Dividends and risk in European banks

Enrico Onali+

Abstract Ceteris paribus, a large dividend payout ratio decreases the capital ratio of a bank. Under deposit insurance regulation, banks with a low capital ratio are encouraged to take on risk. I investigate the relation between dividends and risk in banking, using a sample of 335 banks for the period 2000-2007. Contrary to the extant literature about nonfinancial firms, I find evidence that dividends are positively related to default risk, and negatively related to retained earnings. Similar to nonfinancial firms, dividends are related to insider/outsider agency issues, profitability, and size.

Keywords: Dividend, Bank Risk, Monitoring, Agency Costs JEL classification: G21, G35

An earlier version of this paper was presented at the 22 nd Australasian Finance and Banking Conference (AFBC, Sydney, December 2009), and the UKEPAN conference (Leicester, November 2009). I am indebted to John Goddard and Klaus Schaeck (Bangor University) for reading the paper several times and for giving me valuable feedback. I also wish to thank Balasingham Balachandran (AFBC discussant), Owain ap Gwilym, Philip Molyneux, Lynn Hodgkinson, Angelo Zago, Graham Partington, and seminar participants at Bangor University in November 2009.

Bangor Business School, Bangor University, Gwynedd, LL57 2DG, United Kingdom

e-mail address: e.onali@bangor.ac.uk

1. Introduction Banks are currently subject to intense criticism for their role in the global financial crisis. Inadequate scrutiny of the behaviour of bank executives and senior managers by non-executive directors and large shareholders (insiders) has been one of the causes of the crisis (Treasury Committee, 2009). Bank risk and the dialectic between insiders and outsiders are closely related. Dividends help outsiders monitor insiders because: they lead to more frequent equity issues which imply market scrutiny (Easterbrook, 1984); they discourage the use of financial resources for empire building and perquisites (Jensen, 1986). However, large dividend payout ratios reduce the ability of a bank to build a solid capital buffer. Under deposit insurance regulation, a low capital ratio encourages risk taking (Merton, 1977). Therefore, banks that pay large dividends tend to have low capital ratios, an incentive for risk taking. The nexus between dividend policy and bank risk is crucial, as suggested by a recent statement by the Financial Stability Board: Although many financial institutions have returned to profitability in recent quarters, [] it is important that firms retain these profits in order to rebuild capital to support lending after official support measures have been removed (Financial Stability Board, 2009, p. 1). Banks can replenish their capital hit by the financial crisis of 2007-2009 either by retaining earnings or by issuing new capital. During 2007-2009 many new capital issues have been in the form of hybrid instruments (included in tier 2 of the regulatory capital required by the Basel Accord) rather than common equity capital. Because these instruments do not constitute equity in the sense of residual claim of the shareholders, they imply higher risk for debt holders and incentivise leveraging and risk taking on the part of bank owners (Acharya et al., 2009). This phenomenon takes place because common equity represents a call option on the ownership of a bank, whose exercise price is represented by the value of debt capital (Merton, 1974): if the value of the assets is lower than that of the liabilities, the value of the option (or common equity) is zero. Increasing the fraction of assets funded by capital other than common equity increases the exercise price up to a point where the value of the option is close to zero. Owners of highly-leveraged banks have nothing to lose, and engage in excessive risk taking. Dividends play an important role in this model, as they decrease the value of assets, which implies a decrease in the value of both equity and debt, but benefit only the owners of the bank. Dividends, as well as hybrid instruments, effectively shift the risk from the owners to the debt holders. Recently, it has been suggested that restrictions on dividends should be included in a set of ladder of sanctions for banks that do not satisfy certain regulatory requirements in terms of solvency and liquidity (Brunnermeier et al., 2009). Regulations on dividends would help prevent excessive risk taking because it would force bank owners to bear a substantial part of this risk.

Despite the important role of dividends, the literature on the dividend policy of banks is rather sparse. I examine the relationship between dividends and risk in banks using a sample of 335 banks during 2000-2007. I focus on two measures of risk: default risk and credit risk. I use the Z-score (Boyd and Graham, 1988) to pick up the effects of default risk and the loan loss provision ratio to proxy for credit risk. Because of the link between risk and monitoring, and the role of dividends as a monitoring device, I investigate whether the severity of the conflict between insiders and outsiders (agency problem) increases the level of dividends. I examine whether practices of earnings and capital management may influence the relation between dividends and variables employed to capture risk. I test the life-cycle theory of dividends by examining the role of earned equity. To account for the large number of cases for which the dividend is zero, I use the Tobit regression model. I find evidence of a positive relationship between dividends and default risk. Retained earnings are negatively related to dividends, contrary to the life-cycle theory. Banks where the insiders-outsiders conflict is more severe distribute more dividends. The results for default risk and retained earnings may suggest that, unlike commercial firms, banks need to retain a large portion of their earnings to be sound. Low retention rates result in high default risk. This is an important finding, especially in light of the current debate on the need for banks to avoid paying large dividends. My paper provides also new insights on the possible divergences and similarities in the dynamics of financing decisions of banks and nonfinancial firms. The paper is organised as follows. Section 2 reviews the previous evidence on dividend policy, focussing on the importance of risk and monitoring. Section 3 describes the methodology and the data set. Section 4 reports the main results. Section 5 examines the impact of accounting manipulation on the proxy for credit risk, and investigates the life-cycle theory of dividends. Finally, section 6 summarises and concludes. 2. Related literature and hypotheses Recent literature has investigated whether regulation in the financial sector (in particular deposit insurance and capital adequacy regulation) impinges on the determinants of the financing decisions of banks (Gropp and Heider, 2009). This paper assumes a similar perspective in that it investigates the dynamics of the relationship between dividends and risk. While dividends may impair the long-term ability of a bank to build a solid capital buffer, they may be used by outsiders to monitor insiders. Despite the importance of dividend policy for bank risk, this topic has been rather overlooked by the literature. Bessler and Nohel (1996, 2000) and Cornett et al. (2008) focus on the signalling content of dividends. Casey and Dickens (2000) and Casey et al. (2002) investigate the determinants of the dividend payout ratio and dividend yield, respectively. Boldin and Leggett (1995) investigate the relation between dividends and bank rating. These studies focus on US banks only and provide mixed results as to how dividends relate to bank risk. In sections 2.1 and 2.2 I report relevant literature on the nexus between dividends and bank risk and on the monitoring function of dividends. I present two testable hypotheses built upon such literature.

2.1 Dividends and bank risk Literature on the dividend policy of nonfinancial firms argues that, other things being equal, risk should reduce dividend payments (Rozeff, 1982; Bar-Yosef and Huffman, 1986). Is this true even in banking? Deposit insurance regulation incentivises risk taking in banking (Laeven, 2002). Deposit insurance is equivalent to a put option on the banks asset (Merton, 1977). Accordingly, the value of deposit insurance is positively related to default risk. Dividend policy affects the ability of a bank to build a solid capital buffer: ceteris paribus, a large dividend payout ratio constrains the ability of a bank to retain earnings. Due to the nexus between the dividend payout ratio, capital ratio and the value of deposit insurance, a large dividend payout ratio is likely to lead to high default risk and, because of the positive impact of default risk on the value of deposit insurance, encourages further risk taking. In this framework, prudential capital adequacy regulation should ensure that banks do not engage in excessive risk taking. However, minimum capital requirements could backfire because restrictions on leverage could be circumvented by decreasing asset quality. For example, Iannotta (2006) finds that asset quality is negatively related to capital ratio. For this reason, the capital ratio is probably not a good proxy for default risk. The banking literature commonly employs the Z-score as a measure of default risk (Boyd and Graham, 1988). A more specific measure of risk, related to the loan portfolio, is the ratio loan loss provision to total loans (Iannotta et al., 2007). The literature on dividend policy of nonfinancial firms employs measures of risk such as the beta (Rozeff, 1982), or the standard deviation of residuals from a regression of daily stock returns on returns of the market portfolio (Hoberg and Prabhala, 2009; Li and Zhao, 2008). Other measures are the standard deviation of stock returns or the residuals of a regression of excess returns on the three Fama and French (1992) factors. However, I cannot employ these measures as proxies for risk of unlisted banks, since they require stock returns data. The Z-score includes the standard deviation of accounting returns (returns on assets) in the denominator, and accounting returns and the ratio equity to total assets in the numerator (see table 1). Given the importance of equity for banks, the Z-score appears a much more reliable measure for risk than the standard deviation of accounting returns alone. On the grounds of the impact of deposit insurance regulation on the pricing of a banks assets, the first testable hypothesis of this paper is as follows: H1: bank default risk and credit risk are positively related to dividends. I test H1 by assessing the effect of the Z-score and loan loss provision ratio on the dividend payout ratio.

2.2 Dividends and monitoring In the US, the agency problem refers to the conflicting interests of managers and shareholders. In Western Europe, where many corporations and banks are not publicly held, insider shareholders are so close to the management that the crucial agency problem is between insiders (managers and large shareholders) and outsiders (Faccio et al., 2001). Bank insiders benefit from privileged information with respect to outsiders. Insiders use this information to forecast the future profitability and growth of the bank. This information asymmetry can increase the cost of equity capital. Therefore, it is in the insiders interest to reduce such asymmetry as much as possible. Monitoring reduces information asymmetry, but is costly. Where share ownership is widely dispersed, there is a free-rider problem that discourages outsiders monitoring of insiders (Grossman and Hart, 1980). The conflict between insiders and outsiders may be reduced by paying dividends (Easterbrook, 1984; Jensen, 1986). This monitoring rationale is one of the reasons why dividends are paid. However, dividends are not the only monitoring mechanisms available to outsiders. If other mechanisms exist, dividends may lose their monitoring function (Noronha et al., 1996). This may occur when there is a large outsider shareholder whose incentive to monitor insiders is high (Shleifer and Vishny, 1986), or when the interests of insiders and outsiders are aligned (for example, in the presence of performance-related compensation packages for managers). For nonfinancial firms, loan intensity from relationship banks increases monitoring and decreases payout ratios (Allen et al., 2009). In banking, regulation may provide an alternative monitoring device for outsiders (Filbeck and Mullineaux, 1999). Because of the obvious link between risk and monitoring, I examine the validity of the monitoring rationale of dividends in banking. Monitoring does not necessarily imply less risk when it refers to the conflict between managers and shareholders. Saunders et al. (1990) stress that managers may be risk averse, while bank shareholders are encouraged to increase bank risk under deposit insurance regulation (Merton, 1977). When the interest of managers is aligned to that of shareholders (i.e. if there is a large shareholder or if managers hold a large shareholding) risk taking may increase. As said before, this is likely to be the case in Europe, and therefore risk-averse outsiders may use dividends to monitor risk-seeking insiders.1 My second testable hypothesis is: H2: Dividends are positively related to the severity of the insiders-outsiders (IO) conflict. I test H2 using three proxies for the degree of the IO conflict: the listing on a stock exchange, the absence of a shareholder with more than 25% of the voting rights, and multiple recorded shareholdings. These proxies are positively related to ownership dispersion, which increases the severity of the IO conflict. Results supporting a positive relation between dividends and ownership dispersion would support the hypothesis that dividends can be used by outsiders to monitor insiders.
1

An alternative view to the monitoring rationale for dividends is the rent extraction hypothesis, which posits that dividends are paid to avoid expropriation from large shareholders (Faccio et al., 2001).

3. Methodology and data 3.1 Methodology I investigate the nexus between dividends and risk using the Tobit model. The Tobit model allows me to account for the large number of cases for which the dividend is zero. Following Allen et al. (2009) I assume that the determinants of the decision to pay a dividend are the same as the determinants of the level of the dividend. I estimate a single equation for the dividend payout ratio (the ratio dividends to earnings, DP, and the ratio dividends to equity, DPE) treating all observations for which the dividend is either zero or missing2 as censored. I employ two measures for the dividend payout ratio because: for comparability with previous literature (Rozeff, 1982), I examine the effects of risk on DP; since the use of the ratio dividends to earnings as a dependent variable may cause problems when earnings are zero or negative, DPE offers a sensible alternative. I prefer equity to other possible variables such as total assets given the importance of equity capital in banking. 3 The specification of the model is as follows. Yit is an observable variable, which depends on a latent variable Yit* according to: Yit = Yit* Yit = 0 if Yit* > 0 if Y 0
* it

(1)

where Yit* is the latent variable:

Yit* = 'xit + 'cit + it

(2)

where and are vectors of coefficients, xit is a vector of covariates which proxy for risk (to test H1) and the IO-conflict level (to test H2) and cit is vector of control variables. Due to multiple observations for each bank, the error term in (2) could be heteroskedastic and the OLS model inefficient. To allow for such eventuality, I use a Random Effects (RE) model:4

it = i + it

(3)

Bankscope reports a missing value if the annual report of a bank does not report the amount of the dividend. It is very unlikely that paid dividends do not appear in the annual report of a bank. Therefore, I consider missing dividends for observations for which a value for the net income is reported as cases for which the bank did not pay dividends. This criterion enables me to enlarge the number of observations available and reduce sample selection bias. 3 I am grateful to John Goddard for this remark. 4 The collinearity in variables with data available only for 2007 (e.g. S i) prevents me from using a Fixed Effects (FE) model. Moreover, the currently available fixed-effect tobit models (Honore, 1992) are biased.

2 where it ~ N(0,2), and i ~ N(0, ) is an idiosyncratic time-invariant effect, assumed to be randomly distributed and uncorrelated with xit and cit.

The ratio dividends to earnings tends to be quite volatile because while dividends tends to be sticky (Lintner, 1956), earnings (and as a result equity) may be very volatile. I consider the effect of extreme observations on the relationship between dividends and bank risk in the following univariate and multivariate analysis. Section 3.2 defines the variables that constitute xit and cit. Section 3.3 describes the data. 3.2 Definition of the explanatory variables H1 posits that there is a positive relationship between dividends and two facets of bank risk: default risk and credit risk. The covariates that comprise the vectors xit and cit are defined in table 1. The Z-score (Boyd and Graham, 1988) is negatively related to default risk. Credit risk is proxied by the ratio loan loss provision to total loans. According to H1, the Zscore should be negatively related to the dividend payout ratio, while credit risk should be positively related to the dividend payout ratio. The number of recorded shareholders, the listing on a stock exchange, and the absence of a shareholder with more than 25% of the voting power proxy for the severity of the IO conflict. These proxies are positively related to ownership dispersion. For convenience, I refer to banks where there is no shareholder with more than 25% of the voting rights as Independent banks. According to H2, dividends should be positively related to the severity of the IO conflict. Therefore, the coefficients on my three proxies should be positive. Because of positive and significant correlation between these three variables, I insert them one at a time in the Tobit regressions. The extant literature finds that loans growth, size, and profitability influence the dividend payout ratio. Accordingly, I include several control variables in equation (2) to account for the impact of these factors. I expect a negative coefficient on loans growth because banks that are growing rapidly are likely to retain to more cash than banks that lack growth opportunities (Fama and French, 2001; Rozeff, 1982). Size and profitability (ROA) should have a positive coefficient, because small banks and unprofitable banks are less likely to distribute dividends (Denis and Osobov, 2008; DeAngelo et al., 2004; Fama and French, 2001). Given the multi-country nature of my study, I also control for possible country effects. According to the outcome model of agency theory (La Porta et al., 2000), the legal framework under which the bank operates influences dividend policy. Banks in countries where there is a strong protection for minority shareholders (typically, countries whose legal system is based on common law) should pay larger dividends. Minority shareholders whose rights are inadequately protected may lack the

necessary legal power to induce insiders to pay dividends. 5 I use three country dummy variables (for France, Germany, and Italy) to proxy for the level of protection of minority shareholders rights. I choose the UK as the reference category because it is the only country under investigation where the protection of minority shareholders rights is strong (La Porta et al., 2000). Accordingly, I expect the coefficients on the three country dummies to be negative. [Insert Table 1 here] 3.3 Data I collect consolidated bank accounts data for 335 banks (32 listed) in the four European G-7 countries (France, Germany, Italy and UK) from the Bureau Van Dijk Bankscope database. The sample period is 2000 to 2007. Table 2 reports descriptive statistics for each country and for the whole sample. The first section of the table reports the number of banks in the sample, the number of listed banks, the number of observations for positive, negative and zero DP (dividends/earnings), the number of observations for positive and zero DPE (dividends/equity), and the average DP and DPE. The number of zero payout ratios includes missing values for which it is likely that no dividend was paid. 6 Negative DP refers to cases for which the earnings were negative but a dividend was paid. 7 Most of the banks in the sample are domiciled in the UK (39%). Italian and German banks constitute only 15% and 18% of the sample, respectively. The relatively low number of Italian and German banks is due to the lack of consolidated accounts data for these banks in Bankscope. UK banks have the highest average DP, while Italian banks have the lowest DP. French banks have the highest DPE, while German banks have the lowest DPE. The average DP and DPE for the whole sample are 16.59% and 2.45%, respectively. The second and third sections of table 2 report the mean, standard deviation, and various percentiles of the distribution of the positive DP and DPE. The fourth and fifth sections report the mean, standard deviation, and 50th percentile (median) of the Z-score and loan loss provision ratio. French banks have the highest average positive DP and DPE, while Italian (German) banks have the lowest positive DP (DPE). The value of the median DP is very similar for Germany, Italy and the UK (between 45% and 49%), while for France it is much higher (64%). The value of the median DPE is also highest for France. Dispersion, in terms of both standard deviation and the difference between
5

An alternative to the outcome model is the substitute model: in countries with weak legal protection for minority shareholders, companies use dividends as a means to establish a reputation. The substitute model predicts higher dividends for countries with weak legal protection for minority shareholders (La Porta et al., 2000). 6 A missing value is likely to indicate that no dividend was paid if a value for earnings is reported. Missing values for both dividends and earnings indicate either that the bank did not exist in that year, or that the data were not recorded by Bankscope. 7 All banks that paid a dividend when earnings were negative are from the UK. Two of them are Bank Holding Companies (Ansbacher Overseas Group Ltd and Butterfield Holding Ltd), while the other three are Commercial Banks. Among the latter is Northern Rock Plc, which was bailed out by the UK government in 2007.

the 99th and 1st percentile, is highest in the UK for DP and in France for DPE, and lowest in Italy for DP and Germany for DPE. According to the statistics for the Z-score, UK banks are the most sound, and German banks the least sound. German banks also tend to have a much higher credit risk than banks in the other countries. However, a comparison of the mean and median value for Germany suggests the presence of outliers. I consider the possible influence of these outliers in the subsequent investigation of the determinants of the payout ratio. The descriptive statistics in table 2 are only partially consistent with La Porta et al. (2000). According to the outcome model of agency theory, UK banks should pay higher dividends than banks in Continental Europe. An alternative explanation of country differences in the dividend payout ratios could be the heterogeneous taxation for dividends and capital gains. However, the Tax Advantage of Dividends (TAD) index (La Porta et al., 2000) is higher for German banks than for UK banks. German banks should pay more dividends than UK banks, followed by Italian banks and finally French banks. This is in contrast with what reported in table 2. Moreover, computing the tax effect on dividends is extremely tricky, due to the variation in the actual tax rate paid by individuals and companies in different tax brackets. While concentrating the analysis on both dividends and stock repurchases may seem to solve this problem, this assumes that all the other features of dividends are held constant when considering stock repurchases. This condition is not verified in reality. For instance, stock repurchases are non-binding (unlike dividends), and therefore do not contain the same signalling content of dividends (Allen et al., 2009). [Insert Table 2 here] Table 3 reports the pair-wise correlations between DP, DPE and the continuous explanatory variables in table 1. Five observations for which the DP is negative (earnings are negative but a dividend is paid) and 21 observations for which earnings are zero are excluded. When the DP is negative, large dividends cause the payout ratio to fall instead of increase. This inverse relation between dividends and DP is counterintuitive and the analysis for negative payout ratios should be separated from the analysis of positive payout ratios. When earnings are zero and a dividend is paid the DP is infinite. In my sample there are no such cases, because no dividends are paid when earnings are zero. However, for 21 observations both the dividend and the earnings are zero: given that 0/0 does not exist, I exclude these 21 observations. To allow for the effects of earnings volatility and extreme risk conditions, I run the pair-wise correlations on two sub-samples. The first sub-sample (upper part of the table) includes only observations for which the DP and the loan loss provision ratio (credit risk) are lower or equal than their respective 95th percentiles. Therefore, these correlations take into account cases where the payout ratio and credit risk are under relatively normal conditions. I refer to the pair-wise correlations in the first subsample as model 1. The second sub-sample (lower part of the table) comprises all observations (excluding those for negative DP and zero earnings) and therefore the correlations capture the effects of extreme DP and very high credit risk. It is interesting to see whether these exceptional cases are so important as to drive the correlation

between DP and credit risk. I refer to the pair-wise correlations in the second subsample as model 2. In model 1 DP and DPE are, of course, strongly positively correlated (0.6469). DP is negatively correlated with the Z-score (default risk) and positively correlated with size and profitability. Consistent with H1, default risk is positively related to the payout ratio. However, the correlation between DP and credit risk is negligible. Similarly, DPE is positively correlated with profitability and size, and negatively correlated with the Zscore. However, the correlation with the Z-score is weakly significant (significant at the 10% level). The correlation between DPE and credit risk is also negligible. Credit risk is positively correlated to size and negatively correlated to profitability. Large unprofitable banks bear higher credit risk than small profitable banks. This finding may be due to the too big to fail phenomenon, and may suggest possible moral hazard for large banks. In model 2 the correlation between DP and DPE is stronger than in model 1 (0.6477). The correlation between the DP and the Z-score maintains its sign and significance, and the correlation between DPE and the Z-score increases and becomes strongly significant (1% level). Moreover, the correlation between the DP and credit risk becomes positive and significant, while the correlation between DPE and credit risk remains insignificant. This could be due to the negative relationship between net income and loan loss provision (when earnings smoothing is negligible), as loan loss provision pushes down net income. Considering equity in the denominator of the payout ratio rather than earnings alleviates this problem. The correlation between DP and profitability and size remain positive and significant, although the significance of the relationship with size becomes weak (10% level). The correlation between DPE and profitability and size remain positive and significant. The results shown in table 3 suggest that outliers may drive the correlation between DP and credit risk, and between DPE and the Z-score. Accordingly, I exclude from the subsequent multivariate analysis observations for which DP, DPE, and the loan loss provision ratio are higher than the 95th percentile. [Insert Table 3 here] 4. Results This section presents regression results of my Tobit models. Throughout the discussion, I refer to results significant at the 5% level or better as significant, while I use weakly significant for results that are significant at the 10% level or better. Subsection 4.1 reports the main findings. Sub-section 4.2 reports robustness checks. 4.1 Main results Table 4 reports the estimation results for four Tobit regressions with Random Effects. As before, I exclude observations for which the DP is negative or earnings are zero. I consider only observations for which the DP/DPE and the loan loss provision ratio do not exceed their respective 95th percentile.

10

The upper section of table 4 reports estimation results for models 1 to 4 when DP is used as the dependent variable for equation (2). The results confirm that there is a positive relation between dividends and default risk. However, credit risk is not related to dividends. H1 is only partially confirmed. The coefficients on the three proxies for the IO conflict are positive and significant, consistent with H2. The coefficients on the control variables are either significant with a sign consistent with my expectations or insignificant. As an indication of the economic significance of the results for the Z-score (using column 1 of table 4), a change in the Z-score by one standard deviation (65.98 see table 2), is related to a change in the DP by 13% for all firms (-0.0019 65.98). The average DP for the whole sample (considering observations for which DP is zero) is 16.69%. Therefore, the association between variations in the Z-score on one side and the choice to pay dividends and the amount of dividends on the other side is economically significant. 8 The lower section of table 4 reports estimation results for models 1 to 4 when DPE is used in place of DP as the dependent variable for equation (2). Contrary to what reported in the upper section of table 4, the coefficient on credit risk is positive and significant for model 4 and weakly significant for models 1 to 3. The coefficients on the Z-score remain significant for models 1 to 3, but for model 4 the coefficient is weakly significant. The coefficients on the three IO-conflict proxies are positive and significant. The estimation results for the control variables remain virtually unaltered. As an indication of the economic significance of the results for the Z-score (using column 1 of table 4), a change in the Z-score by one standard deviation (65.98 see table 2), brings about a change in the DPE by 1.32% for all firms (-0.0002 65.98). The average DPE for the whole sample (considering observations for which DPE is zero) is 2.45%. These figures confirm the economic significance of the association between dividends and default risk, already obtained with respect to DP. 4.2 Robustness checks I run further robustness checks to verify whether H1 and H2 are sensitive to the specification of my models. The results are not reported, but are available upon request. First, the inclusion of the 21 observations for which earnings are zero does not change substantially the results reported in table 4 for DP. Credit risk is not related to the payout ratio while default risk is positively related to the payout ratio. The proxies for the IO-conflict are still positive and significant. The coefficients on the control variables are either insignificant or significant with a sign consistent with my expectations. Second, I replace the ratio loan loss provision to total loans with the ratio loan loss provision to equity. Credit risk is not related to either DP or DPE. In table 4, the relationship between credit risk and DP was positive and weakly significant for models 1 and 4; the relationship between credit risk and DPE was significant at the 5% for model 4 and at the 10% for models 1 to 3. Therefore, using equity in place of loans in
8

I prefer using the term association rather than causality as it is hard to determine whether the Z-score causes variations in the dividends payout ratios or vice versa.

11

the denominator of the loan loss provision ratio weakens the already fragile relationship between credit risk and the dividend payout ratio. The results for the Z-score and the other variables remain virtually unaltered. Third, I drop credit risk from the regressions on DP and DPE, since it does not seem to be related to either DP or DPE. For the regressions on DP the coefficient on the Z-score is still significantly negative. The results for the other explanatory variables remain virtually unaltered. For the regressions on DPE the coefficient on the Z-score is still significantly negative, but for models 3 and 4 the significance is weak (10%). The results for the other explanatory variables remain virtually unaltered. Fourth, I include seven year-dummies (eight minus one to avoid perfect collinearity) to allow for variations in dividends due to business-cycle conditions. The year effects are significant when either DP or DPE is the dependent variable. In the regressions for DP, the coefficient on the Z-score is still negative and significant, and the coefficient on credit risk is positive and weakly significant for models 1, 3, and 4 and insignificant for model 2. The results for the other variables remain virtually unaltered. In the regressions for DPE, the coefficient on the Z-score is negative and significant, although for models 2 to 4 the significance is weak (10%). The coefficient on credit risk is positive and significant, although for models 1 to 3 the significance is weak (10%). The coefficients on the other variables do not vary considerably. Fifth, I allow for the effect of Initial Public Offerings (IPOs). Banks that have just gone public are more likely to initiate dividends than nonfinancial firms (Cornett et al., 2008). 16 banks in my sample went public in the period 2000-2007. Accordingly, I create a binary variable (IPO) that takes on the value 1 when a bank becomes public in that year and 0 otherwise. I include the variable IPO in models 1-4, with either DP or DPE as the dependent variable. The coefficient on IPO is positive but insignificant for all eight regressions (four for DP and four for DPE). The results for the other explanatory variables do not vary considerably. 5. Further issues: accounting manipulation and the life-cycle theory of dividends 5.1 The effects of earnings and capital management practices It has been suggested that some banks manipulate the loan loss provision for purposes of income-smoothing and capital management, although the empirical evidence is mixed (Collins et al., 1995; Ahmed et al., 1999). A positive relationship between the ratios of non-discretionary earnings to loans and loan loss provision to loans suggests there is manipulation of the loan loss provision, in order to increase (decrease) net income when non-discretionary earnings are low (high). Loan loss provision decreases earnings but increases regulatory capital (since loan loss reserves are comprised in Tier 2 capital). Therefore, the ability of the loan loss provision to proxy for credit risk may be impaired by practices of earnings and capital management. Table 5 reports estimation results for models 1 to 4, incorporating an attempt to adjust for earnings and capital management practices of the kind described above.

12

Credit risk is replaced in equation (2) by the residuals of a regression of the loan loss provision ratio on the ratio non-discretionary earnings (net income plus loan loss provisions) to loans and the country dummies. In this regression, clustered standard errors allow for possible autocorrelation over the observations pertaining to each bank. The results in Table 5 suggest that the relationship between the adjusted loan loss provision ratio and DP/DPE is insignificant for all regressions. The estimation results for the other explanatory variables remain virtually unaltered: the Z-score is negatively related to DP and DPE, but for the regressions on DPE the coefficient on the Z-score is weakly significant for models 3 and 4; the coefficients on the IO-conflict proxies are positive and significant; the coefficients on the other explanatory variables are either significant with a sign consistent with my expectations or are insignificant. Similar to what I have done in section 4.2, I run robustness checks to improve the reliability of my findings. The results are not reported but are available upon request. First, I allow for the effect of business-cycle conditions using seven yeardummies in addition to the variables already included in my model. The year effects are significant when either DP or DPE is the dependent variable. For the regressions on DP, the coefficients on the adjusted loan loss provision ratio remain insignificant. The coefficients on the Z-score remain significantly negative, although the coefficient on the Z-score for model 4 is only weakly significant. The coefficients on the other explanatory variables are either significant with a sign consistent with my expectations or are insignificant. For the regressions on DPE, the coefficients on the adjusted loan loss provision ratio remain insignificant. The coefficient on the Z-score for model 1 is significantly negative. The coefficients on the Z-score for models 2-4 are negative but only weakly significant. The coefficients on the other explanatory variables are either significant with a sign consistent with my expectations or are insignificant. Second, I allow for the effect of Initial Public Offerings (IPOs). The coefficients on IPO are positive but insignificant for all four regressions on DP and for all four regressions on DPE. For the regressions on DP, the coefficients on credit risk are insignificant, while the coefficients on the Z-score are significantly negative. The coefficients on the other explanatory variables remain virtually unaltered. For the regressions on DP, the coefficients on credit risk are insignificant, while the coefficients on the Z-score are negative and significant for models 1 and 2 and negative and weakly significant for models 3 and 4. The coefficients on the other explanatory variables remain virtually unaltered. 5.2 The life-cycle theory of dividends: do retained earnings matter? DeAngelo et al. (2006) find that the proportion of equity that consists of retained earnings explains dividend payments, supporting the life-cycle theory of dividends. When most of the equity capital is earned rather than contributed dividend payments are more likely. Mature firms tend to have cumulated enough profits to be able to pay dividends. Newly established firms tend to rely on external finance to support their growth. The ratio retained earnings to equity (RETE), a proxy for earned equity, is found to increase the likelihood that a dividend is paid in US nonfinancial firms. However, von Eije and Megginson (2006) do not find a significant relationship between

13

RETE and dividends in Europe. In banking, retained earnings may be more important than in nonfinancial firms to accumulate enough equity capital. Banks may be willing to forego the benefits on dividends (for example, signalling) because of the importance of a solid capital buffer for their reputation, and to stave off regulatory interventions. To examine the impact of RETE on DP and DPE, I run models 1 to 4 with this additional explanatory variable. Before running regressions 1 to 4, I check whether the summary statistics for RETE in my sample are significantly different from those of related studies. Unfortunately, von Eiye and Megginson (2006) do not provide such statistics. Denis and Osobov (2008) report the average ratio retained earnings to equity for France, Germany and the UK over three periods: 1989-1993, 1994-1998, and 1999-2002. Financial firms are excluded from their sample. For the third period (the closest to my sample period) the average ratios are: 57% for France, 23% for Germany, and 38% for the UK. In my sample, the average ratios are: 44% for France, -36% for Germany, 25% for Italy, and 72% for the UK. For Germany, the negative average ratio is due to one very large negative outlier. Once this outlier is excluded, the average ratio is 24%. To alleviate the influence of extreme observations on my estimation, I only consider values for RETE that are above the 5th percentile and below the 95% percentile of the distribution. Table 6 reports estimation results for models 1 to 4 including RETE as an explanatory variable. The number of banks for which data on this variable are available is much smaller than for the other regressors. For the regressions on DP, I have 211 banks and 515 observations. Surprisingly, the coefficients on RETE are all significantly negative. Similar to what reported in tables 4 and 5, credit risk is not significantly related to dividends, while default risk is positively related to dividends. The coefficients on the other explanatory variables are either significant and with a sign consistent with my expectations or insignificant. For the regressions on DPE, I have 207 banks and 512 observations. All coefficients for RETE are significantly negative. The coefficients on credit risk are insignificant, although for model 4 the coefficient is weakly significant. Default risk is positively related to dividends. The coefficients on the other explanatory variables are either significant and with a sign consistent with my expectations or insignificant. These results, along with those for default risk, may indicate that in banking retaining earnings is more important than in other industries. Retained earnings are a result of conservative dividend policy, which is also associated with low default risk. A confirmation of this hypothesis is given by an examination of the pairwise correlations that RETE exhibits with DP, DPE, and the Z-score for different values of RETE. 9 For 0.50 < RETE < 1.00, the correlation with DP is 38.67% (significant at the 1% level), and the correlation with DPE is 34.47% (significant at the 1% level). For 0 < RETE < 0.50, the correlation with DP is 1.49% (insignificant), and the correlation with DPE is 1.75% (insignificant). The pairwise correlations between RETE and the Z-score are as follows: for 0.50 < RETE < 1.00, the correlation is 27.38% (significant at the 1% level);
9

I have chosen the thresholds 0, 0.50, and 1 for RETE because they generate two sub-samples of roughly 250 observations.

14

for 0 < RETE < 0.50, the correlation is 9.10% (insignificant). Therefore, large RETE ratios are associated with low default risk and low payout ratios. Retaining large percentages of profits over time improves the soundness of the bank. However, when RETE is relatively low, the relationship with dividends and default risk becomes weaker. Similar to what I have done in section 4.2, I run robustness checks to improve the reliability of my findings. The results are not reported but are available upon request. First, I allow for the effect of business-cycle conditions using seven yeardummies in addition to the variables already included in my model. The year effects are significant when either DP or DPE is the dependent variable. For the regressions on DP, all four coefficients on RETE remain significantly negative, although for model 4 the coefficient on RETE is weakly significant. The coefficient on the dummy for listed banks becomes insignificant. The coefficients on credit risk are insignificant, while the coefficients on the Z-score are negative and significant. The results for the other explanatory variables are either significant with a sign consistent with my expectations or are insignificant. For the regressions on DPE, all four coefficients on RETE remain significantly negative, although for model 4 the coefficient on RETE is weakly significant. The coefficients on credit risk are insignificant, although the coefficient on credit risk for model 4 is weakly significant. The coefficients on the Z-score remain negative and significant. The coefficients on the other explanatory variables are either significant with a sign consistent with my expectations or are insignificant. Second, I allow for the effect of Initial Public Offerings (IPOs). The coefficients on IPO are positive but insignificant for all four regressions on DP and for all four regressions on DPE. The coefficients on RETE remain significantly negative for all four regressions on DP and all four regressions on DPE. For the regressions on DP, the coefficients on credit risk are insignificant, while the coefficients on the Z-score are significantly negative. The coefficient on listed is positive and weakly significant. The coefficients on the other explanatory variables remain virtually unaltered. For the regressions on DPE, the coefficients on credit risk are insignificant, although for model 4 the coefficient is weakly significant. The coefficients on the Z-score remain negative and significant. The coefficients on the other explanatory variables are either significant with a sign consistent with my expectations or are insignificant. 6. Conclusions and policy recommendations This paper investigates the relation between dividends and bank risk. A positive relationship between risk and dividends, contrary to what found for nonfinancial firms, could be attributed to the existence of deposit insurance regulation, which encourages risk taking in banking. I employ the Tobit model to account for a large number of cases for which dividends are zero. I provide evidence which supports the view that default risk, proxied by the Z-score, is positively related to dividends. Retained earnings are negatively related to dividends. These results are contrary to what found by empirical studies on nonfinancial firms, and suggest that in banking large retention rates are important to build large capital buffers and establish reputation among the public.

15

I also investigate the relation between dividends and the insiders-outsiders (IO) conflict in banking. Consistent with the extant literature on nonfinancial firms, I find that dividends are positively related to the severity of the IO conflict. The results on the IO conflict, unlike those for the relationship between risk and dividends, are robust to all specifications. The findings of this paper have important policy implications. The current debate on whether dividends should be curbed in banks that are not financially sound or with liquidity problems can draw further insights from my analysis. While for the US there is already a system of Prompt Corrective Actions (PCA), European countries do not currently have a system of legally binding PCA. It has been suggested that dividends be included in the variables that should be targeted by PCA (Brunnermeier et al., 2009), and an investigation of the relationship between dividends and bank capital during the 2007-2009 financial crisis backs this view (Acharya et al., 2009). In light of my findings on the relation between dividends and default risk, and considering the importance of retained earnings for financial stability, it would be useful for regulators to set formal requirements for banks that want to distribute dividends. Lack of compliance with such requirements should promptly be followed by restrictions in dividends payments, to be coordinated with complementary measures including, for instance, issuance of new common equity capital.

16

References Acharya V. V., Gujral I., Shin H. S. (2009). Dividends and bank capital in the financial crisis of 2007-2009. Unpublished paper. Ahmed A. S., Takeda C., Thomas S. (1999). Bank loan loss provisions: a reexamination of capital management, earnings management and signaling effects. Journal of Accounting and Economics, 28, 1-25. Allen L., Gottesman A., Saunders A., Tang Y. (2009), The role of banks in dividend policy, New York University, Stern School of Business Working Paper. Bar-Yosef S., Huffman L. (1986). The informational content of dividends: A signalling approach. Journal of Financial and Quantitative Analysis, 21, 47-58. Bessler W., Nohel T. (1996). The stock-market reaction to dividend cuts and omissions by commercial banks. Journal of Banking and Finance, 20 (9), 1485-1508. Bessler W., Nohel T. (2000). Asymmetric information, dividend reductions and contagion effects in bank stock returns. Journal of Banking and Finance, 24 (9), 1831-1848. Boldin R., Leggett K. (1995). Bank dividend policy as a signal of bank quality. Financial Services Review, 4 (1), 1-8. Boyd J. H., Graham S. L. (1988). The profitability and risk effects of allowing bank holding companies to merge with other financial firms: a simulation study. Federal Reserve Bank of Minneapolis Quarterly Review, 12, 320. Brunnermeier M., Crocket A., Goodhart C., Persaud A. D., Shin H. S. (2009). The fundamental principles of financial regulation. Geneva Reports on the World Economy 11. Casey K.M., Dickens R.N. (2000). The effect of tax and regulatory changes on commercial bank dividend policy. The Quarterly Review of Economics and Finance, 40, 279293. Casey K.M., Dickens R.N., Newman J.A. (2002). Bank dividend policy: explanatory factors. Quarterly Journal of Business and Economics, December. Collins J., Shackelford D, Wahlen J. (1995). Bank differences in the coordination of regulatory capital, earnings and taxes. Journal of Accounting Research, 33 (2), 263-292. Cornett M. M., Fayman A., Marcus A. J., Tehranian H. (2008). Dividend signalling: evidence from bank IPOs. Unpublished paper. DeAngelo H., DeAngelo L., Skinner D. J. (2004). Are dividends disappearing? Dividend concentration and the consolidation of earnings. Journal of Financial Economics, 72, 425-456. DeAngelo H., DeAngelo L., Stulz R. M. (2006). Dividend policy and the earned/contributed capital mix: a test of the life-cycle theory. Journal of Financial Economics, 81, 227-254. Denis D. J., Osobov I. (2008). Why do firms pay dividends? International evidence on the determinants of dividend policy. Journal of Financial Economics, 89, 62-82. Easterbrook F. (1984). Two agency cost explanations of dividends. American Economic Review, 72, 65058. Fama E. F., French K. R. (1992). The cross-section of expected stock returns. Journal of Finance, 47, 427-465. Fama E. F., French K. R. (2001). Disappearing dividends: changing firms characteristics or lower propensity to pay? Journal of Financial Economics, 60, 3-43. Faccio M., Lang L. H. P., Young L. (2001). Dividends and expropriation. The American Economic Review, 91 (1), 54-78.

17

Filbeck G., Mullineaux D.J. (1999). Agency costs and dividend payments the case of bank holding companies. The Quarterly Review of Economics and Finance, 39, 409-418. Financial Stability Board. Press Release, 15 September 2009. Greene W. H. (1990). Econometric Analysis. Macmillan, New York. Gropp R., Heider F. (2009). The determinants of bank capital structure. European Central Bank Working Paper Series, No 1096. Grossman S. J., Hart O. D. (1980). Takeover bids, the free-rider problem, and the theory of the corporation. Bell Journal of Economics, 11, 42-64. Gugler K., Yurtoglu B. B. (2003). Corporate governance and dividend pay-out policy in Germany. European Economic Review, 47, 731-758. Hoberg G, Prabhala N. R. (2009). Disappearing dividends, catering, and risk. Review of Financial Studies, 22, 79-116. Honore B. E. (1992).Trimmed lad and least square estimation of truncated and censored regression models with fixed effects. Econometrica, 60, 533-565. Iannotta G. (2006). Testing for opaqueness in the European banking industry: evidence from bond credit ratings. Journal of Financial Services Research, 30, 287-309. Iannotta G., Nocera G., Sironi A. (2007). Ownership structure, risk and performance in the European banking industry. Journal of Banking and Finance, 31, 2127-2149. Jensen M. C. (1986). Agency costs of free cash flow, corporate finance and takeovers. American Economic Review, 76, 323-339. Laeven L. (2002). Bank risk and deposit insurance. The World Bank Economic Review, 16 (1), 109-137. 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. Li K., Zhao X. (2008). Asymmetric information and dividend policy. Financial Management, Winter, 673-694. Lintner J. (1956). Distribution of incomes of corporations among dividends, retained earning and taxes. The American Economic Review, 46 (2), 97-113. Merton R. C. (1974). On the pricing of corporate debt: the risk structure of interest rates. Journal of Finance, 29, 449-470. Merton R. C. (1977). An analytical derivation of the cost of deposit insurance and loan guarantees. Journal of Banking and Finance, 1(2), 3-11. Noronha G. M., Shome D.K., Morgan G.E. (1996). The monitoring rationale for dividends and the interaction of capital structure and dividend decisions. Journal of Banking and Finance, 20, 439-454. Rozeff M.S. (1982). Growth, Beta and Agency Costs as Determinants of Dividend Payout Ratios. Journal of Financial Research, 5, 249-259. Saunders A., Strock E., Travlos N. G. (1990). Ownership structure, deregulation, and bank risk taking. Journal of Finance, 45, 643-654. Shleifer A., Vishny R. W. (1986). Large shareholders and corporate control. The Journal of Political Economy, 94 (3), 461-488. Treasury Committee (2009). Banking Crisis: reforming corporate governance and pay in the City. Ninth Report of Session 2008-2009.

18

Von Eije H., Megginson W. (2006). Dividend policy in the European Union. Unpublished paper.

19

Table 1

* Variables definition for equation (2): Yit = 'xit + 'cit + it

Credit risk

ratio loan loss provisions to total loans of bank i in year t

Z-score

The Z-score (Boyd and Graham, 1988), or the ratio ROA of bank i plus the equity-total assets ratio bank i in year t divided by the standard deviation of the ROA of bank i number of recorded shareholders in 2007+

Recorded shareholders Listed bank

1 if bank i is listed on the stock market in year t , 0 otherwise

Independent bank Loans growth Size

1 if there is no shareholder with more than 25% or direct ownership in 2007+ and 0 otherwise

average annual % rate of growth in loans of bank i between years t-1 and t

log assets of bank i in year t

ROA

return on assets (net income/ average total assets) of bank i in year t

France

1 if bank i is domiciled in France, 0 otherwise

Germany

1 if bank i is domiciled in Germany, 0 otherwise

Italy

1 if bank i is domiciled in Italy, 0 otherwise

Unfortunately, Bankscope provides data for these variables only as of the last accounting year available. However, because these data tend to be sticky, we do not believe this has affected our results considerably.

20

Table 2 Descriptive statistics for dividends/earnings ratio, dividends/equity ratio, default risk, and credit risk (2000-07) France Germany Italy UK ALL Banks 94 61 50 130 355 Sample % 28% 18% 15% 39% 100% Listed Banks 10 15 13 13 51 Positive DP 70 51 22 126 269 Negative DP 0 0 0 5 5 Positive DPE 70 51 22 131 274 Zero* DP/DPE 192 170 103 278 743 Average DP** 19.05% 13.66% 8.41% 19.12% 16.59% Average DPE 3.49% 1.12% 1.31% 2.86% 2.45% Positive DP Mean 71.32% 59.20% 47.77% 61.29% 62.40% SD 60.05% 60.49% 21.41% 65.85% 60.91% p1 1.05% 5.79% 11.50% 5.30% 4.70% p5 7.95% 14.20% 17.96% 7.69% 7.95% p50 63.54% 45.30% 46.16% 47.52% 48.65% p95 151.65% 135.30% 83.74% 132.39% 135.27% p99 415.54% 404.84% 86.67% 423.40% 415.54% Positive DPE Mean 13.05% 4.87% 7.43% 8.93% 9.10% SD 18.78% 3.09% 5.38% 7.93% 11.43% p1 0.09% 1.30% 1.31% 0.36% 0.33% p5 0.99% 1.47% 2.87% 0.48% 0.94% p50 8.77% 4.08% 6.00% 7.35% 6.81% p95 33.36% 11.79% 16.71% 25.71% 26.00% p99 120.87% 15.08% 26.03% 40.52% 50.68% Z-score Mean 41.28 31.24 55.09 63.06 50.95 SD 52.77 58.94 68.87 71.42 65.98 p50 25.63 13.99 30.26 42.37 30.21 Loan loss provision ratio Mean 0.34% 6.03% 0.69% 0.57% 1.68% SD 4.30% 106% 3.90% 3.54% 48.97% p50 0.22% 0.40% 0.41% 0.06% 0.23% Notes: DP = dividends/net income. DPE = dividends/equity. The five observations for which the dividend payout ratio is negative are for: Ansbacher Overseas Group Ltd (UK, Bank Holding Company, BHC), year 2003; Butterfield Holding Ltd (UK, BHC), year 2004; EFG Private Bank Ltd (UK, Commercial Bank), year 2005; London Scottish Bank Plc (UK, Commercial Bank), year 2007; Northern Rock Plc (UK, Commercial Bank), year 2007. I report the statistics only for observations for which the explanatory variables reported in Table 1 are available. Mean is the average positive DP or DPE. SD is the standard deviation. p1, p5, p50, p95, and p99 refer to the 1 st, 5th, 50th, 95th and 99th percentile of the distribution. * Includes number of missing values for the dividend payout when earnings are reported. ** Calculated excluding negative DP for the UK.

21

Table 3 Pair-wise correlations MODEL 1 DP DPE Loans Growth Credit risk ROA Size (log assets) Z-score 0.6469*** -0.0123 0.0354 0.1798*** 0.2057*** -0.0782*** -0.0056 0.0256 0.4990*** 0.1090*** -0.0518* 0.0084 -0.0218 0.0491 -0.0292 -0.3544*** 0.0664** 0.0356 -0.1459*** -0.0292 -0.0407 DPE Loans Growth Credit risk ROA Size

MODEL 2 DP DPE Loans Growth Credit risk ROA Size (log assets) Z-score 0.6477*** -0.0121 0.2393*** 0.0551** 0.0388* -0.0677*** -0.0066 0.0120 0.1330*** 0.0701*** -0.0674*** -0.0014 -0.0128 0.0387 -0.025 -0.0569** 0.0417 -0.0188 -0.0797*** -0.0096 0.0124 DPE Loans Growth Credit risk ROA Size

Notes: DP = dividends/net income. DPE = dividends/equity. I have excluded observations for which DP is negative or for which the earnings are zero. Model 2 considers observations for which DP and credit risk (the loan loss provision ratio) are lower or equal to their respective 95th percentile. * Denotes significance at the 10% level. ** Denotes significance at the 5% level. *** Denotes significance at the 1% level.

22

Table 4 Tobit regression model with Random Effects: Estimation results Dependent variable: Dividends/earnings, DP MODEL 1 Credit risk Z-score Listed bank Recorded Shareholders Independent bank Loans Growth ROA Size (log assets) France Germany Italy Banks Obs Chi
2

MODEL 2 1.7047 -0.0018** 0.2091**

MODEL 3 1.7008 -0.0018**

MODEL 4 1.8849* -0.0016**

1.7961* -0.0019**

0.0057*** 0.2000** -0.0020 10.6686*** 0.0938*** -0.1140 -0.2577** -0.2581** 325 910 53.9983*** MODEL 1 -0.0020 10.1251*** 0.0826*** -0.1061 -0.2774** -0.2998** 325 910 58.9449*** MODEL 2 0.2224* -0.0002** 0.0311** 0.0008*** 0.0233** -0.0006 1.8946*** 0.0129*** -0.0123 -0.0364** -0.0491*** 322 929 -0.0006 1.7292*** 0.0111*** -0.0117 -0.039** -0.0555*** 322 929 -0.0006 1.7301*** 0.0093*** -0.0046 -0.0303* -0.0491*** 322 929 -0.0006 1.8468*** 0.0133*** -0.0143 -0.0343** -0.047*** 322 929 -0.0021 9.8662*** 0.0685*** -0.0565 -0.2128* -0.2544** 325 910 63.7730*** MODEL 3 0.2291* -0.0002** -0.0021 10.7233*** 0.0975*** -0.1309 -0.2397** -0.2458** 325 910 59.0197*** MODEL 4 0.2513** -0.0002*

Dependent variable: Dividends/equity capital, DPE Credit risk Z-score Listed bank Recorded Shareholders Independent bank Loans Growth ROA Size (log assets) France Germany Italy Banks Obs
2

0.2481* -0.0002**

Chi 53.8119*** 60.2411*** 64.3374*** 57.7652*** Notes: For the regressions on DP, I have excluded observations for which the DP is negative or for which the earnings are zero, and observations for which the DP and the loan loss provision ratio are larger than their respective 95th percentile. For the regressions on DPE, I have excluded observations for which DPE and the loan loss provision ratio are larger than their respective 95 th percentile. * Denotes significance at the 10% level. ** Denotes significance at the 5% level. *** Denotes significance at the 1% level.

23

Table 5 Tobit regression model with Random Effects allowing for earnings smoothing: Estimation results Dependent variable: Dividends/earnings, DP MODEL 1 Credit risk Z-score Listed bank Recorded Shareholders Independent bank Loans Growth ROA Size (log assets) France Germany Italy Banks Obs Chi
2

MODEL 2 0.3271 -0.0018** 0.1978**

MODEL 3 0.344 -0.0018**

MODEL 4 0.3926 -0.0016**

0.3593 -0.0019**

0.0059*** 0.2243*** -0.0019 7.1688*** 0.0940*** -0.1033 -0.3190** -0.2553** 325 913 46.4456*** MODEL 1 -0.0019 6.4925*** 0.0833*** -0.0953 -0.3384*** -0.2938** 325 913 50.7856*** MODEL 2 0.0355 -0.0002** 0.0320*** -0.002 6.4806*** 0.0681*** -0.044 -0.2711** -0.2512** 325 913 56.3302*** MODEL 3 0.0398 -0.0002* 0.0009*** 0.0285** -0.0006 0.947*** 0.0123*** -0.0125 -0.0445*** -0.0474*** 322 932 -0.0007 0.792*** 0.0106*** -0.0118 -0.0468*** -0.0539*** 322 932 -0.0007 0.8332*** 0.0086*** -0.0042 -0.0377** -0.0474*** 322 932 -0.0006 0.9636*** 0.0128*** -0.0149 -0.0413** -0.045*** 322 932 -0.002 7.4644*** 0.0980*** -0.1224 -0.2947** -0.2422* 325 913 52.3501*** MODEL 4 0.0463 -0.0002*

Dependent variable: Dividends/equity capital, DPE Credit risk Z-score Listed bank Recorded Shareholders Independent bank Loans Growth ROA Size (log assets) France Germany Italy Banks Obs
2

0.0425 -0.0002**

Chi 43.9401*** 50.7531*** 55.2221*** 49.5816*** Notes: For the regressions on DP, I have excluded observations for which the DP is negative or for which the earnings are zero, and observations for which the DP and the loan loss provision ratio are larger than their respective 95th percentile. For the regressions on DPE, I have excluded observations for which DPE and the loan loss provision ratio are larger than their respective 95 th percentile. The total number of observations (913 for the regressions on DP, and 932 for the regressions on DPE) is larger than that of table 4 (910, 929) because of the different distribution of the loan loss provision ratio allowing for earnings smoothing with respect to the same ratio when earnings smoothing is not allowed for. The difference in the distribution causes 3 additional observations to be allowed in the estimation of equation (2). * Denotes significance at the 10% level. ** Denotes significance at the 5% level. *** Denotes significance at the 1% level.

24

Table 6 Tobit regression model with Random Effects allowing for the influence of the ratio retained earnings/equity: Estimation results Dependent variable: Dividends/earnings, DP MODEL 1 Credit risk Z-score Listed bank Recorded Shareholders Independent bank Retained earnings/equity Loans Growth ROA Size (log assets) France Germany Italy Banks Obs Chi
2

MODEL 2 1.2309 -0.0017** 0.1728*

MODEL 3 1.255 -0.0018**

MODEL 4 1.6421 -0.0017**

1.3991 -0.0019**

0.0038** 0.2014** -0.4178** -0.0004 7.9876*** 0.1151*** -0.2344* -0.2093 -0.3696*** 211 515 -0.4477*** -0.0004 7.5395*** 0.1056*** -0.2298* -0.2549* -0.4084*** 211 515 -0.4241*** -0.0004 7.1298*** 0.092*** -0.1723 -0.1749 -0.3631*** 211 515 62.4581*** MODEL 3 0.2583 -0.0002** -0.3487** -0.0004 8.6012*** 0.1223*** -0.2707** -0.1948 -0.3362*** 211 515 59.5593*** MODEL 4 0.3164* -0.0002**

57.2161*** 60.6012*** Dependent variable: Dividends/equity capital, DPE MODEL 1 MODEL 2 0.2429 -0.0002** 0.0268** 0.3022 -0.0003**

Credit risk Z-score Listed bank Recorded Shareholders Independent bank Retained earnings/equity Loans Growth ROA Size (log assets) France Germany Italy Banks Obs
2

0.0005** 0.0265** -0.0594*** -0.0001 1.6703*** 0.0167*** -0.0347** -0.0333* -0.0627*** 207 512 -0.0641*** -0.0001 1.4122** 0.0151*** -0.0349** -0.0395** -0.0681*** 207 512 -0.0608*** -0.0001 1.4191** 0.0134*** -0.0264 -0.0284 -0.0614*** 207 512 -0.051** -0.0001 1.65*** 0.0176*** -0.039** -0.0314* -0.0578*** 207 512

Chi 55.3262*** 59.9646*** 60.3222*** 57.3947*** Notes: For the regressions on DP, I have excluded observations for which the DP is negative or for which the earnings are zero, and observations for which the DP and the loan loss provision ratio are larger than their respective 95th percentile. For the regressions on DPE, I have excluded observations for which DPE and the loan loss provision ratio are larger than their respective 95th percentile. * Denotes significance at the 10% level. ** Denotes significance at the 5% level. *** Denotes significance at the 1% level.

25

26