Local Culture and Dividends

Local Culture and Dividends Erdem Ucar I empirically investigate whether geographical variations in local culture, as proxied by local religion, affect dividend demand and corporate dividend policy for a large sample of US firms. Firms located in Protestant counties are more likely to be dividend payers, initiate dividends, and have higher dividend yields, while firms located in Catholic counties are less likely to be dividend payers and have lower dividend yields. There is a geographically varying dividend clientele effect consistent with the variations in risk aversion among different cultural groups. My results suggest that firms largely held by local investors determine their corporate policies in line with local culture. I examine the impact of culture, defined as the customary beliefs, ideas, values, social forms, and customs of a social or religious group, on dividend demand and payout policy. 1 Culture plays a major role in individuals lives. Differences in cultural attributes can lead to variations in attitudes or actions. Culture has an influence on decision making via its impact on attitudes, beliefs, and values (Guiso, Sapienza, and Zingales, 2006; Grullon, Kanatas, and Weston, 2010). More importantly, recent studies indicate that cultural groups have different attitudes with respect to financial and economic decisions (La Porta et al., 1998, 1999; Stulz and Williamson, 2003). Religion, as a common measure of culture, is particularly useful in shedding additional light on some phenomena related to investor behavior and corporate decisions (Stulz and Williamson, 2003; Kumar, Page, and Spalt, 2011; Pantzalis and Ucar, 2014). Payout policies or dividend demand are closely related to individuals attitudes regarding financial outcomes. The dividend clientele literature (Miller and Modigliani, 1961) argues that investors characteristics and attitudes toward dividend income vary across investor groups, leading to differences in dividend demand. I posit that culturally induced variations in attitudes are important in explaining issues related to payout policies. Following the prior literature, I use religion as a proxy for culture and empirically investigate whether geographical variations in local religious affiliation affect dividend policy in a large sample of US firms. There is a significant variation in local religious affiliations across the United States, and my results suggest that this variation is an important determinant of dividend demand, as well as payout policies. Focusing on the two major religious affiliations in the United States, I find that there are significant differences between the dividend policies of firms located in predominantly Protestant areas and those located in predominantly Catholic areas. Firms headquartered in counties with large Protestant populations are more likely to be dividend payers, whereas firms from counties with large Catholic populations are less likely to be dividend payers. Similarly, dividend yields are higher (lower) for firms located in predominantly Protestant I would like to thank an anonymous referee, Raghavendra Rau (Editor), Christos Pantzalis, M. Sinan Goktan, Ayca Altintig, Aslihan G. Korkmaz, and seminar participants at the Financial Management Association 2013 Annual Meeting and the Southern Finance Association 2013 Annual Meeting for their helpful comments. Any remaining errors are mine. Erdem Ucar is an Assistant Professor of Finance in the Barowsky School of Business at the Dominican, University of California in San Rafael, CA. 1 See the following dictionary websites or Guiso, Sapienza, and Zingales (2006): http://www.merriam-webster. com/dictionary/culture, http://oxforddictionaries.com/definition/english/culture?q=culture. Financial Management Spring 2016 pages 105 140

106 Financial Management Spring 2016 (Catholic) counties. Furthermore, nondividend payer firms headquartered in Protestant areas are more likely to initiate dividends. My results can be interpreted based on risk aversion and dividend clientele arguments. Prior literature suggests that some investors prefer dividends because they perceive dividends as safe available income for current consumption compared to uncertain future income from equity investments. Firms cater to these demands through dividend payments. The clientele effect argument suggests that investors have different preferences and characteristics such that investors select firms whose payout policies are consistent with their preferences. Prior literature also highlights differences in investor attitudes and preferences by suggesting that different cultures have different attitudes towards finance (Stulz and Williamson, 2003) and indicates a significant variation in risk aversion across different religious affiliations. Protestant culture is more riskaverse than Catholic culture in terms of financial and economic outcomes (Barsky et al., 1997; Hilary and Hui, 2009; Benjamin, Choi, and Fisher, 2010; Kumar et al., 2011; Shu, Sulaeman, and Yeung, 2012). Kumar et al. (2011) demonstrate this point through individual investors risktaking behavior, as well as through corporate decisions and stock returns. In addition, Shu et al. (2012) support this point by analyzing mutual fund risk-taking behavior. My results suggest that this difference in risk aversion plays the largest role in the geographically varying local dividend clientele effect demonstrated in this paper. After controlling for other firm characteristics, I find that geographical variation in local culture, as proxied by local religious affiliation, has an impact on corporate dividend policies. The findings indicate a statistically and economically significant positive (negative) relation between dividend policies and firms headquartered in counties with a large Protestant (Catholic) population. For example, holding other variables constant, a firm headquartered in an area with an approximately 13% (14%) Protestant (Catholic) population has a significantly higher (lower) likelihood of being a dividend payer than a similar firm from an area with no Protestant (Catholic) population. There are similar findings for dividend yield and initiation. My empirical findings hold after a series of robustness checks. For example, my results remain strong after controlling for other local economic, demographic, and cultural factors. In addition, using a sample of firms with location information confirmed by the Compact Disclosure data provides even stronger results. My findings are also robust after accounting for the impact of a greater preference for lottery-type stocks by Catholics as suggested by Kumar et al. (2011) or after controlling for an alternative set of control variables. Additional tests suggest that my results emerge through the local culture channel. The results are stronger for firms largely held by local investors when I examine different local ownership measures. Furthermore, a matched sample test indicates that local religious affiliations lead to a variation in dividend policies. An analysis of corporate relocations also provides additional evidence and suggests a change in dividend policies in line with local cultures in new firm locations. Chief executive officers (CEOs) play an important role in corporate policies. Therefore, I investigate whether local CEO or local culture is the primary determinant of my findings and determine that local culture, as proxied by local religious affiliation, is the main driver of the geographically varying dividend clientele effect in my tests. Dividend policy has attracted a considerable amount of attention in the finance literature. An important strand of the prior literature focuses on cross-sectional variations in payout policies. The dividend clientele (or demand side) perspective, dating back to Miller and Modigliani (1961), investigates the determinants of a firm s propensity to pay dividends. Miller and Modigliani (1961) argue that imperfections, such as transaction costs or taxes, lead dividend clienteles to prefer dividends through variations in investor preferences. Recent studies find results consistent with dividend clienteles based on investor characteristics, such as investor age or income (Graham and Kumar, 2006; Becker, Ivkovic, and Weisbenner, 2011). My results suggest a correlation

Ucar Local Culture and Dividends 107 between geographical variation in dividend demand and investors local cultural or religious characteristics. I highlight a new type of geographically varying clientele effect induced by local culture consistent with variations in risk aversion among different cultural groups. My paper is also related to Baker and Wurgler (2004a, 2004b) who propose that firms cater to investors demands for dividends through dividend payouts when investors consider dividend payer firms more valuable. Moreover, earlier studies suggest that investor risk aversion can lead investors to prefer dividends over future capital gains (Lintner, 1962; Gordon, 1963). My paper suggests that local religion affects investor demand for dividends through its impact on investor risk attitudes and characteristics and firms cater to this dividend demand through their dividend payouts. The paper is also related to the recent literature regarding local bias. Ivkovic and Weisbenner (2005) find that individual investors have a higher propensity to invest in local firms due to an information advantage. Pirinsky and Wang (2006) determine a strong comovement in stock returns for same-location firms. Gao, Ng, and Wang (2011) examine the effect of location on capital structure. The locality of investors plays an important role in my results. My findings highlight the local component of dividends by suggesting a geographically varying dividend clientele effect induced by local culture. Moreover, I note the stronger impact of local culture on dividend policies for firms held primarily by local shareholders. This implies that firms with greater local shareholder bases determine their dividend policies consistent with the investor risk aversion induced by the local culture. Recent studies indicate that culture has an impact on economic and financial outcomes (Grinblatt and Keloharju, 2001; Stulz and Williamson, 2003; Guiso et al., 2006; Eun, Wang, and Xiao, 2014). Eun et al. (2014) suggest that culture is an important omitted variable in the literature. Hilary and Hui (2009) demonstrate the impact of local religious characteristics on corporate risk-taking behavior. Kumar et al. (2011) confirm the effect of local culture on investor behavior and corporate decision making. Pantzalis and Ucar (2014) demonstrate how investor inattention to firm news can be traced to local religious characteristics. My paper expands on this literature by presenting the effect of local culture, as proxied by local religious affiliation, on dividend demand and corporate dividend policies. My paper is closer in spirit to Becker et al. (2011) who examine local dividend clienteles. Becker et al. (2011) argue that firms from areas with a higher proportion of senior citizens have a greater likelihood of being dividend payers due to the dividend clientele based on age. The main distinction between my paper and their paper is that my paper focuses on the impact of local culture, as measured by local religion, on payout policies and suggests a new dividend clientele effect based on local culture. Moreover, my empirical findings are robust to the local senior population effect for a larger sample and suggest a stronger economic impact of local culture on geographically varying dividend policies. The remainder of the paper is organized as follows. The next section presents a short summary of the data and the sample selection method in addition to the summary statistics. Section II provides the empirical results. Section III presents the empirical results of additional tests and robustness checks, while Section IV provides my conclusions. I. Data, Sample Selection, and Summary Statistics A. Data and Sample Selection I follow the sample selection criteria used by recent studies (Grullon et al., 2011) in the related literature. My sample includes US firms with available dividend data and accounting information from Compustat and stock price information from CRSP in the period from 1990 to 2010. The sample excludes utilities and financials categories (SIC codes 4900-4999 or SIC

108 Financial Management Spring 2016 codes 6000-6999). The sample also requires sample firms to have CRSP issue codes of 10 or 11 and available headquarters location information. For some variables, the empirical tests use lagged and leading year information. Thus, the main regression sample loses some firm-year observations (i.e., 80,049 observations in the propensity to pay dividends regressions.) I use firm address information from Compustat in the main tests. I also use firm address information from Compact Disclosure for some robustness tests. I obtain local religious affiliation information from the Association of Religion Data Archives (ARDA) using the 1990, 2000, and 2010 countylevel data sets in order to construct the local religious affiliation variables. County-level economic and demographic variables are from the 1990 and 2000 Censuses, as well as the other available statistics on the US Census website for the years following 2000. I also use the interpolations of the ARDA or Census data in order to construct the local variables for years without available data. My empirical model builds on Becker et al. s (2011), while keeping the dividend policy and firm characteristics variable definitions consistent with the recent literature (Becker et al., 2011; Grullon et al., 2011). I define my dividend policy variables as follows. Dividend Payer is an indicator variable that takes a value of one if the total amount of dividends is greater than zero for a given year, and zero otherwise. Dividend Yield is the ratio of total dividends to the lagged market value. Dividend Initiation is an indicator variable that takes a value of one if a current nondividend payer firm becomes a dividend payer in the following year, and zero if a current nondividend payer firm stays as nondividend payer in the following year. I define the local religious affiliation variables similar to Kumar et al. (2011) by using the ARDA data sets. Prot (Cath) is the fraction of Protestants (Catholics) in a given county where a firm is located. I also use a variable Cpratio, which is the ratio of Catholics to Protestants (ratio of Cath to Prot). Other main county-level control variables for headquarter locations are as follows. Local Seniors is the proportion of individuals who are 65 years old or older in a given county. Income is the county median household income. Median House Value provides the median house price for a given county. Education is the proportion of the population holding college degrees in a given county. I also use the logarithm of county population in the tests that include county-level variables. I define the firm characteristic variables of the main empirical model as follows. Net Income is net income scaled by total assets for a given year. Cash is cash scaled by total assets. I define Q as the sum of the market value of equity and the book value of liabilities scaled by total assets for a given year. Debt is long-term debt scaled by total assets. Log of MV is the logarithm of a firm s market value for a given year. Log of Assets is the logarithm of total assets. Volatility is the standard deviation of monthly stock returns for the previous two-year period. Lagged Return is the monthly stock returns for the previous two-year period. In calculating Volatility and Lagged Return, I require stock return information for at least the previous 12 months to be nonmissing if a firm has stock return information available less than for 24 months. Asset Growth is the logarithm of the total assets growth rate calculated using both the current and the previous year s figures. My tests also include the following firm age group indicator variables consistent with Becker et al. (2011): Age 1-5, Age 6-10, Age 11-15, and Age 16-20. Age 21 and Over is the dropped category in my regressions. I measure firm age based on the time between the date that a firm is listed on the CRSP and the current year. My empirical tests also include state, industry, and year dummy variables. I use the Fama-French (1997) 48 industry classifications. B. Summary Statistics In Table I, I report the summary statistics for dividend policy variables and firm characteristics, as well as the main local economic and demographic variables. Panel A indicates that,

Ucar Local Culture and Dividends 109 Table I. Summary Statistics Dividend Payer is an indicator variable that takes a value of one if the total amount of dividends is greater than zero for a given year and zero otherwise. Dividend Yield is the ratio of total dividends to the lagged market value. Dividend Initiation is an indicator variable that takes a value of one if a current nondividend payer firm becomes a dividend payer during the following year, and zero if a current nondividend payer firm stays as a nondividend payer during the following year. Cath is the fraction of Catholics in the county where a firm is located. Prot is the fraction of Protestants in the county where a firm is located. Cpratio is the ratio of Catholics to Protestants in the county where a firm is located (ratio of Cath to Prot). Local variables in this table are as follows. Local Seniors is the proportion of people who are 65 years old or older in the firm s headquartered county. Income is the median household income in the firm s headquartered county. Education is the proportion of the population with college in the firm s headquartered county. NYE is measure of firm size based on the NYSE equity percentiles for the corresponding period. M/B is the ratio of the market value of assets to the book value of assets. Total Assets provides the total asset value in millions of dollars. ROA is the return on assets as measured by income before depreciation divided by total assets for a given year. Sales Growth is the growth rate of the sales calculated by using the current and previous year figures. Firm Age is based on the number of years between the date a firm initially listed on CRSP and the current year. Mean 25th Percentile Median 75th Percentile Standard Deviation Panel A. Payout Policy Variables (in %) Dividend Payer 24.84% 0.00% 0.00% 0.00% 43.21% Dividend Yield 0.57% 0.00% 0.00% 0.25% 1.23% Dividend Initiation 2.14% 0.00% 0.00% 0.00% 14.47% Panel B. County-Level Variables Cath (%) 26.13% 15.98% 24.31% 36.85% 14.00% Prot (%) 20.11% 10.41% 15.62% 28.40% 12.80% Local Seniors (%) 11.63% 9.73% 11.43% 13.07% 2.97% Income ($) 48,557 38,133 46,030 56,113 13,986 Education (%) 31.95% 25.25% 30.40% 38.68% 9.85% Panel C. Firm Characteristics NYE 24.19 3.00 12.00 38.00 27.02 M/B 2.19 1.11 1.52 2.41 1.97 ROA 0.03 0.00 0.10 0.17 0.27 Sales Growth 21.28% 3.25% 8.63% 25.54% 67.42% Total Assets ($ mil) 1,726.58 32.22 126.38 583.54 12,491.93 Firm Age 13.26 3.47 8.58 18.19 14.33 on average, 25% of the sample firms pay dividends in a given year. Average dividend yield is 0.6% for all of the sample firms. On average, 2.1% of the firms that do not pay dividends in the current year pay dividends in the following year. These dividend policy variables have values consistent with previous studies. Some of the summary statistics for local county characteristics are as follows. The average proportion of Catholics (Protestants) is 26.3% (20.1%) for a firm s headquartered county in a given year. The average local senior citizen population is approximately 11.6% and the average fraction of the local population with a college degree is approximately 32%. County-level variables display summary statistics similar to that of previous studies. Panel C reports the summary statistics for some important firm characteristics. For example, an average firm in the sample has an equity value that is equal to the 24 th percentile

110 Financial Management Spring 2016 of the NYSE equity size distribution for a given year. Thus, an average firm in my study is not a very big firm. The average market-to-book ratio is approximately 2.2, the average return on assets (ROA) is approximately 0.03, and the average sales growth is approximately 22.3% for the sample. The characteristics of the sample firms display a pattern consistent with the prior literature. II. Empirical Results A. Main Tests My main empirical model is similar to the one used by Becker et al. (2011). I control for Net Income, Cash, Q, Debt, Volatility, Lagged Return, Log of MV, Log of Assets, and Asset Growth, and indicator variables for firm age groups, as well as state, industry, and year dummy variables. I also adjust standard errors for heteroskedasticity and cluster them at the firm level in all empirical tests. I use my payout policy variables, Dividend Payer, Dividend Yield, and Dividend Initiation, as the dependent variables in the regression models. I use a Logit regression model in order to analyze the impact of local culture on a firm s propensity to pay dividends and to initiate dividends. I use an ordinary least square (OLS) regression model in the dividend yield tests. I use Cath, Prot, and Cpratio to measure local religious affiliation in my tests. Columns 1-3 of Table II indicate that local religious affiliation has statistically significant coefficients. The coefficient of Prot (Cath) is 2.259( 1.220), while the coefficient of Cpratio is -0.089. These results suggest that there is a positive (negative) relation between the likelihood of being a dividend payer and firm locations with higher fractions of Protestants (Catholics). In Logit regression models, it can be difficult and misleading to interpret the economic importance of variables based on coefficient magnitudes. A better way to interpret the economic importance of the results in a Logit regression model is to focus on the change in odds for the dependent variable when there is a one standard deviation change in an independent variable. Consistent with this method, the first three columns suggest economically significant effects. A one standard deviation greater value in the proportion of Protestants in a firm s county is associated with a 34% greater likelihood in that a firm will pay dividends compared to another firm in a county with a lower fraction of Protestants. Alternatively, the same increase in the proportion of Catholics (Cpratio) in a county decreases the odds that a firm will pay dividends by approximately 15.7% (15.5%). The control variables have the expected coefficient values with a pattern consistent with the prior literature (Becker et al., 2011). The findings suggest that, holding all other variables constant, a firm located in a county with an approximately 13% (14%) Protestant (Catholic) population has a significantly greater (lower) likelihood of being a dividend payer firm than a similar firm in a county with no Protestant (Catholic) population. Columns 4-6 report a similar difference toward dividend policies across US counties by focusing on dividend yields. All of the three religious affiliation coefficients suggest statistically and economically significant effects. A one standard deviation higher value in the proportion of Protestants (Catholics) in a firm s county is associated with a 0.107 (0.053) standard deviation increase (decrease) in the dividend yield compared to another firm located in a different county with a lower proportion of Protestants (Catholics). The final three columns of Table II focus on the impact of the local religious composition on dividend initiation across locations within the United States. In this analysis, only Prot is statistically significant, while Cath and Cpratio are statistically insignificant. This finding suggests a stronger role for Protestants in dividend initiation when compared to Catholics, consistent with my earlier findings. Dividend initiation

Ucar Local Culture and Dividends 111 Table II. Dividend Payout and Religious Affiliation The dependent variable for Columns 1-3 is Dividend Payer. The dependent variable for Columns 4-6 is Dividend Yield, while the dependent variable for Columns 7-9 is Dividend Initiation. Columns 1-3 and Columns 7-9 use Logit regressions. Columns 4-6 use OLS regressions. Dividend Payer is an indicator variable that takes a value of one if the total amount of dividends is greater than zero for a given year and zero otherwise. Dividend Yield is the ratio of total dividends to the lagged market value. Dividend Initiation is an indicator variable that takes a value of one if a current nondividend payer firm becomes a dividend payer during the following year, and zero if a current nondividend payer firm stays as a nondividend payer during the following year. Cath (Prot) is the fraction of Catholics (Protestants) in the county where a firm is located. Cpratio is the ratio of Catholics to Protestants in the county where a firm is located. Net Income is the net income scaled by total assets for a given year. Cash is the cash scaled by total assets for a given year. Q is defined as the sum of the market value of equity and the book value of liabilities scaled by total assets for a given year. Debt is long-term debt scaled by total assets for a given year. Volatility is the standard deviation of monthly stock returns for the previous two-year period. Lagged Return is the monthly stock returns for the previous two-year period. Log of Assets is the logarithm of total assets. Log of MV is the logarithm of a firm s market value for a given year. Asset Growth is the logarithm of the growth rate of total assets calculated using the current and previous year s figures. All of the tests include the following firm age group indicator variables: Age 1-5, Age 6-10, Age 11-15, andage 16-20. Age 21 and Over is the dropped category in the regressions. All of the tests include state, industry, and year dummy variables. Intercept, firm age indicators, state, industry, and year dummy variables are not displayed for brevity. Standard errors are adjusted for heteroskedasticity and clustered at firm level. Robust p-values are in parentheses. (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Cath 1.220 0.005 0.113 (0.005) (0.008) (0.822) Prot 2.259 0.010 1.547 (0.000) (0.000) (0.008) Cpratio 0.089 0.000 0.019 (0.021) (0.004) (0.647) Net Income 3.466 3.459 3.465 0.001 0.001 0.001 3.303 3.304 3.301 (0.000) (0.000) (0.000) (0.001) (0.001) (0.0013) (0.000) (0.000) (0.000) Cash 0.908 0.890 0.897 0.000 0.000 0.000 0.669 0.681 0.667 (0.000) (0.000) (0.000) (0.430) (0.341) (0.366) (0.001) (0.000) (0.001) Q 0.140 0.138 0.141 0.000 0.000 0.000 0.137 0.137 0.137 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.003) (0.030) (0.003) Debt 0.979 0.999 0.984 0.004 0.004 0.004 0.723 0.734 0.723 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) (0.001) (Continued)

112 Financial Management Spring 2016 Table II. Dividend Payout and Religious Affiliation (Continued) (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Volatility 14.354 14.249 14.340 0.019 0.018 0.019 3.494 3.460 3.495 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Lagged Return 0.017 0.018 0.017 0.000 0.000 0.000 0.142 0.142 0.142 (0.421) (0.376) (0.400) (0.378) (0.324) (0.355) (0.000) (0.000) (0.000) Log of MV 0.380 0.380 0.383 0.001 0.001 0.001 0.174 0.174 0.174 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.010) (0.009) (0.009) Log of Assets 0.036 0.042 0.034 0.000 0.000 0.001 0.002 0.004 0.002 (0.498) (0.433) (0.523) (0.431) (0.309) (0.431) (0.982) (0.951) (0.976) Asset Growth 0.635 0.633 0.634 0.001 0.001 0.001 0.443 0.442 0.443 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of Observations 80,049 80,049 80,049 79,964 79,964 79,964 52,363 52,363 52,363 R square 0.433 0.434 0.433 0.270 0.271 0.70 0.110 0.110 0.110 Significant at the 0.01 level. Significant at the 0.05 level. Significant at the 0.10 level.

Ucar Local Culture and Dividends 113 tests indicate that a one standard deviation higher value in the proportion of Protestants in a firm location produces a 20.8% greater likelihood of dividend initiation compared to another firm in a different county with a lower proportion of Protestants. Overall, the dividend yield and initiation tests of Table II support the results displayed in the dividend payer tests. The results in Table II are consistent with my conjecture that Protestants (Catholics) are associated with a higher (lower) demand for dividends. Protestants are associated with a greater degree of risk aversion and, as such, investors in Protestant areas prefer safe dividend income compared to uncertain future income from their equity investments. Alternatively, Catholics are associated with a lower degree of risk aversion (such as greater risk-taking behavior indicated by Kumar et al., 2011.) Thus, Catholic areas prefer fewer dividends. Table II also suggests that firms determine their dividend policies consistent with local investors preferences induced by the local culture. B. Empirical Tests with County-Level Factors In this section, I investigate whether other county-level demographic and economic variables have an influence on my results. In the recent literature, Becker et al. (2011) examine age-based dividend clientele and suggest a positive relation between the proportion of local seniors and dividend demand, along with dividend policies. Therefore, I repeat the main tests and include a variable that indicates the proportion of local senior citizens in Panel A of Table III. The results are very similar to the results reported in Table II. In addition, the local religion coefficients have a stronger effect on dividends than the effect induced by local seniors. For example, Column 2 indicates that a one standard deviation difference in the proportion of Protestants between two firms located in two different counties is associated with a 31% difference in the odds that a firm will pay dividends, while the same difference in the proportion of local seniors is only associated with a 8% difference in the odds that a firm will pay dividends. Similarly, Columns 1 and 3 present a greater impact of Cath and Cpratio on dividends compared to the impact of Local Seniors on dividends. The dividend yield and initiation tests in Table III also present similar evidence. For example, the economic significance of Prot is almost four times that of Local Seniors in my dividend initiation tests. Overall, local culture, as proxied by religious composition, presents a stronger local dividend clientele effect than the age-based local clientele effect demonstrated by Becker et al. (2011). Next, I also include county-level control variables in the empirical tests. Specifically, I include some local demographic and economic factors consistent with the prior literature (median household income, median house value, local education, and the logarithm of the county population, along with the proportion of local senior citizens) in Panel B of Table III. My results still hold and display a pattern similar to the one in Panel A. Although there is some decrease in the magnitude of the religious affiliation coefficients, their economic significance remains robust, while there is a larger decrease in the impact of Local Seniors. For example, Prot has a stronger coefficient in Panel B in the dividend initiation tests and the economic significance of Prot is almost nine times that of Local Seniors. Overall, my findings are robust to other local factors that can affect local dividend clienteles and dividend policies. In unreported tests, I re-examine the main tests after controlling for county-fixed effects. These tests also provide findings similar to the ones presented in Table II. This section suggests that other local characteristics, such as county-level economic and demographic characteristics, do not eliminate the role that local religious affiliation plays in dividend demand and corporate dividend policies.

114 Financial Management Spring 2016 Table III. Dividend Payout, Religious Affiliation, Local Seniors, and Other Local Factors The dependent variable for Columns 1-3 is Dividend Payer. The dependent variable for Columns 4-6 is Dividend Yield, and the dependent variable for Columns 7-9 is Dividend Initiation. Columns 1-3 and Columns 7-9 use Logit regressions, while Columns 4-6 use OLS regressions. County-level local control variables are as follows: Local Seniors, Income, Median House Value, Education,andLog of Population. Local Seniors is the proportion of people who are 65 years old or older in a firm s headquartered county. Income is the median household income in the county in which a firm is located. Median House Value is the median house value in the county in which a firm is located. Education is the proportion of the population with college degrees in the county in which a firm is located. Log of Population is the logarithm of the county population. All of the other variable definitions are provided in Table II. All of the tests include the following firm age group indicator variables: Age 1-5, Age 6-10, Age 11-15,andAge 16-20. Age 21 and Over is the dropped category in the regressions. All of the tests include state, industry, and year dummy variables. In this table, only the local religious affiliation variables are displayed for brevity. Standard errors are adjusted for heteroskedasticity and clustered at firm level. Robust p-values are in parentheses. (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Panel A. Cath 1.364 0.005 0.201 (0.002) (0.004) (0.681) Prot 2.084 0.001 1.446 (0.000) (0.000) (0.013) Cpratio 0.097 0.000 0.014 (0.013) (0.002) (0.734) Local Seniors Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of Observations 80,049 80,049 80,049 79,964 79,964 79,964 52,363 52,363 52,363 R square 0.434 0.434 0.433 0.270 0.271 0.270 0.110 0.110 0.110 Panel B. Cath 1.082 0.005 0.287 (0.024) (0.009) (0.608) Prot 1.652 0.009 1.645 (0.003) (0.000) (0.010) Cpratio 0.071 0.000 0.008 (0.088) (0.003) (0.867) Local Control Variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of Observations 80,049 80,049 80,049 79,964 79,964 79,964 52,363 52,363 52,363 R square 0.436 0.436 0.435 0.272 0.272 0.272 0.111 0.111 0.111 Significant at the 0.01 level. Significant at the 0.05 level. Significant at the 0.10 level. C. Role of Local Investors My findings demonstrate a geographically varying clientele effect induced by local culture and firms maintain dividend policies consistent with this effect. A more direct way to investigate this effect is to focus on the role of local investors. In this section, I examine the differences in the dividend effect between firms with greater local ownership and other firms by using different measures of local stock ownership. First, I focus on empirical tests that highlight the differences between firms located in metropolitan areas and firms from other areas. Previous studies suggest that there are differences between firms located in large metropolitan areas and those located in smaller cities. For example, firms from big cities have an advantage in accessing firm information compared to those from smaller cities or rural areas (Loughran and Schultz, 2005; Loughran,

Ucar Local Culture and Dividends 115 2008). Furthermore, the only-game-in-town effect demonstrated by Hong, Kubik, and Stein (2008) is expected to be weak in large cities with a higher number of firms and stronger in smaller cities with a lower number of firms. Hong et al. (2008) find that firms located in areas with a smaller number of local firms are associated with greater local ownership. One could expect a stronger local bias and, as such, a stronger impact of local culture on dividends for a subsample of firms excluding large metropolitan cities. I investigate whether the dividend effect is different between a subsample of firms located in the three largest metropolitan areas and another subsample that excludes these metropolitan areas in Panels A1 and A2 of Table IV. All of the empirical models in Table IV are the same as the ones in the main tests in Table II. The empirical findings in Panel A1 of Table IV are very similar to the main findings in Table II. The results become stronger for firms located in smaller areas or areas excluding the largest metropolitan regions. This finding is consistent with the higher local bias effect associated with smaller areas. Panel A2 provides the results for firms located in large metropolitan areas and these results are not statistically significant. Most of the results are also weaker in Panel A2. In unreported tests, I also examine the 10 largest metropolitan areas versus other areas. My findings still hold when the 10 largest metropolitan areas are used instead of the 3 largest metropolitan areas. Overall, this analysis provides support to the notion that local religious affiliations affect dividend policies, particularly for firms where local ownership plays an important role. Hong et al. (2008) find that firms located in areas with a lower number of firms experience relatively little competition for local investors dollars (the only-game-in-town effect) and firms located in areas with a smaller number of local firms hold a greater fraction of local ownership. Therefore, examining the number of local firms per capita provides a more direct measure of the local bias effect induced by the only-game-in-town effect. In Panels B1 and B2, I use the Census data and construct a variable by dividing the number of local firms in a given county by county population to identify the number of local firms per capita. Panel B1 of Table IV presents the findings for the sample firms in the lowest tercile of the number of firms per capita. In other words, Panel B1 reports the findings for the sample firms with a higher local bias induced by the only-game-in-town effect. Panel B1 indicates that firms located in areas with a lower number of firms per capita have a pattern similar to the main findings in Table II. There are also stronger coefficient values in Panel B1. Firms in the highest tercile of the number of local firms per capita are expected to have a lower local bias effect and Panel B2 reports the findings for these firms. As opposed to Panel B1, Panel B2 indicates that firms located in areas with a lower bias have a pattern different than the one in the main tests. The religious affiliation coefficients in Panel B2 are not statistically significant for the sample firms where local investors play a smaller role. The tests using the only-game-in-town effect, as measured by the number of local firms per capita, supports the results of the analysis in Panels A1 and A2. The dividend effect demonstrated in this paper is stronger for those firms located in areas associated with greater local ownership. This result also suggests that firms located in areas where local investors hold a significant fraction of ownership in local firms consider local investors culturally induced dividend preferences in shaping their dividend policies. Another way of examining the importance of local investor bases for the impact of local culture on corporate dividend policies is to focus on local stock market participation. Prior literature suggests that there are differences in stock market participation rates among demographic groups. Hong, Kubik, and Stein (2004) find that stock market participation rates are much higher for white households compared to other racial groups. Hong et al. (2004) find that stock market participation rates for white households are almost three times greater than the rates for other racial groups. Hong et al. (2004) also indicate that differences in stock market participation between white households and other racial groups remain robust after controlling for wealth.

116 Financial Management Spring 2016 Table IV. Dividend Payout, Local Religious Affiliation, and Local Stock Ownership The dependent variable for Columns 1-3 is Dividend Payer. The dependent variable for Columns 4-6 is Dividend Yield, while the dependent variable for Columns 7-9 is Dividend Initiation. Columns 1-3 and Columns 7-9 use Logit regressions, while Columns 4-6 use OLS regressions. Panel A focuses on the differences between firms located in big metropolitan areas and firms located in other areas. Panel A1 repeats the main tests for firms located in areas excluding the three largest metropolitan areas (New York, Chicago, and Los Angeles), while Panel A2 repeats the main tests for the subsample of firms that are headquartered in the three largest metropolitan areas. In Panel B, local ownership is proxied by the only-game-in-town effect by following Hong et al. (2008) and it is measured by the number of firms per capita. Panel B1 provides results for the subsample of firms that are located in areas with a lower number of local firms per capita and Panel B2 presents the results for the subsample of firms that are located in areas with a higher number of local firms per capita. Panel C uses the fraction of local white households to measure the local retail stock market participation rate by following Hong et al. (2004). Panel C1 (C2) provides the test results for firms that are located in areas with a higher (lower) fraction of white households. Panel D uses institutional ownership to investigate the role of local ownership. Panel D1 (D2) includes firms with lower (higher) institutional holdings. The main regression control variables are Net Income, Cash, Q, Debt, Volatility, Lagged Return, Log of Assets, andasset Growth. All of the variable definitions are provided in Table II. All of the tests include the following firm age group indicator variables: Age 1-5, Age 6-10, Age 11-15,andAge 16-20. Age 21 and Over is the dropped category in the regressions. All of the tests include state, industry, and year dummy variables. In this table, only the local religious affiliation variables are displayed for brevity. Standard errors are adjusted for heteroskedasticity and clustered at firm level. Robust p-values are in parentheses. (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Panel A. Metropolitan Areas versus Other Areas Panel A1. Firms in Other Areas Cath 1.624 0.007 0.391 (0.001) (0.002) (0.489) Prot 2.269 0.010 1.430 (0.000) (0.000) (0.015) Cpratio 0.137 0.001 0.010 (0.003) (0.0005) (0.831) Number of Observations 71,708 71,708 71,708 71,631 71,631 71,631 47,188 47,188 47,188 R square 0.434 0.435 0.434 0.273 0.274 0.273 0.111 0.111 0.111 Panel A2. Firms in Metropolitan Areas Cath 3.832 0.001 3.306 (0.117) (0.879) (0.338) Prot 2.885 0.010 0.452 (0.479) (0.516) (0.916) (Continued)

Ucar Local Culture and Dividends 117 Table IV. Dividend Payout, Local Religious Affiliation, and Local Stock Ownership (Continued) (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Panel A2. Firms in Metropolitan Areas Cpratio 0.109 0.000 0.199 (0.573) (0.910) (0.300) Number of Observations 8,216 8,216 8,216 8,333 8,333 8,333 4,763 4,763 4,763 R square 0.469 0.468 0.468 0.293 0.293 0.293 0.165 0.164 0.165 Panel B. Number of Local Firms per Capita Panel B1. Firms Located in Areas with Low Number of Local Firms per Capita Cath 1.631 0.008 0.660 (0.007) (0.003) (0.448) Prot 3.275 0.013 1.791 (0.000) (0.000) (0.046) Cpratio 0.178 0.001 0.077 (0.006) (0.002) (0.357) Number of Observations 26,504 26,504 26,504 26,490 26,490 26,490 16,054 16,054 16,054 R square 0.476 0.479 0.476 0.346 0.348 0.346 0.118 0.119 0.118 Panel B2. Firms Located in Areas with High Number of Local Firms per Capita Cath 0.077 0.008 0.519 (0.923) (0.786) (0.472) Prot 0.035 0.003 0.741 (0.973) (0.480) (0.502) Cpratio 0.018 0.000 0.066 (0.751) (0.852) (0.182) Number of Observations 26,402 26,402 26,402 26,405 26,405 26,405 18,079 18,079 18,079 R square 0.411 0.411 0.411 0.212 0.212 0.212 0.107 0.107 0.108 (Continued)

118 Financial Management Spring 2016 Table IV. Dividend Payout, Local Religious Affiliation, and Local Stock Ownership (Continued) (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Panel C. Local Stock Market Participation Panel C1. Firms in Areas with High Fraction of White Households Cath 2.242 0.008 1.857 (0.001) (0.037) (0.032) Prot 2.229 0.009 2.807 (0.006) (0.031) (0.001) Cpratio 0.206 0.001 0.093 (0.002) (0.019) (0.176) Number of Observations 27,493 27,493 27,493 27,511 27,511 27,511 16,935 16,935 16,935 R square 0.424 0.423 0.424 0.272 0.271 0.271 0.124 0.126 0.123 Panel C2. Firms in Areas with Low Fraction of White Households Cath 0.849 0.001 1.675 (0.358) (0.621) (0.095) Prot 1.277 0.017 1.263 (0.299) (0.000) (0.460) Cpratio 0.063 0.001 0.092 (0.406) (0.443) (0.171) Number of Observations 26,512 26,512 26,512 26,496 26,496 26,496 17,812 17,812 17,812 R square 0.452 0.452 0.452 0.263 0.265 0.263 0.136 0.136 0.136 (Continued)

Ucar Local Culture and Dividends 119 Table IV. Dividend Payout, Local Religious Affiliation, and Local Stock Ownership (Continued) (1) (2) (3) (4) (5) (6) (7) (8) (9) Dependent Variable: Dividend Payer Dividend Yield Dividend Initiation Panel D. Institutional Ownership Panel D1. Firms with Low Institutional Holdings Cath 1.767 0.003 0.772 (0.021) (0.082) (0.358) Prot 2.799 0.008 2.332 (0.000) (0.001) (0.019) Cpratio 0.152 0.000 0.050 (0.031) (0.070) (0.496) Number of Observations 26,928 26,928 26,928 26,966 26,966 26,966 20,442 20,442 20,442 R square 0.341 0.342 0.341 0.1243 0.126 0.124 0.118 0.119 0.118 Panel D2. Firms with High Institutional Holdings Cath 1.162 0.003 0.257 (0.086) (0.193) (0.749) Prot 2.28 0.006 0.092 (0.005) (0.076) (0.930) Cpratio 0.048 0.000 0.005 (0.420) (0.178) (0.934) Number of Observations 25,680 25,680 25,680 25,756 25,756 25,756 13,994 13,994 13,994 R square 0.39 0.391 0.39 0.339 0.339 0.339 0.104 0.104 0.104 Significant at the 0.01 level. Significant at the 0.05 level. Significant at the 0.10 level.

120 Financial Management Spring 2016 Thus, I focus on the demographic characteristics provided by the Census data and use the fraction of local white households as a proxy for local retail stock market ownership. This method helps to identify the dividend effect for different levels of local stock participation, particularly for retail investors. Panel C1 of Table IV reports that firms located in areas with higher local market participation rates, as measured by the highest tercile of the percentage of local white households, elicit stronger findings than the main findings of Table II for most of the tests in Panel C1. However, the results for firms located in areas with lower local market participation rates, as measured by the lowest tercile of the percentage of local white households, are weaker and statistically insignificant in almost all of the tests in Panel C2. This table underlines the notion that the impact of local religion on dividend policies is stronger for firms headquartered in counties with high local ownership measured by high stock market participation. These tests suggest that firms consider local shareholder preferences on corporate policies in order to attract local investors money when local investors play an important role. Another way of investigating the importance of the locality of investors for my findings is to examine firms with different levels of institutional ownership. Prior literature indicates that firms with lower institutional ownership are expected to have higher levels of local and retail investor ownership. Coval and Moskowitz (1999) suggest that individual investors are more likely to own local stocks when compared to institutional investors. I examine the differences between firms with high and low institutional holdings. For every firm without missing information, I use institutional ownership figures from 13F filings in order to calculate the average annual institutional ownership. I then rank the sample into terciles based on this variable. Firms in the lowest (highest) tercile constitute the low institutional holding subsample serving as the high (local) local ownership subsample in the empirical tests. I rerun all of the main regressions for these two subsamples in Panels D1 and D2 conjecturing that my results are expected to be stronger for firms where local retail investors constitute a greater fraction of the investor bases. 2 The results in Panel D1 are similar to my earlier findings. Consistent with my assumption, the results are more pronounced and stronger for the subsample of firms with low institutional holdings. I find more pronounced results for firms with higher local ownership, as measured by lower institutional ownership. On the other hand, the results for the subsample of firms with high institutional holdings are either weaker or statistically insignificant in Panel D2. 3 This table provides additional evidence regarding the dividend clientele effect induced by local culture, particularly for dividend payer and dividend yield tests. I also examine the local religious affiliation coefficient differences across groups of firms with different levels of local ownership for all the tests of Table IV in the appendix. The results of the tests that focus on the local religious affiliation coefficient and statistical significance of differences in the appendix provide support to the findings of Table IV. Overall, Table IV helps to shed additional light on the channel through which local culture affects geographically varying dividend policies. Local cultural or religious 2 The logit regression model for dividend initiation regressions in the last three columns of Panels D1 and D2 does not converge when industry-fixed effects based on Fama and French s (1997) 48 industry classifications are used. This result can be attributed to the small number of observations in the dividend initiation compared to the dividend payer and the dividend yield tests. Thus, I use industry-fixed effects based on one-digit SIC codes in in the last three columns of Panels D1 and D2. The first six columns of Panels D1 and D2 have industry-fixed effects based on Fama and French s (1997) 48 industry classifications as in the other tests. When I employ one-digit SIC code classifications in the first six columns of Panels D1 and D2, I find results similar to the ones reported in this table. 3 In addition, in unreported tests, I divide the sample into subsamples based on time and compare the results between the earlier and later subperiods. I thank an anonymous referee for this suggestion. Most of the results are more pronounced for the earlier subperiod in this analysis. This finding is consistent with the expectation of a reduced level of local individual investor ownership with the rise of institutions over time. These results can be provided on request.