# Dividend policy as a multi-purpose mechanism; the case of conventional and Islamic banks before and after the 2008 crisis.

1. IntroductionConventional and Islamic banks differ in the components of their capital structures. Conventional banks (CBs) receive deposits and treat them as liabilities. In this case, the depositor is a lender, and the bank is the borrower bearing all the risks. Therefore, to avoid interest payment default, CBs strike a balance in their asset-liability management. Islamic banks (IBs), on the other hand, receive deposits under profit and loss sharing agreements. According to these agreements, depositors are considered partners, implying a possible loss of part, or even all, of the amount deposited without legal obligation on the bank. Because clients bear the risks, IBs are tempted to take more risks with these deposits for the reward of making higher returns. Therefore, IBs presumably enjoy more profitability during times of economic prosperity. Conversely, at times of economic downturns they struggle more for recovery.

With this background, there is reason to believe that CBs and IBs differ in the way they set their financial policies including, among others, the dividend policy. There is also reason to believe that they responded differently to the recent global financial crisis that started in 2008. To provide evidence of these beliefs, the determinants of dividend policy of the two bank types were investigated before and during the time of the crisis. The paper addresses several important questions. What are the determinants of dividend policy for CBs and IBs before and after the crisis? Did their purposes of paying cash dividends change in response to the crisis? If they did, what are they? And why did they change?

The results of this research should add more to the understanding of how dividend policy is set by CBs compared to IBs and provide theoretical and practical contributions.

The next section of the paper provides a discussion of the relevant literature on dividend policy focusing on the banking industry, and on the mixed banking systems in particular. The objective of the review is to develop our research conceptual framework and hypotheses which set out in a separate section. Following that, the scope of the data and the research methodology is addressed before presenting the results of the various tests and model estimations. The last section concludes with conclusions drawn from the research and some observations.

2. Literature review

In this paper, two strands of research relevant to the dividend policy of banks are recognized. The first is concerned with the relationship between firm value and its dividend policy that was initiated by Miller & Modigliani (1961). Their original proposition is that dividend policy is irrelevant to value. They claim that, if shareholders prefer an appreciation in stock market value, then profit should be retained by the firm to fund further growth. Therefore, the firm should pay no dividends. Alternatively, if they prefer cash, then they can always sell some of their shares. Again, no cash dividend distribution is necessary. They argue that shareholders are indifferent to the dividend policy of the firm. This argument implies that when firms pay, cut, increase or decrease cash payouts, stock traders do not respond and the stock market price remains unchanged. In the literature of corporate finance, this is called the "dividend irrelevance" theory. The first to criticize this theory were Lintner (1962) and Gordon (1963). They both believe that the Miller & Modigliani (1961) proposition is based on unrealistic assumptions. They believe that investors value a dollar of dividend today more than a dollar of price appreciation in the future. Miller & Modigliani (1961) disagreed and called this "bird-in-the-hand" theory. A third theory is called the "Tax preference" theory. It maintains that, where dividends are taxed more than capital gains, some investors prefer firms that retain profit to finance further growth. Because when they get cash dividend, a significant part of it will be deducted as taxes. This theory implies that an increase of dividend payout leads to lower value.

For those who believe that the dividend policy does matter to firm value (e.g., Barely et al., 1995; Fama & French, 1998; Dickens et al., 2003; and Bhattacharyya, 2007 for a review), it is crucial to investigate the determinants of this policy. This forms the second strand of the literature. To explain changes in dividend policy, the literature recognizes three main theories: the "clientele" theory, the "signaling" theory, and the "agency" theory. A review of the literature on these theories is summarized by Bhattacharyya (2007).

When Miller & Modigliani (1961) introduced their famous dividend-irrelevance proposition, it was criticized for being unrealistic as they ignored the effect of tax treatments. Opposing scholars (e.g., Peterson et al., 1985; Amihud & Murgia, 1997; Allen et al., 2000; Baker & Wurgler, 2004; DeAngelo & DeAngelo, 2006; and Komrattanapanya & Suntraruk, 2014) believe that dividend policy is relevant and that different clienteles have different preferences. Other researchers provided proofs of the signaling effects of cash payouts. Bhattacharyya (1979) developed a signaling proposition based on the notion of information asymmetry. He argued that managers have deeper and more detailed information about the future cash flows of the firm and they convey this information by means of additional dividends as a signaling mechanism. A similar argument was made by Miller & Rock (1985), William (1988), and Bernheim & Wantz (1995). Advocates of the agency hypothesis believe that profitable firms with no further growth opportunities should pay a cash dividend to prevent executives from misusing in non-profitable investments. The hypothesis implies that accumulation of cash is associated with an increased agency cost. This notion is supported by Rozeff (1982), Easterbrook (1984), Moh'd et al. (1995), La Porta (2000), DeAngelo et al. (2004).

As the main driver of economic strength, the banking industry was extensively investigated in relation to dividend policy. For those who believe that dividend distribution is relevant to the value of the firm, it was important to understand how dividend policy is determined. It is even more important when there is a mixed banking market. This is particularly true for the Gulf Cooperation Council (GCC) region and other Islamic countries where most of the Islamic banks in the world coexist with their conventional counterpart. Many research attempts, within these mixed-banking markets, have been made to study the differences in dividend policies of IBs versus CBs. Hassan et al. (2003), for example, provided evidence of dividend signaling and agency cost effects in an interest-free banking system. By studying the dividend policy of the Malaysian mixed-banking market, Ameer (2008) provided evidence that the decision to cut or increase dividend is determined differently. He did not differentiate between IBs and CBs. A comparison between the dividend policies of IBs and CBs was later investigated by Eng, Yahya, & Hadi (2013). The dividend policy of IBs was found to be affected only by lagged dividends, implying the signaling effect, while CBs' dividend was found to be affected only by profitability.

Zameer et al. (2013) explored the determinants of dividend policy in the mixed-banking industry of Pakistan. They found that profitability, pervious dividends, and ownership structure have a positive impact on dividend while liquidity had a negative impact. Size, risk, growth, and agency cost were found irrelevant to dividend policy. In his short paper on modeling dividend policy in the banking industry of the GCC region, Al-Ajmi (2010) did not differentiate between Islamic and conventional banks but provided mixed results about signaling and agency effects. He did, however, provide evidence of the effect of current earnings and previous dividend on cash payout policy. Athari et al. (2016) conducted a study of dividend policy in a dual-banking system covering some GCC countries in addition to Egypt and Jordan. They provided two agency explanation for the dividend payout. They found that IBs used dividends to mitigate risk and agency problem associated with accumulation of cash. Alternatively, CBs were found to follow the outcome model of dividend policy when dividend policy was more stable.

Al-Kayed (2017) compared the determinants of dividend policies of IBs to CBs. She found that IBs' dividend policy is affected by profitability, previous dividends, and leverage. She also found that for CBs, the dividend policy was affected by the same determinants in addition to liquidity and growth. The Agency theory was not tested.

Although some of the studies reviewed have covered the period before and after the 2008 global financial crisis, none of them has catered for its effect on bank dividend policy, especially when comparing between IBs and CBs. In the next section, we make use of the reviewed literature to develop our conceptual framework, dependent and explanatory variables, and research hypotheses.

3. Conceptual framework, variables, and hypotheses

The research framework in this paper recognizes three main propositions to explain dividend decisions exercised by Islamic and conventional banks in a mixed banking system. With the assumption of the fundamental differences between the capital structures of IBs versus CBs, one would assume different dividend policies and determinants for the two types of banks based on the status of the economy (before and after the 2008 global financial crisis). To pursue the investigation, the model will be based on the three most established theories of dividend policy: the signaling, pecking order, and the agency effect theories. If we can provide evidence of the adoption of different forms of effects in different economic occasions, then banks are using dividend payouts for different purposes according to different needs.

As we are focusing on the GCC region where taxes are of a lesser concern, we excluded the clientele theory. We elected nine explanatory variables that were discussed in the literation, three variables to represent each theory. These variables are:

For signaling effect, we chose

- Previous dividend ([div.sub.t-1]). This variable is calculated as the dividend per share in the previous year. When firms maintain a certain pattern in payouts despite variability in profitability performance, they signal their commitment to their shareholders (see for example Lintner, 1956; Miller & Modigliani, 1961; DeAngelo et al., 1992; Ben Naceur et al. (2006); and Al-Ajmi, 2010). Briston & Tomkins (1970) found that there was evidence of dividend policy stability in periods of a major change in the UK tax system. Eng, Yahya, & Hadi (2013) provided evidence of a positive association between payouts and previous dividend for IBs but not for CBs. Al-Kayed (2017) found a positive relation for CBs and IBs in Saudi Arabia. We, therefore, hypothesize that previous dividend positively affects dividend policy.

- Future earnings (geps). This is calculated as the change in earnings per share for the current year. There is ample evidence that the payment of dividends is seen by stock traders as a signal for future prospects of the firm. Miller & Modigliani (1961) argue that investors interpret change in dividend payout as a change in management view of future earnings. A statistically significant interaction effect between earnings and dividend announcement was reported by Kane et al. (1984). Nissim & Ziv (2001) reported similar results. Our hypothesis concerning this predictor is that future earning positively affects dividend policy.

- Deposits (lndep). It is calculated as the natural log of this year's deposits. Dividends signal on the future availability of cash (more deposits for banks). This variable could also be used to test the agency effect. The Miller & Modigliani (1961) argument on future cash flows could be applied to deposits for banks. Bhattacharyya (2007) provides a review discussion of this variable. Eng, Yahya, & Hadi (2013) did not support the hypothesis for either CBs or IBs. The hypothesis here is that deposits positively affect dividend policy.

For the pecking order effect, we chose:

- Profitability (eps). This variable is defined as the year's earnings per share. Companies prefer internal funding over external funding to finance dividend distribution. So, higher profitability is associated with more dividend payouts. DeAngelo et al. (1992) argue that earnings do explain variation in dividend payouts. Various profitability proxies were employed by Barclay et al. (1995), Adedeji (1998), and Al-Malkawi (2007). Ameer (2008) provides evidence to support this hypothesis for Malaysian banks. He did not compare between CBs and IBs. Eng, Yahya, & Hadi (2013) did not support the hypothesis for either CBs or IBs. Al-Kayed (2017) found a positive relation for CBs and IBs in Saudi Arabia. For non-banking firms, Ben Naceur et al. (2006) provided evidence of the positive effects of earnings. The hypothesis here is that profitability positively affects dividend policy.

- Investment (invtota). This is a proxy for a greater investment opportunity calculated as the percentage of investments on total assets. The relationship with dividend payouts is negative; when opportunities exist, they distribute less dividends. The hypothesis was supported by Al-Ajmi (2010) for most of the banks in the GCC region. However, he did not attempt to compare between CBs and IBs. Our hypothesis relevant to this predictor is that the investment ratio negatively affects dividend policy.

For agency effect, we chose:

- Agency cost (ac). The capital ratio is used as a proxy for agency cost. Agency theory stipulates that higher agency cost indicates management misconduct and may be relevant to the accumulated cash when there are less profitable opportunities. The association to payout ratio could go either way. It could be positive supporting the argument that more cash encourages paying more dividends. Or, it could be negative supporting the argument that with management misconduct lesser dividends, than it should be, is paid out to shareholders. As in Berger & Patti (2006) and Pratomo & Ismail (2007). The hypothesis here is that agency cost negatively affects dividend policy.

- Size (lnta). Large firms, typically measured by the natural log of their assets, generate more cash and the accumulated cash encourages management misconduct when profitable opportunities do not exist. The most of the existing literature on the size (a representative variable of agency) supports that larger firms should pay more dividends to overcome the agency problem. Eriotis (2005), Al-Malkawi (2007), and Denis & Osobov (2008) provided evidence that size is positively associated with higher cash payouts supporting the agency concept. Zameer et al. (2013) did not provide support for this hypothesis. Zameer et al. (2013) provided no support for this hypothesis. Our research hypothesis relevant to this predictor is that size positively affects dividend policy.

- Fixed Assets Growth (fatota). This variable is calculated as the percentage of net fixed assets over total assets. If firms have no growth investment opportunities (investing in fixed assets) and still pay dividends, then there is a reason to believe they want to avoid agency problems. Although Zameer et al. (2013) did not support it as a hypothesis in his research, Eng, Yahya, & Hadi (2013) found a positive association for the case of CBs but not for IBs. Al-Kayed (2017) found a positive relation for CBs, but not for IBs, in Saudi Arabia. Our hypothesis regarding this predictor is that fixed assets positively affect dividend policy.

- Risk index (ri). The ri as proposed by Sinkey (1988) and Sinkey & Nash (1993) is defined as([[sigma].sub.ROA] + E/A)/[[sigma].sub.ROA]. We add this predictor as we believe that an increase of risk index, which implies a decrease in the level of general risk the bank is facing, should be positively associated with cash payouts. The argument here is that banks may use cash payouts to lower the risk leading to lower agency problems. Our hypothesis with this predictor is that higher risk index (lower level of risk) positively affects dividend policy.

For the dependent variable, we chose:

- Dividend yield (dy). This variable is calculated as the current of divided per share over current price. Most of the research use the dividend payout ratio (current dividend divided by net income) as the dependent variable (see, e.g., Amidu & Abor (2006) and Komrattanapanya & Suntraruk, 2014).

A description of the variables is presented in Table 1.

4. Data and methodology

The data was collected using Bayanati financial & banking data from the Institute of Banking Studies in Kuwait. This is an online (Institute of Banking Studies, 2018) database for all companies listed in the Kuwait Stock Exchange. It also provides fundamental data for all banks in the GCC region. Initially, the attempt was to include all listed banks in the GCC region for the period from 2003 to 2016. Unfortunately, some were omitted because of either their short lives or lack of significant data. We also targeted strongly balanced panels, but this would have cut the sample even further. We ended up with a total of 76 banks, 50 of which are CBs and 26 are IBs.

Ultimately, the objective was to understand how the dividend policy of each bank type is set and to explore to what extent the 2008 global financial crisis affected how dividend policy is determined by each bank type.

The investigation starts with avoiding some of the common problems associated with causal model estimations. Therefore, we start with removing any multicollinearity among the proposed nine predictors for each bank type. This is done by estimating an OLS model regressing the dependent variable, dividend per share, on all nine predictors. The objective is to perform a variance inflation factor (VIF). Cutoff values of less than 10, 5, or even 2.5 are commonly used to eliminate variables with possible multicollinearity among predictors. We use a value of less than 2.5. We also use Fisher-type unit root method based on augmented Dickey-Fuller tests to test for data stationery. Variable remains when the null hypothesis of unit root existence is rejected. The Shapiro-Wilk W test for normal data is selected as it handles unbalanced panel data. This is important for deciding on which method to use for testing differences in means. The student t-test is preferred when data is normal. Alternatively, the nonparametric Kruskal-Wallis equality-of-populations rank test must be employed.

To explore causal effects, the nature of the data compiled for this study calls for panel data analysis. This type of analysis takes two forms, the random effect form and the fixed effect for. We adopt the Hausman test to select the appropriate model. This is a test that analyses the coefficients resulting from estimating tow OLS regressions, one with random effects and another with fixed effect. When the hypothesis of the test is rejected, a fixed effect model is more appropriate. Otherwise, the random effect model is more efficient, but the fixed effect model can still be consistent.

A general panel data fixed effect model may be written as following:

[y.sub.it] = [[alpha].sub.i]+[X.sub.it][beta] + [u.sub.it] for t = 1,...,T and i =1,...,N (1)

Where [y.sub.it] is the dependent variable for panel i at time t, [[alpha].sub.i] is the unobserved panel effect, [X.sub.it] is 1 x k explanatory variables matrix, [beta] is the calculated 1 x k coefficients matrix, and [u.sub.it] is the error term.

The general panel data random effect model can be written as

[y.sub.it] = [[alpha].sub.i] + [X.sub.it][beta] + [u.sub.it] + [[epsilon].sub.it] (2)

Where [u.sub.it] is the between-entity error and [[epsilon].sub.it] is the within entity error.

5. Diagnostics and preliminary testing

We start our diagnostics by checking the multicollinearity among the predictors for the CBs dataset. The results of the OLS regression and VIF test are presented in Table 2.

The results indicate high VIF value for four predictors with possible collinearity. By excluding these variables, we end up with six possible predictors. Table 3 presents the results of re-estimating the model with the six predictors. The new VIF values indicate the problem is solved.

After the VIF examination, we end up with six predictors for CBs causal investigation. To decide on fixed or random effect modeling, we perform the Hausman test. This test requires running equations (1) and (2) and comparing their estimates. The results of this test are presented in Table 4.

The conclusion of the Hausman test is that it not possible to use the random effect option to the model panel data regression. The fixed-effect model is more appropriate. The results of this test show a [chi square] score of 93.45 and a p-value of 0.000 rejecting the null hypothesis to accept random effect.

Regressing dps on geps, lndep, eps, debtota, invtota, ac, Inta, and ri with the VIF test in the 1st run produced the VIF values for IBs dataset with an obvious multicollinearity problem in four predictors as shown in Table 5.

The 2nd run of the OLS regression after removing three of the four problematic variables, produced VIF values for the remaining predictors higher than the 2.5 cutoff point. We have ended up with seven potential explanatory variables for IBs causal investigation.

To decide on which effect model to adopt, again, we use the Hausman test for the IBs dataset in the manner we did with the CBs dataset. The results of this test are shown in Table 6.

As indicated at the bottom of the Table 6, a [chi square] of 22.00 with a p-value of 0.0025 rejects the null hypothesis that the random effect appropriate. Therefore, for IBs dataset we would be using the panel data regression with fixed effect for causal investigation.

Table 7 presents the correlation matrix of the dependent variable and the potential explanatory variables for the CBs dataset.

A significant high positive correlation value between the dependent, dps, variable and the predictors; [div.sub.t-1] and eps indicate potential positive causal effects. The relatively low correlation values among the potential independent variables confirm the independency of the variables and low collinearity. A summary statistics of the variables for CBs dataset are presented in Table 8.

The zero minimum values for the dps and [div.sub.t-1] designate nonpayment of cash dividend in some years. The high standard deviation of the investment to total assets ratio designates the volatility of the profitability indicator of CBs which is possibly caused by the global financial crisis during the period of the study. A further and more detailed description of the variables (Table 9) shows a comparison of the variables' means before and after the crisis.

Most of the variables exhibit a logical decline which can be explained by damaging consequence of the global crisis in 2008. However, we can observe that CBs exhibited a remarkable increase in the growth of earning per share ratio indicating a rapid recovery from the crisis.

The correlation matrix for the selected variable of the IBs dataset is presented in Table 10. The significant correlation values between the dependent variable [div.sub.t-1] and two predictors eps and lnta indicate possible causal effects. The low correlation values among predictors confirm variables' independence.

Summary statistics of the selected variables for the IBs dataset is presented in Table 11. A high standard deviation value for the risk index indicates on the high variability in the level of risks, which is possibly a result of the financial crisis.

A comparison of the means before and after the crisis for the IBs dataset is illustrated in Table 12. Unlike CBs, IBs exhibit a noticeable deterioration in the growth of their earning per share ratio after the crisis signifying a struggle to recover from the crisis. A simple comparison of the same ratio between the two types of banks, demonstrates that CBs exhibit better resilience. The decline in most of the values following the crisis is logical.

Before moving on to the causal models, we present two preliminary tests of normality and data stationary for the CBs dataset and IBs dataset. The results of the Shapiro-Wilk normality test for CBs dataset is depicted in Table 13.

The normality null hypothesis is rejected for all variables. The result implies that, for a comparison between means, a nonparametric rank test is more appropriate that the student t-test for the CBs dataset. Table 14 presents the results of the unit root test for data stationery, again, for the CBs dataset.

The test rejects the null hypothesis that data contains unit root implying that data is stationary for all selected variable of the CBs data set. The results of the nonparametric Kruskal-Wallis-of-populations rank test is shown in Table 15. The table is useful for detecting the variables that changed significantly in response to the 2008 crisis.

The Table 15 shows that except for the fixed assets ratio and risk index, the ranks of all remaining variables are significantly different indicating significant differences in means for CBs.

The results of the Shapiro-Wilk W test of normality for IBs (Table 16) shows that data of the selected variables are not normally distributed which, again, means that we cannot use parametric tests to compare means.

The results of the Fisher unit root test (Table 17) indicate that only [div.sub.t-1] variable contains a unit root for IBs dataset. According to this result, this variable should be removed as a predictor.

Table 18 exhibits the results of the Kruskal-Wallis rank test which indicates that [div.sub.t-1] variable is significantly different after the crisis. Similarly, dps, geps, eps, and lnta exhibit significant ranks differences.

6. Panel regression estimation

The causal fixed effect model represented in equation (1) will be estimated for three datasets. The first dataset contains conventional banks only, without considering the effect of the crisis. This is important for a comparison of these results with those after the inclusion of the crisis' effects. The second dataset includes only the panels representing CBs for the years before 2008. The third dataset includes the CBs for the period from 2008 to 2016. This is important to show the significant effect of the crisis when the periods of economic stability is separate from the period of economic instability. It may also provide evidence of how important it is to reflect such effects. Many financial studies were conducted using the time period covering the years 2008 and neglected the effect of the crisis.

The results of estimating the model for CBs using the 1st dataset is shown in Table 19. Two variables only ([div.sub.t-1] and eps) are positively significant at the 5% significance level.

The results imply that the dividend policy of CBs is determined only by previous dividend and the current earnings per share. The question here is whether these results will hold when separating the data into two datasets: before and after the crisis. To answer this question, we re-estimated the model for the "before-crisis" dataset. The results of this estimation are presented in Table 20. The previous dividend effect did not hold for this dataset. Only the current earnings per share have a positive and significant effect of the dividend policy of conventional banks.

Now, the question is "What factor can predict dividend policy for the third dataset?". The results of the third run of the panel regression model are depicted in Table 21. Four variables have significant positive effects. These are previous dividends, earning per share, investment ratio, and the risk index.

We follow the same procedures to investigate the dividend policy of IBs. The results of estimating the fixed effect panel regression for the complete dataset of IBs are presented in Table 22. A significant F value of 7.27 is indicative of good model specification.

Three predictors have a positive significant effect on dividend policy. These are earnings per share, investment ratio, and size. The risk index is also positively significant at the 10% level.

Running the model for "before-crisis" dataset reveals weak model specification as indicated by the F value of 0.61 with prob>F = 0.7177 (Table 23).

The weak specification may be a result of a relatively small number of observations compared to 16 groups. For Islamic banks, the weak data sample is undermining any possible inference relevant to the period before the crisis. The reason for a smaller sample for the period before 2008 is the fact that there were fewer banks at that time. Gradually, the increase of the number of listed banks led to the larger sample size after 2008. This is reflected in the next run of the model.

The results for the "after crisis" dataset show that the earning per share predictor is the only predictor that is significant at the 5% level (Table 24). Though, two other predictors (growth in earnings per share and size) are significant at the 10% level.

7. Discussion

We start our discussion with a summary of the causal model results as illustrated in Table 25.

We can understand from the outcome of this study that the dividend policy of CBs robustly confirms the pecking order hypothesis. Conventional banks seem to continue adopting the policy of preferring internal funding over external funding. The association of higher dividend payouts with the higher earnings is evident from the correlation relationship discussed earlier, and the significant causal effect of earning per share on cash dividends paid out to shareholders. Earnings per share variable is the only robust predictor that appeared to be statistically significant in all three data sets. The result confirms the pecking order effect evidence provided by Ameer (2008) in the study conducted on Malaysian banks. Although to a lesser extent, the previous dividend per share predictor can also be considered robust as it appeared to be statistically significant in two of the three datasets, the "entire-dataset" and the "after crisis" dataset. This result implies that CBs, despite the dramatic decrease in profitability following the crisis, were striving to signal their ability to maintain a pattern of cash payment for shareholders. This is an evidence of the adoption of a cash payout as a signaling instrument. It confirms previous earlier propositions made by Lintner (1956) and the evidence provided by DeAngelo et al. (1992) from their study on the NYSE. The final interesting result for CBs is the positive and significant effect of the risk index on dividend payouts. We note here that a higher risk index implies a lower level of risks associated with lower agency problem. What we may understand from this result is that, and in response to the global financial crisis, CBs adopted the agency effect with cash payouts to mitigate management misconduct at times of economic instability. The result is consistent with the arguments made by Rozeff (1982) and DeAngelo & DeAngelo (2004).

For IBs, earning per share is the only robust predictor with a significant positive effect on dividend policy. It appeared statistically significant for the "entire" dataset and the "after crisis" dataset. The argument here is that IBs in the GCC region are following the pecking order to distribute cash dividends to shareholders. They prefer internal funding over external funding. Looking at the "entire" dataset, we can observe that IBs use dividend payouts to signal the strength of their financial positions. This is confirmed by the significant and positive effect of the investment ratio on dividend policy. Our explanation for the positive effect is that IBs use dividend payouts to signal their ability to pay a cash dividend in the presence of investment opportunity which also requires funding. This latter result contradicts with the evidence of negative association provided by AL-Ajmi (2010) for the same region. The positive and significant effect of bank size on dividend policy for the "entire" dataset implies that IBs are gradually adopting payment of cash dividends as a mechanism to mitigate management misconduct and avoid potential agency problems. This result is consistent with the arguments made by Zameer et al. (2013) but contradicts the evidence provided by Al-Kayed (2017) who found a positive association for CBs only.

8. Research limitations

This study suffers from two obvious limitations. The first one is the, relatively, small sample of IBs compared to that of CBs. This limitation is tied to the shorter time series of some of the IBs included in the sample which may undermine the confidence in the results generalization relevant to IBs in the GCC region. The second limitation is the fact that it focuses on a regional market instead of considering the wider global market. Investigating this subject at the global market level would, certainly, increase the confidence to generalize on the results beyond the regional market.

9. Concluding remarks

Considerable literature has been accumulating following Miller & Modigliani (1961) controversial argument on dividend policy. Since then, the majority of this literature provided evidence on its relevance to firm value. Many forms of dividend effects were progressively developed, including, but not limited to, signaling, pecking-order, and agency. Guided by these forms of effects, we developed a conceptual framework to investigate the applicability of these forms in a mixed banking industry before and during a major economic downturn. The main proposition was that, if CBs and IBs differ in their capital structures they must be different in the way they set their dividend policies. We adopted a panel data regression model with fixed effect to test this proposition.

The results reveal that, at times of economic prosperity (before the 2008 crisis), CBs used dividend payout as a signaling instrument and a pecking-order instrument. At times of economic instability (after the 2008 crisis), they used it as an agency-protection instrument. A logical interpretation of this result is that CBs realized that agency problems during the time of economic instability is more damaging to the value of the firm calling for more controlling measures. For IBs, the results indicate that dividend policy is used in the form of pecking-order mechanism irrelevant to the status of the economy. It is interesting to note that, only at the time of economic stability, IBs used dividend payout as an instrument against agency problems. A possible interpretation of this result is that IBs have realized the agency problem during this period and acted sooner to mitigate its effects. The absence of this instrument after the crisis may be due to the significant decrease in IBs' growth of earnings compared to CBs as discussed earlier.

The main conclusion of the results is that dividend policies of CBs and IBs are determined differently. Dividend policy for each bank type differed according to economic status and the individual needs of bank types. The change in the form-of-effects implies that dividend policies are used as a multi-purpose mechanism. That is the form of effect changes as the need changes. One important theoretical implication of this argument is that the multi-purpose notion developed in this paper may be a practical interpretation of the so-called "dividend puzzle" of Black (1976).

One final note on the results is that, for both bank types and irrelevant to economics status, earnings per share is the only robust predictor of dividend policy serving the purpose of pecking-order effect. This result indicates that, when it comes to considering new investment opportunities, banks in the GCC region, prefer internal funding over external funding.

The results of this paper show that dividend payouts represent an important instrument that can be informative to investors about the future perspectives of the firm. They also serve as a growth funding instrument and a protection instrument against agency risks. To maintain the objective of value maximization, and depending on their particular needs, bankers should know which form of dividend policy to adopt. For instance, in certain occasions, dividend policy can serve only as a signaling instrument, while in other occasions it should be adopted as an investment funding instrument or as a protection mechanism against management misconduct.

References

Adedeji, A. (1998). Does the pecking order hypothesis explain the dividend payout ratios of firms in the UK? Journal of Business Finance and Accounting, 25(9-10), 1127-1155.

Al-Ajmi, J. (2010). Modelling the dividend policy of banks in Gulf Cooperation Council countries. Applied Economics Letters, 17(14), 1423-1428.

Al-Kayed, L. T. (2017). Dividend payout policy of Islamic vs conventional banks: case of Saudi Arabia. International Journal of Islamic and Middle Eastern Finance and Management, 10(1), 117-128.

Al-Malkawi, N. H. A. (2007). Determinants of corporate dividend policy in Jordan: an application of the Tobit model. Journal of Economic and Administrative Sciences, 23(2), 44-70.

Allen, F., Bernardo, A. E., & Welch, I. (2000). A theory of dividends based on tax clienteles. The Journal of Finance, 55(6), 2499-2536.

Ameer, R. (2008). Product market competition, regulation and dividend payout policy of Malaysian banks. Journal of Financial Regulation and Compliance, 16(4), 318-334.

Amidu, M., & Abor, J. (2006). Determinants of dividend payout ratios in Ghana. The Journal of Risk Finance, 7(2), 136-145.

Amihud, Y., & Murgia, M. (1997). Dividends, taxes, and signaling: evidence from Germany. The Journal of Finance, 52(1), 397-408.

Athari, S. A., Adaoglu, C., & Bektas, E. (2016). Investor protection and dividend policy: The case of Islamic and conventional banks. Emerging Markets Review, 27, 100-117.

Baker, M., & Wurgler, J. (2004). A catering theory of dividends. The Journal of Finance, 59(3), 1125-1165.

Barclay, M. J., Smith, C. W., & Watts, R. L. (1995). The determinants of corporate leverage and dividend policies. Journal of Applied Corporate Finance, 7(4), 4-19.

Ben Naceur, S., Goaied, M., & Belanes, A. (2006). On the determinants and dynamics of dividend policy. International Review of Finance, 6(1-2), 1-23.

Berger, A. N., & Di Patti, E. B. (2006). Capital structure and firm performance: A new approach to testing agency theory and an application to the banking industry. Journal of Banking and Finance, 30(4), 1065-1102.

Bernheim, B.D., Wantz, A., 1995. A tax based test of the dividend signaling hypothesis. American Economic Review, 85, 532-551.

Bhattacharya, S. (1979). Imperfect information, dividend policy, and "the bird in the hand" fallacy. Bell Journal of Economics, 10(1), 259-270.

Bhattacharyya, N. (2007). Dividend policy: a review. Managerial Finance, 33(1), 4-13.

Black, F. (1976). The dividend puzzle. Journal of Portfolio Management, 2(2), 5-8.

Briston, R. J., & Tomkins, C. R. (1970). Dividend policy, shareholder satisfaction, and valuation of shares. Journal of Finance, 2(1), 17-24.

DeAngelo, H., DeAngelo, L., & Skinner, D. J. (1992). Dividends and losses. The Journal of Finance, 47(5), 1837-1863.

DeAngelo, H., DeAngelo, L., & Stulz, R. (2004). Dividend policy, agency costs, and earned equity (Working Paper, No. w10599). National Bureau of Economic Research. Cambridge, MA.

DeAngelo, H., & DeAngelo, L. (2006). The irrelevance of the MM dividend irrelevance theorem. Journal of Financial Economics, 79(2), 293-315.

Dickens, R. N., Casey, K. M., & Newman, J. A. (2003). Bank dividend policy: explanatory factors. Quarterly Journal of Business and Economics, 41(1-2), 3-12.

Denis, D. J., & Osobov, I. (2008). Why do firms pay dividends? International evidence on the determinants of dividend policy. Journal of Financial Economics, 89(1), 62-82.

Easterbrook, F. H. (1984). Two agency-cost explanations of dividends. The American Economic Review, 74(4), 650-659.

Eriotis, N. (2005). The effect of distributed earnings and size of the firm to its dividend policy: some Greek data. International Business and Economics Journal, 4(1), 67-74.

Fama, E. F., & French, K. R. (1998). Taxes, financing decisions, and firm value. The Journal of Finance, 53(3), 819-843.

Gordon, M. J. (1963). Optimal investment and financing policy. The Journal of finance, 18(2), 264-272.

Hassan, K. M., Farhat, J. B., & Al-Zu'bi, B. (2003). Dividend signaling hypothesis and short-term asset concentration of Islamic interest-free banking, Islamic Economic Studies, 11(1), 1-30.

Institute of Banking Studies (2018). Financial and banking data. Retrieved October 16, 2018, from http://168.187.84.149/Bayanati/

Pratomo, W. A., & Ismail, A. G. (2007). Islamic bank performance and capital structure (MPRA Paper No. 6012). Retrieved October 17, 2018, from http://mpra.ub.uni-muenchen.de/6012/

Kane, A., Lee, Y. K., & Marcus, A. (1984). Earnings and dividend announcements: is there a corroboration effect? The Journal of Finance, 39(4), 1091-1099.

Komrattanapanya, P., & Suntraruk, P. (2014). Factors influencing dividend payout in Thailand: A Tobit regression analysis. International Journal of Accounting and Financial Reporting, 3(2), 255-268.

La Porta, R., Lopez-de-Silanes, F., Shleifer, A., & Vishny, R. W. (2000). Agency problems and dividend policies around the world. The Journal of Finance, 55(1), 1-33.

Lintner, J. (1956). Distribution of incomes of corporations among dividends, retained earnings, and taxes. The American Economic Review, 46(2), 97-113.

Lintner, J. (1962). Dividends, earnings, leverage, stock prices and the supply of capital to corporations. The Review of Economics and Statistics, 44(3), 243-269.

Miller, M. H., & Modigliani, F. (1961). Dividend policy, growth, and the valuation of shares. The Journal of Business, 34(4), 411-433.

Miller, M. H., & Rock, K. (1985). Dividend policy under asymmetric information. The Journal of finance, 40(4), 1031-1051.

Moh'd, M. A., Perry, L. G., & Rimbey, J. N. (1995). An investigation of the dynamic relationship between agency theory and dividend policy. Financial Review, 30(2), 367-385.

Nissim, D., & Ziv, A. (2001). Dividend changes and future profitability. The Journal of Finance, 56(6), 2111-2133.

Peterson, P. P., Peterson, D. R., & Ang, J. S. (1985). Direct evidence on the marginal rate of taxation on dividend income. Journal of Financial Economics, 14(2), 267-282.

Rozeff, M. S. (1982). Growth, beta and agency costs as determinants of dividend payout ratios. Journal of Financial Research, 5(3), 249-259.

Sinkey, J. F. (1988). Commercial bank financial management. Upper Saddle River, N. J.: Prentice Hall.

Sinkey J. F., & Nash R. C. (1993). Assessing the riskiness and profitability of credit-card banks. Journal of Financial Services Research, 2, 127-150.

Eng, S. H, Yahya M. H., & Hadi, A. R. A. (2013). The dividend payout policy-A comparison on Malaysian Islamic and conventional financial institutions. Journal of WEI Business and Economics, 2(2), 12-20

Williams, J. (1988). Efficient signalling with dividends, investment, and stock repurchases. The Journal of Finance, 43(3), 737-747.

Zameer, H., Rasool, S., Iqbal, S., & Arshad, U. (2013). Determinants of dividend policy: A case of banking sector in Pakistan. Middle-East Journal of Scientific Research, 18(3), 410-424.

Talla M Aldeehani

Department of Finance & Financial Institutions, College of Business Administration, Kuwait University

corresponding e-mail: talla[at]cba(dot)edu{dot}kw

address: Department of Finance & Financial Institutions, College of Business Administration, Kuwait University, P.O.Box 13055, Safat 13055

DOI: http://dx.doi.org/10.15208/beh.2019.3

TABLE 1. VARIABLES DESCRIPTION VARIABLE DESCRIPTION PROXY [div.sub.-1]t Previously paid Percentage of previous dividend over dividend price geps Growth in future Change in earnings per share earnings lndep Size of deposits Calculated as the natural log of deposits eps Profitability Calculated as net income over number of shares invtota Size of investments Calculated as the percentage of investments over total assets ac Agency cost Calculated as the capital ratio Inta Size of bank Natural log of total assets fatota Fixed assets Calculated as net fixed assets to growth total assets ri Risk Index Calculated as the standard deviation of return on assets added to the equity ratio all divided by the standard deviation of return on assets dy Dividend yield Percentage of current dividend per (dependent) share over price VARIABLE USED BY [div.sub.-1]t Lintner (1956); Miller & Modigliani (1961); DeAngelo et al. (1992); Ben Naceur et al. (2006); Al-Ajmi (2010); Briston & Tomkins (1970); Eng, Yahya, & Hadi. (2013) geps Miller & Modigliani (1961); Kane et al. (1984); Nissim & Ziv (2001) lndep Bhattacharyya (2007); Eng, Yahya, & Hadi. (2013) eps DeAngelo et al. (1992); Barclay et al. (1995); Adedeji (1998); Ben Naceur et al. (2006); Al-Malkawi (2007); Ameer (2008); Eng, Yahya, & Hadi. (2013); Al-Kayed (2017) invtota AL-Ajmi (2010) ac Berger & Patti (2006); Pratomo & Ismail (2007) Inta Eriotis (2005); Al-Malkawi (2007); Denis & Osobov (2008); Zameer et al. (2013) fatota Zameer et al. (2013); Eng, Yahya, & Hadi. (2013); Al-Kayed (2017) ri Sinkey (1988); Sinkey & Nash (1993) dy Amidu & Abor (2006); Ben Naceur et al. (2006); Komrattanapanya & Suntraruk, (2014) TABLE 2. RESULTS OF THE PRELIMINARY OLS REGRESSION AND VIF FOR CONVENTIONAL BANKS (CBS) VARIABLE COEF. STD. ERR. T P>| T | [95% CONF. INTERVAL] [div.sub.t-1] .494052 .0324784 15.21 0.000 .4302682 geps .0004283 .0004727 0.91 0.365 -.0004999 lndep -.0012079 .0271931 -0.04 0.965 -.054612 eps .1639327 .0151275 10.84 0.000 .1342239 debtota .8141158 .9596008 0.85 0.397 -1.070431 invtota .0584972 .044357 1.32 0.188 -.028615 ac .9347817 .9575347 0.98 0.329 -.9457076 lnta -.0022739 .027766 -0.08 0.935 -.0568031 fatota .0295739 .8918612 0.03 0.974 -1.72194 ri .0008221 .0003175 2.59 0.010 .0001984 _cons -.8186654 .9618839 -0.85 0.395 -2.707696 VARIABLE [95% CONF. INTERVAL] VIF [div.sub.t-1] .5578358 1.49 geps .0013565 1.01 lndep .0521962 43.38 eps .1936414 1.52 debtota 2.698663 102.44 invtota .1456093 1.44 ac 2.815271 99.50 lnta .0522554 37.58 fatota 1.781088 1.15 ri .0014457 1.17 _cons 1.070365 TABLE 3. RESULTS OF THE PRELIMINARY OLS REGRESSION AND VIF FOR CONVENTIONAL BANKS (CBS) AFTER REMOVAL OF PROBLEMATIC VARIABLES VARIABLE COEF. STD. ERR. T P>|T| [95% CONF. INTERVAL] [div.sub.t-1] .4915182 .031974 15.37 0.000 .4287257 .5543107 geps .0004483 .0004719 0.95 0.343 -.0004785 .001375 eps .1615718 .0146366 11.04 0.000 .1328275 .190316 invtota .0534518 .0375203 1.42 0.155 -.0202328 .1271364 fatota .3403701 .8422678 0.40 0.686 -1.313726 1.994467 ri .0009154 .0003068 2.98 0.003 .0003129 .0015179 _cons -.0227208 .0168801 -1.35 0.179 -.055871 .0104293 VARIABLE VIF [div.sub.t-1] 1.45 geps 1.01 eps 1.42 invtota 1.03 fatota 1.03 ri 1.09 _cons TABLE 4. RESULTS OF THE HAUSMAN TEST FOR CONVENTIONAL BANKS (CBS) VARIABLE COEFFICIENTS (b-B) DIFFERENCE sqrt(diag(V_b-V_B)) (b) fe (B) re S.E. [div.sub.t-1] .3320877 .4915182 -.1594304 .0199423 eps .0000308 .0004483 -.0004174 .0000959 ri .1774903 .1615718 .0159186 .0124859 invtota .1328597 .0534518 .0794079 .0736235 fatota .7859691 .3403701 .445599 .7769467 geps .0018788 .0009154 .0009635 .0011805 Notes: b = consistent under Ho and Ha; obtained from panel regression model estimator. B = inconsistent under Ha, efficient under Ho; obtained from panel regression model estimator. Test: Ho: difference in coefficients not systematic. [chi square] (6) = (b-B)'[(V_b-V_B)^(- 1)](b-B)=93.45, Prob>chi2 = 0.0000. TABLE 5. THE VIF TESTS BEFORE AND AFTER THE REMOVAL OF PROBLEMATIC PREDICTORS 1ST RUN: INCLUDING ALL VARIABLES VARIABLE VIF 1/VIF ac 127.50 0.007843 debtota 122.56 0.008159 lnta 7.04 0.142083 lndep 6.24 0.160272 [div.sub.t-1] 2.19 0.456677 eps 1.94 0.515272 invtota 1.49 0.673159 fatota 1.26 0.796422 ri 1.20 0.833043 geps 1.08 0.924165 Mean VIF 27.25 2ND RUN: AFTER REMOVING HIGH VIF VARIABLES VARIABLE VIF 1/VIF [div.sub.t-1] 2.10 0.476941 eps 1.94 0.516417 lnta 1.43 0.698236 invtota 1.18 0.845868 fatota 1.10 0.909413 ri 1.09 0.915711 geps 1.06 0.942521 Mean VIF 1.41 TABLE 6. RESULTS OF THE HAUSMAN TEST FOR ISLAMIC BANKS (IBS) VARIABLE COEFFICIENTS (b-B) DIFFERENCE sqrt(diag (b) fe (B) re (V_b-V_B)) S.E. [div.sub.t-1] .4521232 .5722702 -.120147 .0262 geps -.0001974 -.0004162 .0002188 .0005405 eps .1004533 .1290801 -.0286268 .0161353 invtota .1439666 .0978009 .0461658 .0956167 lnta .0113903 .0084274 .0029628 .0105557 fatota -.0645429 -.1350706 .0705277 .6684216 ri .0015018 .0000782 .0014236 .0010676 Notes: b = consistent under Ho and Ha; obtained from panel regression model estimator. B = inconsistent under Ha, efficient under Ho; obtained from panel regression model estimator. Test: Ho: difference in coefficients not systematic. [chi square] (6) = (b-B)'[(V_b-V_B)^(-1)](b-B) = 22.00, Prob>chi2 = 0.0025. TABLE 7. VARIABLES' CORRELATION MATRIX FOR CONVENTIONAL BANKS (CBS) DATASET VARIABLE dps [div.sub.t-1] geps eps invtota dps 1.0000 [div.sub.t-1] 0.6920 1.0000 geps -0.0078 -0.0584 1.0000 eps 0.6242 0.5337 -0.0027 1.0000 invtota 0.0612 0.0495 -0.0090 -0.0122 1.0000 fatota -0.0300 -0.0397 -0.0209 -0.0279 -0.1517 ri 0.2883 0.2563 -0.0365 0.2336 0.0627 VARIABLE fatota ri dps [div.sub.t-1] geps eps invtota fatota 1.0000 ri -0.0665 1.0000 TABLE 8. SUMMARY STATISTICS FOR CONVENTIONAL BANKS (CBS) DATASET VARIABLE OBS. MEAN STD. DEV. MIN MAX dps 668 .1865023 .1979475 0 2.266229 [div.sub.t-1] 617 .1934515 .199378 0 2.266229 geps 618 .6762588 11.2603 -8.423469 265.6776 eps 668 .4439088 .4232255 -2.867297 2.735895 invtota 668 .2707241 .1468702 .0287392 .9203521 fatota 668 .0092773 .006408 .0004146 .0827879 ri 668 25.2879 18.33361 -2.510582 123.5363 TABLE 9. COMPARISON OF MEANS FOR CONVENTIONAL BANKS (CBS) DATASET VARIABLE CRISIS STATUS MEAN STD. ERR. [95% CONF. INTERVAL] dps Before .2424889 .0161877 .2106991 After .1539455 .0084091 .1374314 [div.sub.t-1] Before .2412436 .0128616 .2159856 After .1664017 .0100051 .1467535 geps Before .253935 .1504851 -.041591 After .9167617 .705402 -.4685226 eps Before .5906036 .0338167 .5241936 After .3723012 .0182767 .336409 invtota Before .2993945 .0105474 .2786813 After .2466978 .0066143 .2337084 fatota Before .009396 .0005038 .0084067 After .0091001 .0002844 .0085416 ri Before 25.05314 1.266225 22.5665 After 25.23576 .8864034 23.49503 VARIABLE [95% CONF. INTERVAL] dps .2742787 .1704595 [div.sub.t-1] .2665015 .18605 geps .549461 2.302046 eps .6570136 .4081934 invtota .3201076 .2596871 fatota .0103853 .0096585 ri 27.53978 26.9765 TABLE 10. CORRELATION MATRIX FOR ISLAMIC BANKS (IBS) DATASET VARIABLE dps [div.sub.t-1] geps eps invtota lnta dps 1.0000 [div.sub.t-1] 0.8079 1.0000 geps 0.0491 0.0157 1.0000 eps 0.6892 0.6676 0.1678 1.0000 invtota -0.0220 -0.0342 -0.0095 -0.1365 1.0000 lnta 0.4035 0.4323 0.0346 0.3475 -0.3236 1.0000 fatota 0.1145 0.1456 -0.0480 0.1075 -0.1373 0.2240 ri -0.0586 -0.0901 0.0862 -0.0242 -0.1778 0.0908 VARIABLE fatota ri dps [div.sub.t-1] geps eps invtota lnta fatota 1.0000 ri 0.1831 1.0000 TABLE 11. SUMMARY STATISTICS FOR ISLAMIC BANKS' (IBS') VARIABLES VARIABLE OBS. MEAN STD. DEV. MIN MAX dps 267 .104675 .152217 0 .7 [div.sub.t-1] 241 .1091911 .157669 0 .7 geps 241 -.0788719 3.650523 -37.72849 19.74044 eps 267 .2389055 .3234104 -.5473449 2.292303 invtota 267 .1425336 .0977161 0 .6437339 lnta 267 8.656587 1.321699 5.242333 11.41535 fatota 267 .0135623 .013323 0 .075341 ri 267 20.98328 20.07043 1.582506 130.0466 TABLE 12. COMPARISON OF MEANS FOR ISLAMIC BANKS (IBS) DATASET VARIABLE CRISIS STATUS MEAN STD. ERR. [95% CONF. INTERVAL] dps Before .1610499 .0248299 .1121374 After .0869177 .0092883 .0686207 [div.sub.t-1] Before .1505052 .0248887 .1014769 After .093933 .0102524 .0737368 geps Before .3667386 .1758421 .0203477 After -.243444 .3147877 -.8635437 eps Before .4782576 .0576427 .3647074 After .1631106 .0148684 .1338215 invtota Before .1431949 .0108466 .1218282 After .139813 .0069763 .1260705 lnta Before 8.230411 .1550768 7.924926 After 8.989282 .0908015 8.810412 fatota Before .0142881 .0020761 .0101983 After .013389 .0009128 .0115909 ri Before 17.97171 1.329654 15.35243 After 20.10519 1.457966 17.23314 VARIABLE [95% CONF. INTERVAL] dps .2099623 .1052146 [div.sub.t-1] .1995334 .1141293 geps .7131295 .3766556 eps .5918077 .1923998 invtota .1645616 .1535555 lnta 8.535897 9.168152 fatota .0183779 .015187 ri 20.59099 22.97723 TABLE 13. RESULTS OF THE NORMALITY TEST FOR CONVENTIONAL BANKS (CBS) DATASET VARIABLE OBS. W V z PROB.>Z dps 668 0.83098 73.874 10.475 0.00000 [div.sub.t-1] 617 0.83388 67.567 10.223 0.00000 geps 618 0.05801 383.699 14.438 0.00000 eps 668 0.83929 70.240 10.352 0.00000 Invtota 668 0.92983 30.670 8.335 0.00000 fatota 668 0.75375 107.626 11.391 0.00000 ri 668 0.80632 84.653 10.807 0.00000 TABLE 14. RESULTS OF DATA STATIONARY TEST FOR CONVENTIONAL BANKS (CBS) DATASET VARIABLE Z-STATISTIC P-VALUE dps -5.1015 0.0000 [div.sub.t-1] -4.2699 0.0000 geps -17.3496 0.0000 eps -6.0794 0.0000 invtota -5.8474 0.0000 fatota -5.9968 0.0000 ri -4.3335 0.0000 TABLE 15. KRUSKAL-WALLIS EQUALITY-OF-POPULATIONS RANK TEST FOR CONVENTIONAL BANKS (CBS) DATASET VARIABLE STATUS OBS. RANK SUM [chi square] PROB. dps Before 272 104122.50 28.746 0.0001 After 396 119323.50 [div.sub.t-1] Before 223 80378.00 29.080 0.0001 After 394 110275.00 geps Before 223 77282.00 15.028 0.0001 After 395 113989.00 eps Before 272 109136.00 54.871 0.0001 After 396 114310.00 invtota Before 272 103355.00 25.486 0.0001 After 396 120091.00 fatota Before 272 92008.00 0.175 0.6760 After 396 131438.00 ri Before 272 89054.00 0.620 0.4310 After 396 134392.00 TABLE 16. RESULTS OF THE NORMALITY TEST FOR ISLAMIC BANKS (IBS) DATASET VARIABLE OBS. W V z PROB.>Z dps 267 0.83194 32.299 8.111 0.00000 [div.sub.t-1] 241 0.83463 29.040 7.823 0.00000 geps 241 0.35057 114.046 11.000 0.00000 eps 267 0.82710 33.230 8.177 0.00000 Invtota 267 0.94230 11.089 5.616 0.00000 lnta 267 0.98836 2.238 1.880 0.00000 fatota 267 0.78765 40.811 8.657 0.00000 ri 267 0.72752 52.367 9.239 0.00000 TABLE 17. RESULTS OF DATA STATIONARY TEST FOR ISLAMIC BANKS (IBS) DATASET VARIABLE Z-STATISTIC P-VALUE dps -1.9207 0.0274 [div.sub.t-1] -0.9812 0.1632 geps -12.3417 0.0000 eps -5.1928 0.0000 Invtota -1.4542 0.0729 lnta -8.9637 0.0000 fatota -2.9242 0.0017 ri -10.7401 0.0000 TABLE 18. KRUSKAL-WALLIS RANK TEST FOR ISLAMIC BANKS (IBS) DATASET VARIABLE STATUS OBS. RANK SUM [chi square] PROB. dps Before 82 12557.00 7.266 0.0070 After 185 23221.00 [div.sub.t-1] Before 65 8691.00 2.957 0.0722 After 176 20470.00 geps Before 65 8952.00 5.122 0.0236 After 176 20209.00 eps Before 82 14279.00 31.968 0.0001 After 185 21499.00 invtota Before 82 10960.00 0.002 0.9614 After 185 24818.00 lnta Before 82 8226.00 22.517 0.0001 After 185 27552.00 fatota Before 82 10121.50 2.216 0.1366 After 185 25656.50 ri Before 82 11843.00 2.158 0.1419 After 185 23935.00 TABLE 19. RESULT OF ESTIMATING MODEL (1) FOR THE 1ST DATASET OF CONVENTIONAL BANKS (CBS) CBS VARIABLE COEF. STD. ERR. t P>|t| [div.sub.t-1] 0.3320877 0.0376833 8.81 0.000 (***) geps 0.0000308 0.0004816 0.06 0.949 eps 0.1774903 0.0192387 9.23 0.000 (***) invtota 0.1328597 0.0826328 1.61 0.108 fatota 0.7859691 1.1458890 0.69 0.493 ri 0.0018788 0.0012197 1.54 0.124 _cons -0.0482333 0.0368321 -1.31 0.191 [R.sup.2](between) 0.8426 Number of observations F(6,561) 10.88 Number of panels Prob.>F 0.0000 VARIABLE [95% CONF. INTERVAL] [div.sub.t-1] 0.2580701 0.4061054 geps -0.0009150 0.0009767 eps 0.1397016 0.2152791 invtota -0.0294478 0.2951673 fatota -1.4647880 3.0367260 ri -0.0005170 0.0042746 _cons -0.1205790 0.0241124 [R.sup.2](between) 617 F(6,561) 50 Prob.>F Note: (***) Significant at 1% level, (**) significant at 5%, and (*) significant at 10%. TABLE 20. RESULT OF ESTIMATING MODEL (1) FOR THE 2ND DATASET OF CONVENTIONAL BANKS (CBS) VARIABLE COEF. STD. ERR. t P>|t| [div.sub.t-1] 0.0270311 0.1186488 0.23 0.820 geps -0.0011633 0.0055047 -0.21 0.833 eps 0.1618791 0.0336563 4.81 0.000 (***) invtota 0.0475939 0.1931634 0.25 0.806 fatota -2.1843960 2.4803760 -0.88 0.380 ri -0.0010567 0.0025095 -0.42 0.674 cons 0.1734069 0.0848220 2.04 0.042 (**) [R.sup.2](between) 0.2695 Number of observations F(6,169) 8.98 Number of panels Prob>F 0.0004 VARIABLE [95% CONF. INTERVAL] [div.sub.t-1] -0.2071936 0.2612557 geps -0.0120302 0.0097036 eps 0.0954382 0.2283200 invtota -0.3337301 0.4289178 fatota -7.0809060 2.7121150 ri -0.0060107 0.0038972 cons 0.0059597 0.3408541 [R.sup.2](between) 223 F(6,169) 48 Prob>F Note: (***) Significant at 1% level, (**) significant at 5%, and (*) significant at 10%. TABLE 21. RESULT OF ESTIMATING MODEL (1) FOR THE 3RD DATASET OF CONVENTIONAL BANKS (CBS) VARIABLE COEF. STD. ERR. t P>|t| [div.sub.t-1] 0.1331740 0.0346830 3.84 0.000 (***) geps 0.0002902 0.0003220 0.90 0.368 eps 0.2104372 0.0249786 8.42 0.000 (***) invtota 0.1710216 0.0870423 1.96 0.050 (**) fatota 1.7996050 1.5483340 1.16 0.246 ri 0.0025853 0.0012116 2.13 0.034 (**) _cons -0.0706368 0.0400253 -1.76 0.078 (*) [R.sup.2](between) 0.7673 Number of observations F(6,338) 19.78 Number of panels Prob>F 0.0000 VARIABLE [95% CONF. INTERVAL] [div.sub.t-1] 0.0649522 0.201396 geps -0.0003432 0.000924 eps 0.1613041 0.25957 invtota -0.0001912 0.342234 fatota -1.2459790 4.84519 ri 0.0002021 0.004969 _cons -0.1493668 0.008093 [R.sup.2](between) 394 F(6,338) 50 Prob>F Note: (***) Significant at 1% level, (**) significant at 5%, and (*) significant at 10%. TABLE 22. RESULT OF ESTIMATING MODEL (1) FOR THE 1ST DATASET OF ISLAMIC BANKS (IBS) VARIABLE COEF. STD. ERR. t P>|t| geps -0.0015464 0.0018472 -0.84 0.403 eps 0.1842663 0.0300565 6.13 0.000 (***) invtota 0.2628021 0.1308466 2.01 0.046 (**) lnta 0.0285231 0.0131881 2.16 0.032 (**) fatota -0.0544825 0.9028336 -0.06 0.952 ri 0.0021673 0.0012704 1.71 0.089 (*) _cons -0.2680603 0.1312568 -2.04 0.042 (**) [R.sup.2](between) 0.4235 Number of observations F(6,209) 7.27 Number of panels Prob>F 0.0000 VARIABLE [95% CONF. INTERVAL] geps -0.0051879 0.002095 eps 0.1250135 0.243519 invtota 0.0048539 0.52075 lnta 0.0025243 0.054522 fatota -1.8343100 1.725345 ri -0.0003371 0.004672 _cons -0.5268172 -0.0093 [R.sup.2](between) 241 F(6,209) 26 Prob>F Note: (***) Significant at 1% level, (**) significant at 5%, and (*) significant at 10%. TABLE 23. RESULT OF ESTIMATING MODEL (1) FOR THE 2ND DATASET OF ISLAMIC BANKS (IBS) VARIABLE COEF. STD. ERR. t P>|t| [95% CONF. INTERVAL] geps -0.0095361 0.0164576 -0.58 0.565 -0.0427259 eps 0.0695102 0.0735905 0.94 0.350 -0.0788992 invtota 0.1190729 0.5041191 0.24 0.814 -0.8975802 lnta 0.0498091 0.0440894 1.13 0.265 -0.0391057 fatota -0.7941536 2.0391940 -0.39 0.699 -4.9065800 ri 0.0022663 0.0037625 0.60 0.550 -0.0053215 _cons -0.3250799 0.3583253 -0.91 0.369 -1.0477120 [R.sup.2](between) 0.6516 Number of observations F(6,43) 0.61 Number of panels Prob>F 0.7177 VARIABLE [95% CONF. INTERVAL] geps 0.023654 eps 0.21792 invtota 1.135726 lnta 0.138724 fatota 3.318273 ri 0.009854 _cons 0.397552 [R.sup.2](between) 65 F(6,43) 16 Prob>F TABLE 24. RESULT OF ESTIMATING MODEL (1) FOR THE 1ST DATASET OF ISLAMIC BANKS (IBS) VARIABLE COEF. STD. ERR. t P>|t| geps -0.0021066 0.0011165 -1.89 0.061 (*) eps 0.3294499 0.0470530 7.00 0.000 (***) invtota 0.0882965 0.0868488 1.02 0.311 lnta -0.0323122 0.0168495 -1.92 0.057 (*) fatota -0.1171132 0.8354682 -0.14 0.889 ri -0.0004156 0.0009566 -0.43 0.665 _cons 0.3207109 0.1619881 1.98 0.050 (***) [R.sup.2](between) 0.4539 Number of observations F(6,144) 8.40 Number of panels Prob>F 0.0000 VARIABLE [95% CONF. INTERVAL] geps -0.0043135 0.0001 eps 0.2364460 0.422454 invtota -0.0833668 0.25996 lnta -0.0656165 0.000992 fatota -1.7684790 1.534253 ri -0.0023064 0.001475 _cons 0.0005293 0.640893 [R.sup.2](between) 176 F(6,144) 26 Prob>F Note: (***) Significant at 1% level, (**) significant at 5%, and (*) significant at 10%. TABLE 25. SUMMARY RESULTS OF THE CAUSAL MODEL HYPOTHESIS SUPPORTED FOR CBS DATASET PREDICTOR ENTIRE BEFORE AFTER [div.sub.t-1] ([check]+***) ([check]+***) geps eps ([check]+***) ([check]+***) ([check]+***) invtota ([check]+**) lnta fatota ri ([check]+**) HYPOTHESIS SUPPORTED FOR IBS DATASET PREDICTOR ENTIRE BEFORE AFTER [div.sub.t-1] geps ([check]+*) eps ([check]+***) ([check]+***) invtota ([check]+**) lnta ([check]+**) ([check]+*) fatota ri ([check]+*) Note: ([check]) Hypothesis is supported; (+) Positive effect; (***) Significant at 1% level, (**) significant at 5%, and * significant at 10%.

Printer friendly Cite/link Email Feedback | |

Author: | Aldeehani, Talla M. |
---|---|

Publication: | Business and Economic Horizons |

Geographic Code: | 7SAUD |

Date: | Jan 1, 2019 |

Words: | 10073 |

Previous Article: | The value of education in the labour market. How realistic are student expectations? |

Next Article: | Cointegration and causality between the GCC stock indices and gold indices. |

Topics: |