“Why do financial services companies pay dividend? Evidence from Qatar Stock Exchange”

This study identifies the dividend policy determinants of banks and other financial institutions listed on Qatar Stock Exchange (QSE) for a period from 2009 to 2015 through studying the impact on eight factors on banks’ dividends per share. Three models were adopted to investigate the determinants of the dividend policy and the factors that affect a bank’s decision to pay out dividends. The findings indicate that the previous year’s dividends per share, earnings per share, cash flow per share, firm size and return on average equity are positively related to the current year’s dividends per share, as hypothesized. The study shows that the leverage position, bank’s life cycle and growth opportunities are negatively related to the dividend payment. The study also reveals that banks and financial institutions in Qatar do a bit of “earnings smoothing” when comparing the earnings figures with the cash flow.


INTRODUCTION
Dividend policy, along with its related decisions, remains one of the frequently researched areas in finance.Given that profit and shareholders' value maximization are the core economic and financial targets that companies seek to satisfy, dividend policy stays on the top of researches.With an attempt to maximize shareholders' wealth, financial managers seek to assess and trigger the optimal level of dividends that would entail satisfying shareholders, in which they have to tradeoff between payouts and retentions.Many propositions developed by numerous researchers to understand the payout policy have led to different and, in many cases, contradicting findings.Black (1976) referred to dividends as a "puzzle" to describe the uncertainty inherited in the divided policy, for which he argued that the more analysts look into the dividend policy issue, the more it gets complicated to generalize or comprehend its determinants.Black's (1976) dividends puzzle theory was supported by Bhattacharyya (2007) who adopted a different approach to understand the dividends theory through means of developing a model that links the retention ratio with executive management compensations and hypothesized a positive association between them in an equilibrium situation.Bhattacharyya (2007) found that dividends issue remains as an unsolved and complicated puzzle.DeAngelo and DeAngelo (2006) criticized Miller and Modigliani (1961) irrelevance theory of dividends payout to firms and investors' value and wealth, as they suggested that the theory resulted in restricting researchers' view about dividends payout, and, hence, they disagree with Black Investment Management and Financial Innovations, Volume 14, Issue 3, 2017 (1976) and Bhattacharyya (2007) for classifying dividends as an unsolved "puzzle".Extensive previous research works have been conducted to examine the dividend policy in developed countries, relatively a few similar studies have been used to understand emerging countries' payouts policies and its determinants.
Qatar, which was reclassified by Morgan Stanley Capital International (MSCI), Standard & Poor's (S&P) and Dow Jones Indices in 2014 from a frontier risky market to an emerging market, is one of the countries that one can hardly find any dividend policy studies conducted on.Given the recent reclassification, more of investors' attention is expected to be drawn on Qatar financial market behavior.Hence, this study intends to fill a gap with this respect.
The main objectives of the present empirical study are: 1. To apply Lintner's dividend model (1956) on banks and financial institutions listed on Qatar Stock Exchange (QSE) to examine the impact of lagged dividends and profitability on banks' dividend policy.2. To investigate the impact of cash flows per share on banks' dividends.3. To ascertain and examine the impact of other dividend policy determinants on the dividend payment.4. To examine the factors that would affect a firm's probability to pay out dividends.
The remaining sections of the present study are organized as follows: section 1 provides a literature review of previous similar empirical studies conducted on emerging markets generally and others conducted on the banking sector specifically.Section 2 presents a brief overview of Qatar's banking sector.Section 3 and 4 describe the variables and data used in the study.Section 5, 6, 7, 8 and 9 present the empirical findings and analyze the results.Final section provides a short summary and concludes the study.

PREVIOUS STUDIES
Although the controversial issue of dividend policy has been a major focus of many researchers who developed multiple models, theories and pursued numerous empirical studies to understand the payout decisions and dividends determinants, the key credit is directed to Lintner (1956).Lintner (1956) was the first to attempt to formulate a model for the potential dividend payment determinants, in which he conducted interviews with 28 companies' managers and found that the two main factors are the current year earnings and the lagged dividends payment.On the other hand, Miller and Modigliani (1961) theory indicates when perfect and complete market conditions are satisfied, in which there are no taxes or transaction or agency costs and all firms are classified into the same risk level, the dividend policy does not matter, as it does not result in any impact on the shareholders' wealth or firm value.The first examination of Lintner's model on emerging markets was done by Mookerjee (1992) who prepared an empirical study on Indian stock market and found that Lintner's model well explains the behavior of the studied companies' dividends.The study cov-ered a period of 33 years, from 1949 to 1981, in which firms were found to be over-focused on the essentiality of dividend payment, no matter how the performance of the company was, even if it needed external borrowing.Another study carried out on Bursa Malaysia by Nassir and Mohamed (1993) concluded that, as proposed by Lintner (1956), firms maintain a target payout ratio based on the long-term sustainable profitability.
From Middle East context, according to Al Yahyaee (2006) who studied the dividend policy of financial and non-financial firms in Oman, both types of firms are affected by profitability, firm size and business risk.However, the dividend policy of the nonfinancial sector is significantly affected by leverage, age and government ownership, while the financial firms are not affected by those factors.Moreover, Al-Malkawi (2007) who examined the behavior of all the firms listed on Amman Stock Exchange (Jordan) for a period from 1989 to 2000 agreed that size, age and profitability of firms have a significant positive relationship with the dividend policy.However, the study found that dividend policy is negatively related to leverage, a finding that is agreed by Al Kuwari (2009) who used a panel data of Gulf Co-operation Council (GCC) listed non-financial firms for a fiveyear period.Some empirical studies had a concentrated scope on the dividend policy determinants specifically in the banking sector.Of the first researchers who limited their study scope to the banking sector are Gupta and Walker (1975).They conducted their research on a sample of 980 banks during 1965-1968 and revealed that there is a positive association between dividends and current year earnings, earnings variation from one year to another, cumulative earnings and total assets growth.On the contrary, negative associations exist between dividends and liquidity position.Lee (2009) examined the dividend policy determinants of Korean banks from 1994 to 2005 and found that there exists a positive association between banks' dividend payout ratio and their size and profitability.Kinfe (2011) investigated the dividend payout determinants for a sample of six banks in Ethiopia from 2006 up to 2010 using Lintner's model, where the results showed that the dividend payout ratio is positively related to firm size, negatively related to liquidity, while is unaffected by growth, profitability or leverage.A similar study by Marfo-Yiadom and Agyei (2011) on sixteen banks in Ghana for the years 1999-2003 concluded that the dividend policy is significantly and positively affected by leverage, profitability, collateral capacity and changes in the dividend payment.However, maturity and growth were found to have a significant negative impact on the payout ratio.Maladjian and El Khoury (2014) who conducted a study to reveal the determinants of dividend payout ratio by examining four Lebanese banks listed in Beirut Stock Exchange concluded that the lagged dividend payment positively affects the dividend policy and is found to be the most important variable.The study also found that firm size and risk profile have a significant positive relationship with dividend payout, while the profitability and growth have a significant negative relationship.

HYPOTHESES DEVELOPMENT, DATA AND METHODOLOGY
The main objective of the analysis is to test the relationship b etween the dividends paid by banks and financial institutions and the selected independent variables.Following is the F-test hypothesis for the overall model: H0: There is no statistically significant relationship between the dividends payment and the lagged DPS, EPS, CFPS, ROAE, SIZE, AGE, REVG and LR.
H1: There is a statistically significant relationship between the dividends payment and at least one of the variables, the lagged DPS, EPS, CFPS, ROAE, SIZE, AGE, REVG and LR.
However, the significance of each of the independent variable will be tested separately.

DATA SOURCES
To examine the dividend policy determinants of Qatar's banks and other financial institutions, a database is constructed using unbalanced panel method.
The unbalanced panel data were used, as all banks and financial institutions should meet one criterion: being listed on the stock exchange and having published annual reports for the selected period.One bank failed to meet this criterion and therefore has been excluded from the study.Many previous studies have used the unbalanced panel estimation, including Omet (2004)

RESEARCH MODELS
Based on the previous literatures, three research models are developed to test the hypotheses developed so far.The impact of a number of variables on the Dividends per Share (DPS) of Qatar's banks and financial institutions in a given year is investigated.Fixed Effect Pool Panel Regression is employed to identify the most significant determinants of dividend payments in the banks and financial institutions in Qatar and to test the developed hypotheses using the models shown below.

Model A
The first model that attempted to identify the dividend policy determinants is the Lintner model ( 1956).According to the model, each company has a target dividend payout ratio.The targeted dividend level ( ) .According to Lintner, dividend level should be gradually adjusted to the targeted payout ratio, and firms' management will curtail or eliminate dividends only in extreme circumstances or as a last resort.
Accordingly, the difference between two consecutive year's dividend levels is represented by: ( ) where t D -current period's dividends per share; 1 t D − -previous period's dividends per share; α - reluctance to curtail dividends; c -adjustment factor; e -error term.
By substituting equation (1) in equation ( 2), we get: By rearranging the equation, we get the current period's dividends per share as follows: ( ) We therefore arrive to Lintner's basic model, which is the regression format that will be studied separately (model A): A modified version of Lintner's basic model (modified model A) will be used in the present study, in which the EPS in equation ( 5) will be replaced by CFPS to examine the impact of cash on a firm's dividend paying capacity.

Model B
Model

Model C
Model C tries to identify the factors that would impact a company's likelihood to pay out dividends, and includes the same independent variables identified in model B. Given the purpose of model C, in which factors affecting a firm's payment of dividends are to be identified, and to account for the nature of the dependent variable (DPS), the logit model will be adopted.

Regression
Given the nature of the cross-sectional data collected for 7 years to conduct the study, and with an attempt to control of the unobserved variables that influence the dividend payments and its policies, the pooled panel regression was adopted.

Hausman test
Panel data regression model includes two variations, which are the fixed effects and the random effects.The Hausman test was run using the Stata tool to determine whether the fixed or random effects fit the study data.The following table (Table 2) shows the data.Note: * significant at 5%.
Therefore, the fixed effect pool panel regression will be implemented in the study, especially that it is the preferred model to control for unobservable variables.

ANALYSIS OF DATA SPECIFICATIONS AND CHARACTERISTICS
In order to detect the characteristics and the nature of the regressed data, tests including residuals normal distribution, homoscedasticity, serial correlation, multicollinearity and linearity were used.Given that no correlation between the independent variables is above 0.75 (Malhotra, 2004), we conclude multicollinearity is not a serious issue.

Test 3. Serial correlation
This assumption states that the error terms (disturbances) have independent distribution, where the correlation and covariance of the different residuals are zero.For the purpose of testing residuals autocorrelation, the Wooldridge test (2002) will be conducted, given that it fits the data model used in this study, which are the unbalanced panel data.7 shows the results of this test, where the p-value reported at 0.109 is greater than 0.05.We therefore cannot reject that the sample is drawn from a population that is normally distributed.

Descriptive statistics
Table 8 presents the descriptive statistics of the dependent and independent variables, which will be used to analyze some variables' behaviors and conduct simple analysis.Given that we are concerned about the relationship between the CFPS and the DPS, in comparison to the EPS and DPS, we can observe that the average EPS account for 90.5% of the average CFPS.Given that the mean EPS is smaller than the CFPS, the mean Dividends Payout Ratio (DPR) is higher on the basis of EPS in comparison to CFPS, which is 63.4% and 57.4%, respectively.The two ratios being relatively close to each other provide an indication that the average reported earnings of the banks and financial institutions listed on QSE for the selected period are conservative in comparison to the cash flows and no noticeable deviation is observed.
Based on the reported statistics, the coefficient of variation (CV) of the EPS is 0.88 and the CV of the CFPS is 0.90, which indicates that there is only a slightly higher variability in relation to the mean for the CFPS relative to the EPS.However, the variance ratio of the DPS over EPS is 0.30 (2.01^2/3.66^2),and over CFPS is 0.24, while the variance ratio of the EPS over CFPS is 0.80.Given that the CV of the CFPS is slightly higher than the CV of EPS while the variance ratio of the EPS over the CFPS is found to be 0.80, we can conclude that the studied firms and financial institutions exhibit a little degree of earnings smoothing.

RELATIONSHIP BETWEEN FIRMS' PERFORMANCE AND DPS CHANGES
Table 9 reflects the association between a firm's performance and the changes in its dividend payment.The studied banks and financial institutions were split between two groups: loss-incurring and profit-generating.In order to better understand the behavior of firms, a distinction was made between profitable firms that witnessed increase in profit, and those that had decreasing profits.The observations were subsequently divided into the three groups (loss incurring, profit increasing, and profit decreasing banks), along with the occurrences of dividends omission, maintenance, increase and decrease among those categories.As observed, the majority (67%) of the loss-incurring firms omitted the payment of dividends in the year during which losses were reported.Of the profitable firms, only 7% of the two groups declined to pay dividends, 30% of them increased the DPS in comparison to the previous period, and an equivalent percentage maintained their DPS in the same period.With respect to the firms that witnessed profit increases, the majority (41%) increased their dividend payment during the year of reported profit increase, and 32% of which maintained their dividends level, while only 8% did not pay any dividends during the years with profit increases.However, the majority (67%) of the firms that reported decreases in profit reduced their DPS level and 22% of which maintained their level of dividends.Those findings indicate that the profitable firms, firms reporting profit increases tend to either increase or maintain their dividends level, while firms reporting profit decrease tend to either decrease or maintain their dividend level.This finding is supported by Al Ajmi and Abo Hussain (2011).

Model (B). Extended Lintner's model empirical results -dividend payment determinants
Table 11 below presents the results of the extended Lintner's model, which is conducted on two versions to gauge the relationship between different variables and a firm's dividend payment: one including the EPS and the other independent variables which are previous year's DPS, the size, firm's leverage level, age of the firm, ROAE, and revenue growth, while the second one includes CFPS instead of the EPS, along with the other independent variables.The findings are discussed below.

Previous Year's Dividends per Share (PYDPS)
The results of both versions of model (B) support the findings of model (A), in which the coefficient of the PYDPS is positively related to the current year DPS, and is highly statistically significant, which indicates that the DPS has a positive relationship with the PYDPS, as expected.This finding indicates that the lagged dividends per share are important determinants of the dividends paid by banks and financial institutions in Qatar in any given year, which supports that those banks are reluctant to express inconsistency or deductions in the level of their dividends, which might lead to unsatisfied shareholders that do not have confidence in the banks' ability to generate stable earnings that allow them to distribute dividends, and hence resulting in an adverse impact on the bank's image.

Earnings per Share (EPS) and Cash Flow per Share (CFPS)
Similar to the findings of model (A) and its modified version, the EPS and CFPS coefficients are positively related to the current year DPS and are highly significant.This provides a strong signal of the importance of those variables in determining the DPS, in which the likelihood of paying dividends increases with the EPS and CFPS, as hypothesized.Given that the earnings of banks are split between retained earnings and dividends, higher earnings indicate higher banks' ability to pay out dividends, as the earnings generated would be sufficient to pursue activities using internal financing with retained earnings, as well as distributing dividends out to shareholders.The finding can also provide support to the signaling theory, in which firms tend to distribute dividends back to shareholders to signal that they are healthy, enjoy a solid financial position, and are able to generate sufficient earnings.Therefore, the more earnings generated, the more the dividend payment.

Firm Size (SIZE)
While the firm size is found to have a positive coefficient sign as expected, it has a non-significant relationship with the DPS.This finding is in-line with Al Ajmi and Abo Hussain (2011) but contra- The positive relationship indicates that as banks and financial institutions in Qatar grow in size, as they win more creditor's trust and confidence in their ability to owe their obligations, and hence they become more able to obtain external financing at more favorable financing cost in comparison to smaller banks, becoming less reliant on internal financing and therefore having higher ability to pay out dividends.However, a probable reason for the non-significant relationship between size and DPS can be attributed to the fact that since all banks and financial institutions in Qatar, and regardless of their size, are monitored and governed by the Central Bank of Qatar (CBQ), they should follow the CBQ policies and rules.

Leverage Ratio (LR)
Banks and financial institutions' leverage position is found to be negatively related to dividends.However, both versions of the regression results reveal that leverage is non-significant and hence is not considered as one of the DPS determinants, which contradict with what was reported by Agrawal and Narayanan (1994) and Aivazian et al. (2006).However, similar findings were reported by Al Ajmi and Abo Hussain (2011).The negative relationship between the leverage position and dividends could be due to two reasons.First, as banks become more highly leveraged, bondholders tend to impose higher restrictions on their earnings in order to be consistently able to service the committed fixed financing cost, and hence the leverage position negatively affects dividend payment.Second, following the financial crisis of 2007, the Bank for International Settlement (BIS) imposed additional restrictions on banks to enhance their financial position and increase their stable funds.Therefore, banks may have started taking the necessary measures in order to be able to meet the additional restrictive rules and regulatory buffers once they become compulsory, and hence lowering their dividend distribution and increasing their earnings retention.The non-significant relationship between the leverage position and the DPS could be justi-fied by the common nature of the core business activities carried out by all banks and financial institutions, in which they function as an intermediary between those with excess funds (lenders), and others who are in shortage (borrowers).Hence, the leverage position could be considered to not being unique to specific bank, and varies at a lower range, and is therefore excluded from being a significant explanatory variable to the DPS.

Firm Age (AGE)
Age, measured in this study by the retained earnings relative to common equity, was found to be negatively related to DPS, which indicates that chances for dividend payment decrease as a bank moves along the life cycle, a finding that contradicts our expectation of the existence of a positive relationship.The coefficient of the AGE is statistically significant and hence is one of the determinants of the dividend payments.The negative relationship found could be explained by the availability of many opportunities in the market; the more banks become well-established, the more they have unique branding and strong image in the eyes of customers and investors, the more they become able to enforce their strong existence in the market, the more they can go ahead with profitable expansion opportunities driven by the long history and strong relationship with customers and investors.

Return on Average Equity (ROAE)
As hypothesized, the ROAE as another measure of firm profitability was found to be positively related to DPS.However, according to Table 11 below, ROAE is non-significant and hence is not a determinant of the dividend payment.This finding contradicts with Maladjian and El Khoury (2014) who found that ROAE is negatively related to DPS.As the ROAE is a measure of profitability, its positive relationship with the DPS was expected.The higher ROAE level that a bank can stand at, the more profitable it is deemed to be.The positive relationship is also expected by shareholders, as it indicates the bank's ability to generate more earnings per a currency invested by the shareholders in its capital, and consequently, the higher its ability in distributing dividends to shareholders.

Revenue Growth (REVG)
As hypothesized, revenue growth is negatively related to dividends.However, the REVG coefficient is non-significant and hence the growth opportunities is not a primary determinant for dividend payment.The revenue growth relationship with DPS is consistent with the assumption that the more opportunities available in the market, which vary from introducing new financing products, pursuing expansion opportunities locally, regionally and internationally, as well as expanding the banking network among others, the more the chance that a bank would retain a larger portion of its earnings as an internal cheaper source of financing, than paying out dividends.Hence, the negative relationship can be justified.
As can also be concluded from the regression results table, the p-value of the F-test indicates that the overall model is highly significant in explaining the dependent variable (DPS).Owning to the results, and given that some of the regressed independent variables were found to be significant in determining the DPS, and with evidence from the F-test results that propose the overall model significance, there is no multicollinearity detected as the tests result support, and do not contradict, each other.

Model (C). Potential factors affecting banks' probability of dividend payment
Model (C) attempts to identify the factors that would affect a bank's decision and probability to pay out dividends in a given year.As Table 13 below shows, the previous year's dividend payment, earnings per share, banks' size and leverage level are all statistically significant and hence do affect a bank's decision to pay out dividends.The results show that all of these variables have the hypothesized signs of the relationship with the dividends, as all had positive signs, except for the leverage ratio that is negatively related to dividends.The results of the modified model (C) where CFPS is used in place of EPS show similar findings.The only difference between the two versions of model (C) is that the modified version includes the bank's age, referring to its life cycle, as a significant factor positively affecting a bank's probability of paying dividends.

CONCLUSION
The main objective of the present study was to identify the dividend policy determinants of banks and financial institutions listed on Qatar Stock Exchange (QSE) for a period from 2009 to 2015, through studying the impact on eight factors on banks' dividends per share.The fixed effects pool panel regression was implemented to run models (A) and (B).The findings revealed that the previous year's dividends per share, earnings per share, cash flow per share, firm size and return on average equity are positively related to the current year's dividends per share, as hypothesized.The study also found that the leverage position, bank's life cycle and growth opportunities are negatively related to the dividend payment.The negative relationship between a bank's leverage position and its dividend payment was explained by the possibility of higher restrictions on earnings imposed by the debt-holders due to the committed fixed debt cost and the Bank of International Settlement (BIS) additional restrictive requirements for Basel III following the financial crisis 2007, in which banks were required to enhance their liquidity position and maintain higher amounts in stable funds, which could negatively affect banks' dividend payment.The negative relationship between the life-cycle and dividends was explained by the possibility of the availability of plentiful growth opportunities in the market.
However, only four factors were found to be significant, which are lagged dividends, earnings per share, cash flow per share and life cycle.The study also found that the majority of loss-incurring firms tend to omit dividend payment during the period of reported losses, while the majority of firms with increasing profits tend to either increase or maintain their dividends level, and the majority of decreasing profits firms were found to either decrease or maintain their dividends level.

r
and the firm's earnings for the year ( ), t E as follows: and Al Ajmi and Abo Hussain (2011).Panel techniques were recommended by Baltagi (2001) who points out to the effectiveness of panel data, as it takes into account the heterogeneity that avails among individual banks, and enables to conduct the study of different factors with enhanced efficiency and less collinearity.Maladjian and El Khoury (2014) provided similar recommendations for the panel techniques as well.The sample size includes 12 firms for the prior mentioned period, totaling 84 observations.The data were collected from the firms' specific annual reports.

Table 2 .
Results of the Hausman test

Table 4
Chatterjee and Price (1977)of the VIF tests, where test 1 includes the CFPS along with the other independent variables, and test 2 includes the EPS instead.As indicated byChatterjee and Price (1977), VIF coefficient that is greater than 10 is considered as an indication for the presence of multicollinearity.As observed from the table below, the calculated VIFs are well below the limit of 10, and hence we can conclude that no strong correlation exists between the independent variables, which supports the findings of the Pearson tests.

Table 4 .
Results of the Variance Inflation Factor (VIF)Two tests will be conducted to test homoscedasticity.Koenker test and White test.P-values greater than 0.05 level of significance result in not rejecting the null hypothesis.Table5below shows that the p-values resulting from both tests are greater than 0.05, indicating that there is homoscedasticity and the error variance is constant across all independent variables.Therefore, no corrective measures have to be taken.

Table 5 .
Results of Koenker test and White test

Table 6
presents the results of the Wooldridge test, where the significance statistics indicate that residuals autocorrelation does not present in the study.As the p-value resulting from the test is greater than 0.05, the null hypothesis cannot be rejected, and we therefore conclude that serial correlation does not present in this study.

Table 3 .
Pearson correlation matrix

Table 6 .
Results of the Wooldridge testThis test is conducted to detect whether the residuals are normally distributed.Violation for the residuals normality might lead to inefficient, or at worst, misleading parameters and results.In order to detect whether the error terms are normally distributed, Shapiro-Wilk test is used in this study.For the Shapiro-Wilk test, two extreme residuals outliers were excluded from the sample.Where the p-value resulting from running the test is greater than 5% level of significance, we accept the null hypothesis and conclude that the residuals are normally distributed.Table

Table 8 .
Variables descriptive statistics

Table 10
(2011).The results of model (A) and its modified version clearly outline the importance of the PYDPS, EPS and CFPS on determining the current period's dividends.

Table 9 .
Association between a firms' performance and its changes in the dividend payment

Table 10 .
Results of Lintner's model (model (A) and (modified model (A) with Aivazian et al. (2006) who, based on his study findings, found that there is a significant positive relationship between size and dividend payment, and Ben Naceur et al. (2006) who found that dividends are negatively related to firm size. dicts

Table 11 .
Results of Lintner's extended model (model (B)) and modified extended model (B)

Table 12 .
Summary of model (B) results of coefficients relationship direction and significance