝐸𝐺𝑖𝑡_ℎ = 𝛼 + 𝛽1∆𝐷𝑖𝑣𝑡_−1,3+ 𝛽2𝑅𝑂𝐴𝑖𝑡+ 𝛽3𝐴𝐺𝑖𝑡_1,3,5+ 𝛽4𝑃𝐸𝐺𝑖𝑡_−1,3,5+ 𝜖𝑖𝑡 (3)

(28)

27 As can be seen in Table 9 and Table 10, columns (1) (2) (3), illustrate the signalling hypothesis across one-year past dividend growth for the two universes. The results presented are either too small to be considered, or are statistically insignificant across all time horizons, and thus the hypothesis is rejected for the one year past dividend growth horizon. These results conclude, that a single year increase in dividends does not necessarily provide additional information to investors regarding future earnings growth in the context of either Emerging or Frontier markets. These findings are in line with those found in Flint et al (2010), whom while successfully demonstrated a correlation between payout and earnings growth, failed to hold the same correlation with dividend growth on the one-year horizon. An added modification to the test which this study offers, is examining signalling strength of increased dividend payments across the three-year horizon as well, noted in columns (4) (5) (6). In this case, the relationship is as expected- positive and significant across all three-time horizons for Emerging markets, however remains insignificant for Frontier markets. From these results we can deduce that in the case of Emerging markets, when controlling for Mean Reversion in earnings, Return On Asset, and Asset Growth, dividend increases do provide valuable additional information useful for investors regarding future earnings growth. These findings are in line with the theory noted by Avaizian and Booth (2001), which explained that due to high volatility in Emerging markets, dividend signalling strength – expressed by short term dividend increases followed by profitability measure increases – appears weak in the extreme short time periods, however still found evidence to point to its existence across longer horizons. In the case of Frontier markets, increased volatility of earnings might be the reason for the remaining weakness of dividend signalling across the three-year horizon.

(29)

28

Table #9: Testing for Signalling Hypothesis across One- Three- and Five- Year

Earnings Growth as a function of One- and Three-Year Dividend Growth – Emerging

Markets Universe

(1) (2) (3) (4) (5) (6)

Earnings Growth Three Year Earnings Growth Five Year Earnings Growth One Year Earnings Growth Three Year Earnings Growth Five Year Earnings Growth

Past One- Year Dividend Growth -0.0274 0.00288 -0.00136

(0.0104) (0.00200) (0.00186)

Past Three-Years Dividend Growth 0.145* 0.0412*** 0.0338***

(0.0931) (0.0111) (0.00774)

ROA -7.702*** -3.468*** -2.076*** -7.263*** -3.416*** -1.993***

(1.071) (0.175) (0.0605) (1.050) (0.175) (0.0537)

One-Year Asset Growth 1.150*** -0.0783*** -0.0165* 1.279*** -0.0530*** -0.0203**

(0.123) (0.00995) (0.00754) (0.136) (0.0111) (0.00721)

Three-Year Asset Growth 0.895** 0.601*** -0.0474* 1.056** 0.559*** -0.0704**

(0.376) (0.0791) (0.0249) (0.397) (0.0584) (0.0241)

Five-Year Asset Growth 2.000*** 0.744*** 0.959*** 1.876*** 0.781*** 0.968***

(0.363) (0.0823) (0.0470) (0.405) (0.0641) (0.0442)

Past One-Year Earnings Growth -0.111*** -0.0123*** -0.00587** -0.102*** -0.00982*** -0.00500**

(0.0207) (0.00229) (0.00206) (0.0137) (0.00202) (0.00174)

Past Three-Years Earnings Growth -0.190*** -0.114*** -0.0644*** -0.289*** -0.120*** -0.0800***

(0.0584) (0.0202) (0.0187) (0.0590) (0.0225) (0.0175)

Past Five-Years Earnings Growth -0.307** -0.168*** -0.133*** -0.433** -0.198*** -0.155***

(0.119) (0.0246) (0.0104) (0.151) (0.0303) (0.00929)

Constant 0.329** 0.254*** 0.156*** 0.264* 0.243*** 0.149***

(0.138) (0.0128) (0.00781) (0.126) (0.00951) (0.00819)

Observations 12,700 11,210 10,969 11,754 10,363 10,155

Number of groups 2,985 2,719 2,662 2,821 2,557 2,516

Table 9 illustrates the results for the Emerging market universe for equation (3) using the Fixed Effects model with Driscoll-Kraay Standard errors- illustrated in the parentheses. The revised model is therefore: 𝐸𝐺𝑖𝑡ℎ = 𝛼 + 𝛽1∆𝐷𝑖𝑣𝑡−1,3+ 𝛽2𝑅𝑂𝐴𝑖𝑡+ 𝛽3𝐴𝐺𝑖𝑡ℎ+ +𝛽4𝑃𝐸𝐺𝑖𝑡−1,3,5+

𝜖𝑖𝑡 Columns (1)-(3) illustrate the results of the model using past one year dividend growth, Columns (4)-(6) present the corresponding results

(30)

29

Table #10: Testing for Signalling Hypothesis across One- Three- and Five- Year

Earnings Growth as a function of One- and Three-Year Dividend Growth – Frontier

Markets Universe

(1) (2) (3) (4) (5) (6)

Earnings Growth Three Year Earnings Growth Five Year Earnings Growth One Year Earnings Growth Three Year Earnings Growth Five Year Earnings Growth

Past One- Year Dividend Growth 0.00239*** 0.000401*** -0.000329

(0.000512) (0.000938) (0.000120)

Past Three-Years Dividend Growth 0.0230 -0.000168 -0.00401*

(0.0134) (0.00420) (0.00210)

ROA -6.610*** -2.658*** -1.665*** -4.683*** -2.543*** -1.649***

(0.776) (0.190) (0.106) (0.770) (0.212) (0.0787)

One-Year Asset Growth 1.139*** -0.0177 -0.0209 1.134*** -0.0278 -0.0134

(0.141) (0.0314) (0.0180) (0.132) (0.0397) (0.0223)

Three-Year Asset Growth 1.151** 0.751*** -0.0367 1.234** 0.725*** -0.0163

(0.481) (0.198) (0.0502) (0.531) (0.180) (0.0350)

Five-Year Asset Growth 1.130* 0.456*** 0.776*** 0.216 0.366*** 0.729***

(0.521) (0.0971) (0.0934) (0.394) (0.0834) (0.0713)

Past One-Year Earnings Growth -0.0789*** -0.00690 -0.00512 -0.0758*** -0.000301 -0.00294

(0.0186) (0.00455) (0.00567) (0.0166) (0.00486) (0.00573)

Past Three-Years Earnings Growth -0.390** -0.139*** -0.0966*** -0.0491 -0.145*** -0.0830***

(0.159) (0.0241) (0.0225) (0.101) (0.0331) (0.0217)

Past Five-Years Earnings Growth 0.361 -0.230*** -0.147*** -0.239* -0.246*** -0.138***

(0.245) (0.0535) (0.0191) (0.110) (0.0604) (0.0287)

Constant 0.224*** 0.217*** 0.148*** 0.184*** 0.220*** 0.148***

(0.0595) (0.0130) (0.0129) (0.0558) (0.0130) (0.0137)

Observations 2,595 2,379 2,367 2,440 2,245 2,224

Number of groups 744 698 701 733 693 681

Table 10 illustrates the results for the Frontier market universe for equation (3) using the Fixed Effects model with Driscoll-Kraay Standard errors- illustrated in the parentheses. The revised model is therefore: 𝐸𝐺𝑖𝑡ℎ = 𝛼 + 𝛽1∆𝐷𝑖𝑣𝑡−1,3+ 𝛽2𝑅𝑂𝐴𝑖𝑡+ 𝛽3𝐴𝐺𝑖𝑡ℎ+ +𝛽4𝑃𝐸𝐺𝑖𝑡−1,3,5+

𝜖𝑖𝑡 Columns (1)-(3) illustrate the results of the model using past one year dividend growth, Columns (4)-(6) present the corresponding results

(31)

30

5.2.2. Free Cash Flow Hypothesis

Arnott and Asness (2003) bring forth the possibility of inefficient empire building as an additional potential explanation to the low payout low earnings growth relationship noted in this study. This idea is consistent with the Free Cash Flow Hypothesis, which explains, that due to factors such as: managerial compensation schemes, tax consideration, local legislations, or desire to empire build, when a firm has abundant free cash flows managers hold the risk of sub-optimally investing in potentially negative NPV projects (Murphy 1985). Such inefficient, sub-optimal investments; while reducing potential dividend payments to shareholders, simultaneously also create the foundations for a negative effect on earnings growth in the future. This theory has been further tested by Lang and Litzenberg (1989), as they found discrepancies in share price movements as a reaction to dividend increases, based on a firm’s level of profitable potential investments. In their study they noted a more positive market reaction to increased dividends for a firm with low potential investment opportunities, compared to a lower reaction in the case where dividends were raised despite abundant positive NPV projects available.

In order to test this hypothesis, a firms positive investment opportunity is measured using Tobin’s-Q. Tobin’s-Q, measured as the firms market value of equity plus the book value of total debt, over book value of total assets. This measure is often used as a proxy for a firm’s profitable growth opportunities by assessing a company’s projected benefit from an asset- as it is priced based on the firms market value of equity, versus the assets replacement cost. If Tobin’s Q is lower than one, this indicates a firm with poor investment opportunities, and thus overinvestment poses a larger risk (Lang & Litzenberg, 1989). A firm with Tobin’s Q above one, suggests the opposite to be true. Based on these assumptions, and the assumption that the market values firms efficiently, it can be expected that an interaction term between Tobin’s Q and Payout will be negative and significant. These results will support the theory that, if the Free Cash Flow Hypothesis holds, when growth opportunities are low (hence Tobin’s Q below one), the positive relationship between earnings growth and dividend payout will be stronger. Additionally, a positive Tobin’s Q coefficient will be further in line with the theory suggesting that positive growth opportunities will lead to higher earnings growth. The revised model is now presented as:

𝐸𝐺𝑖𝑡ℎ = 𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+ 𝛽2𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝐸𝑃𝑖𝑡+ 𝛽5𝐿𝐸𝑉𝑖𝑡+ 𝛽6𝐴𝐺𝑖𝑡ℎ+ 𝛽7𝑃𝐸𝐺𝑖𝑡−1,3,5+ 𝛽8𝑇𝑄𝑖𝑡

+ 𝛽9𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡∗ 𝑇𝑄𝑖𝑡+ 𝜖𝑖𝑡 (4)

(32)

31 other available alternatives. In case this assumption is held, a rejection of this theory can only be achieved through statistically significant coefficients with the opposing expected signs. However, as is more probable to be the case in this examination, the lack of strong statistical significance across the variables suggests that either there is insufficient data to conclude the hypothesis, or, that in the context of Emerging and Frontier markets the assumption of complete market efficiency must be significantly relaxed. Therefore, it can be concluded, that our data does not provide sufficient evidence to support Free Cash Flow Hypothesis as an explanation to the positive relationship between dividend payout and future earnings growth.

Table #11: Testing for Cash Flow Signalling Hypothesis using Tobin’s Q – Emerging

Markets Universe

(1) (2) (3)

INDEPENDENT VARIABLES One Year Earnings Growth Three Year Earnings Growth Five Year Earnings Growth

Dividend Payout 0.285*** 0.155*** 0.0771*** (0.0551) (0.0109) (0.00526) Firm Size -0.225*** -0.139*** -0.0900*** (0.0508) (0.0104) (0.00365) Return on Assets -3.286*** -1.933*** -1.236*** (0.931) (0.172) (0.105) Earnings Yield -1.271*** -0.697*** -0.403*** (0.223) (0.0375) (0.0241) Leverage Ratio -0.0737 0.176*** 0.137*** (0.206) (0.0341) (0.0102) Asset Growth 1.421*** 0.700*** 0.638*** (0.111) (0.0449) (0.0354)

Past One-Year Earnings Growth -0.0641*** 5.19e-05 0.000945

(0.0105) (0.00153) (0.000908)

Past Three-Years Earnings Growth -0.0124 -0.0585*** -0.0384**

(0.0413) (0.0123) (0.0135)

Past Five-Years Earnings Growth -0.0709 -0.117*** -0.0916***

(0.0538) (0.0201) (0.00671) Tobin’s Q -0.00199 0.000850** 0.000636*** (0.00267) (0.000334) (0.000171) Payout x Tobin’s Q -0.00835*** -0.000608* -0.000271* (0.00230) (0.000294) (0.000148) Constant 5.376*** 3.367*** 2.178*** (1.210) (0.241) (0.0883) Observations 20,396 13,933 9,991 Number of Firms 4,104 3,155 2,574

Table 11 illustrates the results for the Emerging market universe for equation (4) using the Fixed Effects model with Driscoll-Kraay Standard errors- illustrated in the parentheses. The regression is similar to that seen in (1), with the added variables of Tobin’s Q (TQ in equation) and the interaction variable between Tobin’s Q x Payout (Payout*TQ in equation). The revised model is therefore:

𝐸𝐺𝑖𝑡_ℎ = 𝛼 + 𝛽1𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡+ 𝛽2𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽3𝑅𝑂𝐴𝑖𝑡+ 𝛽4𝐸𝑃𝑖𝑡+ 𝛽5𝐿𝐸𝑉𝑖𝑡+ 𝛽6𝐴𝐺𝑖𝑡_ℎ+ 𝛽7𝑃𝐸𝐺𝑖𝑡−1,3,5+ 𝛽8𝑇𝑄𝑖𝑡+ 𝛽9𝑃𝑎𝑦𝑜𝑢𝑡𝑖𝑡∗ 𝑇𝑄𝑖𝑡+ 𝜖𝑖𝑡. The bottom section of the table presents the total observations used in each regression model, and the total

An Examination of the Relationship between Dividend Payout and Future Earnings Growth within Emerging and Frontier Markets.