Dividend reputation, dividend yield and stock returns in Korea

Purpose – Although it has often been studied in ﬁ nance research, the relationship between dividend yields and stock returns remains an unresolved issue, especially in the Korean stock market. When ﬁ rms continue to pay non-decreasing dividends for three or ﬁ ve years, they may establish a dividend reputation, which could affect this relationship. The author found ﬁ rms that pay more dividends, larger ﬁ rms, older ﬁ rms, more pro ﬁ table ﬁ rms, less leveraged ﬁ rms, ﬁ rms with less volatile returns, ﬁ rms with foreign holdings of more than 5%, and ﬁ rmswith more concentrated ownership build dividend reputations. The author also found that the relationship between dividend yields and future stock returns depends on a ﬁ rm ’ s dividend reputation. The evidence shows that when ﬁ rms with higher yields have dividend reputations, they produce higher future returns, whereas there is no signi ﬁ cant relationship between yields and returns for ﬁ rms with no reputation. These results are inconsistent with the ﬁ ndings of studies that use developed market data. In addition, when larger ﬁ rms with higher growth potential and ﬁ rms with less concentrated ownership have dividendreputations, futurereturns arehigher.


Introduction
There are conflicting opinions in finance regarding the relationship between dividend yields and stock returns. Although Gordon and Shapiro's (1956) well-known "Gordon growth model" and Campbell and Shiller's (1988) dividend-ratio models suggest that dividend yield is predictive of stock returns, empirical studies have found a variety of relationships between dividend yields and stock returns. In an aggregated level, Fama and French (1988), Hodrick (1992), Kothari and Shanken (1997), Naranjo et al. (1998), Lewellen (2004), Campbell and Yogo (2006), Chiquoine andHjalmarsson (2009), Ferreira andSanta-Clara (2011) and Golez (2014) indicate that dividend yields have a strong positive relationship with expected returns on the market. On the other hand, Goetzmann and Jorion (1993), Wolf (2000), Lanne (2002) and Welch and Goyal (2008) find no significant evidence indicating that dividend yields can forecast stock market returns. Goyal and Welch (2003) argue that prior to 1990, the conditional dividend yield could reliably predict future equity premia in-sample, however, this predictive ability of dividend ratios is no longer applicable. Ang and Bakaert (2007) find that dividend yields predict excess returns only at short horizons and do not have any long-horizon predictive power.
Studies using data from individual securities also report conflicting findings. Litzenberger and Ramaswamy (1979), Blume (1980), and Keim (1985) show that returns for dividend-paying stocks tend to increase as dividend yield increases. In this regard, practitioners use high dividend yield strategies as one profitable stiletto outperform market returns in many markets. McQueen et al. (1997) and Visscher and Filbeck (2003) show that high dividend yield stocks outperform the market. Chen et al. (1990) use a single stock market factor model to document the positive relationship between dividend yields and expected stock returns, but they find no reliable cross-sectional relationship between dividend yields and risk-adjusted expected returns. Furthermore, Miller and Scholes (1982) report no significant relationship between expected returns and dividend yields. Filbeck and Visscher (1979) also demonstrate that high dividend yield stocks do not systematically outperform the market index using the UK data.
Prior studies suggest various reasons for these mixed results. For instance, Keim (1985) proposes it is due to the tax effect on dividend income, while Maio and Santa-Clara (2015) show that there is a positive relationship between dividend yields and stock returns for the aggregate stock market, but not for portfolios of small and value stocks.
Further, Chen et al. (1990) find that at least part of the relationship between stock returns and dividend yields can be attributed to dividend-related changes in risk measures. As another possible reason, this study examines the dividend reputation effect on the relationship between dividend yields and stock returns. That is, the relationship is expected to differ depending on whether or not a firm has established a dividend reputation. Wilson (1985) and Gillet et al. (2008) explain that a reputation is the result of people's perceptions of a firm, which are based on observed past behavior and especially on the assumption that this behavior will continue in the future. Thus, they describe a reputation as built over time, based on observable behaviors or firm characteristics. Further, it is generally known that companies consider dividend reputations when determining their dividend policies. In this, regard Lintner (1956), Fama and Babiak (1968), Brav et al. (2005) and Leary and Michaely (2011) indicate that firms are obviously concerned with the stability of dividends. The companies tend to keep their stream of dividend payments stable despite transitory shocks in their earnings. Syed et al. (2018) claim that this tendency has direct implications for a firm's market reputation as a payer, thus, that the literature views a reputation as an essential consideration to decide dividend policy. For example, La Porta et al. (2000) introduce their substitution model that firm insiders pay high and steady dividends to establish their market reputation.
Therefore, the literature implies that not only is building a dividend reputation among existing and potential investors a crucial factor in corporate dividend policies but investors' evaluations of corporate dividend policies also differ depending on whether firms have built dividend reputations. That is, the relationship between dividend yields and stock returns is expected to vary by firm dividend reputation.
The important issue in this study is how to measure a dividend reputation. Prior studies focusing on corporate reluctance to cut or change dividends associated with a reputation primarily investigates dividend smoothing, which assumes that companies adjust the dividend in the given year based on dividend only in the previous year. As Wilson (1985) and Gillet et al. (2008) explain, however, a reputation is established over a long period of time, not just one year. We need to observe a dividend stream for several years to identify whether firms earn a reputation from their dividend. Additionally, most methodologies of studies on dividend smoothing weigh increases and reductions in dividends equally, JDQS 29,1 although Brav et al. (2005) report that firm managers express a strong desire to avoid dividend cuts except in extraordinary circumstances. Indeed, investors prefer increases of dividend payments to decreases.
On the other hand, Kang et al. (2019) propose a measure for a dividend reputation, complementing this existing literature. They distinguish firms with a dividend reputation which pay non-decreasing dividends for several years (e.g. at least three years) from firms that experience reductions in dividends for the same period more than once, thus, cannot build a reputation. Their assumption is that investors expect that firms will keep or increase its dividends in the next period based on observed past behavior of steadily paying nondecreasing dividends over time; consequently, the firms establish a dividend reputation. As their measure takes long-term dividend series and the firms' reluctance to reduce dividends into account, its identification of dividend reputations closely corresponds to investors' practical expectations and companies' payout policies. Thus, I use their measure to investigate factors that affect whether firms build dividend reputations and how a dividend reputation influences the relationship between dividend yields and stock returns.
This study uses data from the Korean financial markets, where there is a very unclear relationship between dividend yields and stock returns. Choi (2016) finds that the dividend price ratio of Korean stocks does not predict next year's stock return.  and Chung and Kim (2010) show that dividend yields in the Korean stock market do not have the predictive power on stock returns. Kim and Seo (2011) cannot find any significant relationship between the dividends and returns on individual stocks. Further, while Kim (2018) reports that a high dividend yield portfolio outperforms the Korean stock market from 1987 to 2017, the KRX High Dividend index that consists of 50 stocks with high dividend yields among the KOSPI and KOSDAQ universe since October 2014 produce an accumulative monthly return of 6.60% from the beginning to December 2019, less than the KRX 100 index of 13.25% for the same period. This ambiguous association in the Korean market is more severe than in developed countries, which could be attributed to several emerging market features. According to Jeong (2013), Korea has experienced massive economic growth over the decades and firms with high growth opportunities are attracted to paying dividends to signal their information to outside investors, at the same time, to retain a large part of earnings to assist with their efficient projects. Indeed, Korean firms are known to tend to pay extremely less dividends in comparison with the developed markets. Moreover, Kim and Yi (2006), Jeong (2013) and Kang et al. (2014) discuss that many of Korean firms belong to business groups linked with their affiliated firms via circularshareholdings, thus, they are likely to have a large discrepancy between voting rights (control) and cash flow rights (ownership) of controlling shareholders. They argue that this has the possibility to cause agency problems, in which controlling shareholders expropriate minority shareholders' wealth. Even Korea also has relatively weak protections for outside minority shareholders (Bae et al., 2002). Consequently, emerging market firms with poor corporate governance such as Korean stocks might pay dividends to mitigate high agency costs. On the other hand, Lee (2011) documents those corporate dividends in family business groups sometimes act as a means of transferring capital within a business group in the Korean market. With these emerging market features, as various market reactions to stock dividend yields may be offset in the Korean stock markets, the relationship between dividend yields and stock returns is unclear. Consequently, I suggest that a dividend reputation Hasan important influence on the relationship and is an important dividend policy for companies in emerging markets.
This study examines the determinants of dividend reputation building for firms, and the association between dividend yields and stock returns for firms with and without dividend Dividend reputation reputations. The evidence from logit regressions and Fama and MacBeth's (1973) crosssectional regressions indicates that firms that pay more dividends, larger firms, older firms, more profitable firms, less leveraged firms and firms with less volatile returns are more likely to establish dividend reputations, measured as no reduction in regular cash dividends over three, five or seven years. These results are consistent with the literature discussing that mature firms tend to initiate or increase dividend payments with diminishing investment opportunities (Fama and French, 2001;Grullon et al., 2002). Indeed, firms that are able to retain or increase regular dividend payments to build a dividend reputation would be mature, are likely to earn stable profits and be less risky. Interestingly, the results for growth opportunities measured by sales growth and market-to-book ratio are mixed. Firms with dividend reputation do not have a significant relationship with sales growth but they have a higher market-to-book ratio. Flavin and O'Connor (2017) argue that firms with larger growth opportunities (higher market-to-book ratio) in Korea tend to pay larger dividends to build a reputation, in contrast to the findings for firms in more developed regimes. Supporting their findings, the results in this paper imply that firms with larger growth opportunities are likely to build dividend reputation, regardless of past growth performance. Moreover, I find that among small firms, firms with a low market-to-book equity ratio are likely to build a dividend reputation, but among bigger firms, firms with a higher market-to-book ratio tend to build a reputation. The result of this study, in which firms with foreign holdings of more than 5% are apt to pay non-decreasing dividend streams, supports this interpretation. In addition to corporate governance, firms with more concentrated ownership establish a dividend reputation. This result does not support the agency theory for corporate dividend policies (Jensen, 1986) or dividend smoothing (Leary and Michaely, 2011).
I also find that the relationship between dividend yields and future stock returns depends on a firm's dividend reputation. I divide firms within the KOSPI and KOSDAQ universe into two groups based on whether the firm establishes a dividend reputation and investigate separately for each group the relation using panel regressions and Fama-MacBeth crosssectional regressions. The evidence indicates there is a significantly positive relation between dividend yields and future returns (after including control variables) in reputationestablished firms, whereas no reputation firms do not have any significant relation. Similarly, results from equal regressions using the entire sample of the one-year future stock return on the dividend reputation dummy variable, the dividend yield and the interaction terms between the reputation dummies and some relevant variables including the dividend yield provide reliable evidence of the reputation effect on the relation between yields and returns. Although the coefficients of the yield are not significant, those of interaction terms between the reputation dummy and the yield have significantly positive coefficients, which means that only reputation-established firms among those with higher yields are expected to produce higher future returns. This result runs counter to Kang et al. (2019) that show interaction terms between the reputation dummy and yields have significantly negative coefficients in the regressions of monthly excess returns in the USA. The document that firms with a dividend reputation tend to have less risk compared to firms without a dividend reputation and the expected return of firms with a dividend reputation will be lower given the dividend yield, taking the positive relationship between yields and returns in the USA into account. However, Korean stocks, having rapidly grown, have no significant association between dividend yields and returns. Nevertheless, when reputation-established firms provide higher dividend yields in Korea, investors will be willing to expect their higher value. Especially when larger firms with higher market-to-book ratios have a dividend reputation, the future returns are higher. That is, markets also positively react on JDQS 29,1 mature firms with higher growth opportunities paying a non-decreasing dividend series as signals.
For ownership, although firms with more concentrated ownership are apt to build a dividend reputation, their future returns tend to be lower when they have a dividend reputation. In other words, investors may react on these companies negatively in Korean markets. As firms with more concentrated ownership have lower agency costs and need not solve the agency problem by paying non-decreasing dividend streams regardless of earnings, investors may regard these continuous non-decreasing dividend payouts as expropriation of minority shareholders' wealth in the Korean environment with poor corporate governance and weak protections for minority shareholders. Additionally, I find that the interaction has no effect between foreign holdings and dividend reputation on stock returns.
The rest of the paper is organized as follows. Section 2 describes the data and variables in this study. Section 3 identifies the determinants of dividend reputation-building for Korean markets. Section 4 examines the reputation effect on the relationship between dividend yields and future returns. Section 5 concludes the paper.

Data and variables
I use all firms traded on the KOSPI and KOSDAQ markets of the Korean Stock Exchange (KSE) from January 1998 to December 2019. Mutual funds, REITs, ETFs, SPACs and preferred stocks are excluded. All financial data used in this paper, including dividends, come from a Data Guide. As most firms in Korea that regularly pay dividends to make dividend payments once a year, the sample includes only common stocks that pay yearly regular cash dividends; firms that pay quarterly or semiannual regular cash dividends more than once during the sample period are excluded. I also eliminate firms that pay occasional special dividends or stock dividends more than once during the sample period. The presence of these dividends precludes determining whether firms have a dividend reputation because these may complement or substitute for regular cash dividends. To reduce the bias related to the market-microstructure, I only consider stocks with prices above 1,000 KRW during the sample period.
Following Kang et al. (2019), I divide the sample firms into two groups based on regular dividend streams: (1) RPT5 includes reputation-established firms that continue to pay non-decreasing dividends per share for at least five years ending on the fiscal year-end date and (2) NR includes no-reputation firms that pay decreased dividends per share from a previous year over the recent five years.
I create a dummy variable, RPT5, which equals one for all firms in RPT5 and zero for all firms in NR. In other words, if a firm maintains previous dividend payments or increases dividends for the latest five years in year t, RPT5 = 1 for year t þ 1. In their study using US stock data, Kang et al. (2019) define firms that pay non-decreasing dividends for at least three years as reputation-established firms and Kang (1997) documents that it usually takes around three years for a firm to establish a reputation. As Korean companies make annual dividend payouts, they provide fewer dividend payments during the same period when compared to US firms, which generally pay regular quarterly cash dividends; I observe the reputation-building behavior of firms for five years. As a robustness test, I use other reputation-building periodsthree years following Kang et al. (2019) or seven yearswith no change in the main results. This is discussed in more detail later.

Dividend reputation
As variables associated with dividends, I use dividends per share(Div) and dividend yield (DY). The dividend per share is measured by Log(Div), which is the natural logarithm of the cash dividend per common share plus one and dividend yield is measured by Log(DY), which is the natural logarithm of the Divscaled by the closing stock price at the end of the fiscal year plus one. In addition, DDY is the yearly change in dividend yield for an individual firm, which is defined as the difference in the dividend yield (in logarithm) between the current and previous years.
Based on the literature, I use several sets of variables to control for firm-specific determinants of dividend payout policy. For example, Fama and French (2001), Grullon et al. (2002) and Denis and Osobov (2008) find that the propensity to pay dividends is higher among larger firms, more profitable firms, firms with fewer investment opportunities and firms with less leverage. It is possible for these characteristics to be associated with reputation building through dividend policy. I first consider two proxies for firm maturity: firm size and age. Many studies argue that larger and older firms faceless information asymmetry (Frank and Goyal, 2003;Lemmon and Zender, 2010;Leary and Michaely, 2011). Firm size, Log(MV), is measured using the natural logarithm of the firm's market capitalization at fiscal year-end and firm age, Log(Age), is measured as the natural logarithm of the number of years the firm has been listed on the KSE (i.e. KOSPI or KOSDAQ) plus one. Firm profitability is captured by ROA and Cashflow. ROA is the operating income before depreciation scaled by the book value of total assets and Cashflow are cash flows from operations scaled by total assets. The past one-year return, r 1Y , measures a firm's performance in the stock market over the past one year. Cashflow is also used as a proxy for the agency costs of free cash flow, as in Jensen (1986). To capture investment opportunities, the sales growth rate, GS and market-to-book ratio, MB (Fama and French, 2001;Grullon et al., 2002;Grullon and Michaely, 2007;Adam and Goyal, 2008;Denis and Osobov, 2008;Leary and Michaely, 2011). GS is calculated using the annual average of the latest three-year sales growth rate and MB is the market capitalization divided by the book value of equity at fiscal year-end. Leverage, LVRG, is measured by total liabilities over total assets. As proxies for volatility, I include the volatility of both earnings and stock returns as measures of risk and information asymmetry (Brennan and Subrahmanyam, 1996;O'Hara, 2003;Leary and Michaely, 2011). s r1Y measures the volatility of stock returns, estimated by the annual standard deviation of weekly stock returns and following Chay and Suh (2009), s ROA measures earnings volatility or cash uncertainty, estimated by the standard deviation of ROA over the prior five years. I further include measures of corporate governance. La Porta et al. (2000), Brav et al. (2005), Michaely and Roberts (2006) and Leary and Michaely (2011) show that corporate governance is related to dividend policies, which capture exposure to agency problems. Officer (2010) also, documents that predicted dividend payers with weak governance are more likely to pay dividends than dividend payers with strong governance. La Porta et al. (2000) predict that firms use dividend policy as a substitute for weak governance. Thus, the proportion of stock held by the largest shareholder, LargeSh, and the proportion of stock held by minority shareholders, MinorSh, are included as proxies for governance. The ownership of the largest shareholder reported by Data Guide covers the largest shareholder and related parties' holdings, which is a more comprehensive measure of ownership concentration. Additionally, firms that want to signal to foreign investors through dividend payouts may seek to establish a dividend reputation. For example, Sul and Kim (2006) show that higher foreign ownership is related to higher dividend yields; therefore, foreign holdings, ForeignSh, are included. Sul and Kim (2006) find that firms with foreign holdings of more than 5% have a significant impact on the increase in dividend payouts; hence, I add a dummy variable, FMsh, which equals 1 if foreign holdings are more JDQS 29,1 than 5% and 0 otherwise. Finally, the market return, r mt , is a value-weighted average of index returns on the KOSPI and KOSDAQ markets and the risk-free rate, r ft , is measured using the CD91 rate. Table 1 defines all variables used in this study.
All variables, except returns, are winsorized at the 1st and 99th percentiles to reduce the influence of outliers. Consequently, the final sample consists of 2,137 firms and 24,343 firmyear observations. Table 2 presents the descriptive statistics of the variables for the sample firms in the full sample period. According to the table, the distributions of the variables is not too extreme to investigate by winsorizing. The average dividend per share is 295 won and the median is 60 won. The average dividend yield is 1.654% and the median is 1%. The average and median of the yearly change of dividend yield is À0.014 and 0, which means Table 1.

Variable definitions Variable Definition
Dummy variables for dividend reputation RPT3 One if a firm continues to pay non-decreasing dividends for at least three years and zero otherwise RPT5 One if a firm continues to pay non-decreasing dividends for at least five years and zero otherwise RPT7 One if a firm continues to pay non-decreasing dividends for at least seven years and zero otherwise Variables for firm dividend payments Log(Div) Natural log of Korea won value per common share of distributions resulting from cash dividends at the end of the fiscal year plus one Log(DY) Natural log of cash dividends per common share divided by the closing stock price at the end of the fiscal year plus one DDY The natural log of the dividend yield in a given year minus the natural log of the dividend yield in the previous year

Variables for firm characteristics Prc
Closing stock price at the end of the fiscal year Log(MV) Natural log of the number of common shares outstanding multiplied by the closing stock price at the end of the fiscal year Log(Age) Natural log of one plus the time (in years) from the firm's first listing date in KOSPI or KOSDAQ markets plus one ROA Operating income before depreciation divided by the book value of total assets Cashflow Cash flows from operating divided by the total assets r 1Y Past one-year stock returns GS Five-year average of one-year growth in sales MB Market value of common equity divided by the book value of common equity LVRG Liabilities divided by the total assets s r1Y Standard deviation of weekly stock returns over the one-year period s ROA Standard deviation of ROAs over the prior five years LargeSh Proportion of the stock held by the largest shareholder and the related parties MinorSh Proportion of the stock held by the minority shareholders ForeignSh Proportion of the stock held by the foreign investors FMsh One if the foreign holding is more than 5% and zero otherwise Variables related to markets r mt Value-weighted average of index returns of the KOSPI and KOSDAQ r ft Proxy of the risk-free rate, the CD91 rate Dividend reputation that most of the firms do not always increase dividend yield, but on average, they maintain their dividend yields.

Determinants of reputation building
3.1 Characteristics of reputation-established firms I compare two groups, reputation-established firms (RPT5) that continue to pay nondecreasing dividends per share for at least five years and no-reputation firms (NR) that decrease or cut dividends more than once in the prior five years. Table 3 shows the separate statistics for RPT5 and NR firms and the results of a t-test of the differences in their variable means. The number of reputation-established firms is much smaller than no-reputation firms. RPT5 contains 2,910 firm-years, while NR contains 21,433 firm-years. Div and DY for RPT5 firms are significantly larger than those of NR firms. According to previous studies such as Fama and French (2001), Denis and Osobov (2008) and He (2012), firms with a consecutively stable dividend stream may pay relatively higher dividends based on their length of maturity and profitability. Further, as Table 3 shows, prices of RPT5 firms are higher than those of NR firms; thus, the relatively high DY of RPT5 firms is due not to low prices but absolutely higher dividend payouts.
According to Log(MV) and Log(Age) in Table 3, RPT5 firms are significantly larger and older; thus, they are more mature. This indicates that relatively mature firms probably pay stable non-decreasing dividend streams. As ROA and Cashflow are significantly higher for RPT5 firms, reputation-established firms are associated with higher profitability and larger cash flows from firm operations. However, stock market performance, r 1Y , differs from the operational performance. RPT5 firms have lower stock returns over the past year than NR firms, but the difference between the two groups is not significant. The results regarding investment opportunities are mixed; the GS of RPT5 firms is significantly lower than that of NR firms, but the difference in MB between the two groups is insignificant. The results from Mean (2) Sth (3) N (4) Mean (5) Sth (6) Difference [(2)- (5) Table 3. Differences between firms with and without dividend reputation Dividend reputation LVRG show that reputation-established firms have relatively low debt. The results of the firms' risk measures, s r1Y and s ROA , show that firms with reputations have less stock market and operational risk and these are consistent with prior studies. That is, firms that are more exposed to risk have difficulty maintaining their previous dividend level. RPT5 firms are larger, older and less risky, which indicates that these firms are associated with less asymmetric information, based on Brennan and Subrahmanyam (1996), O'Hara (2003) and Leary and Michaely (2011). Finally, results from LargeSh, MinorSh and ForeignSh reveal that firms with concentrated ownership and firms with relatively high foreign ownership do not decrease dividend payouts over the previous five years. This implies that building a dividend reputation is not related to principal-agent conflicts, as explained by La Porta et al. (2000) and Brav et al. (2005).
These results are consistent with the literature regarding dividend smoothing in Korea. Jeong (2013) finds that larger firms, lower growth firms, less risky firms and firms with more concentrated ownership tend to smooth dividends more, which implies that neither information asymmetry-based models nor agency cost-based models are supported for dividend payout policies in Korea.
The correlation coefficients among the study variables are presented in Table 4. I report the Spearman correlation coefficients between RPT5 and other variables and the Pearson correlation coefficients among the variables except for RPT5. P-values are in parentheses. Overall, the correlations between the reputation dummy variable, RPT5, and other variables are not high; most absolute values are less than 0.2. Among them, Log(DY), Log(MV) and Log(Age) are the most positively correlated with the reputation dummy, consistent with the findings in Table 3. That is, firm maturity has a greater impact on non-decreasing and dividend payout decisions than firm profitability. s r 1Y has also a relatively high, negative correlation with the reputation dummy.

Logit regressions
Next, I use a set of regressions to examine the relationship between reputation dummies and other relevant variables identified as factors that may impact firm dividend payouts. Because the dependent variables are binomial variablesthe reputation dummies, RPT3, RPT5 or RPT7 -I use logit regressions. The explanatory variables are the dividend yield in the previous year, Log(DY) t-1 , Log(MV) t , Log(Age) t , ROA t , CashFlow t , r 1Y,t , GS t ,MB t ,LVRG t , s r1Y,t and s ROA,t . I use either LargeSh t or MinorSh t to measure ownership concentration because the correlation coefficient between LargeSh and MinorSh is very high (-0.694). Also, I add either ForeignSh or FMsh for the foreign ownership variable, which are the proportion of stock held by foreign investors and the dummy equal to 1 for firms with foreign holdings of more than 5% and 0 otherwise, respectively. Dividend yields in the previous year are used to solve the potential endogeneity problem. The logit regressions include year dummies and industry dummies based on major sections of the KSIC (Korean Standard Industrial Classification) and t-statistics are based on robust standard errors clustered by firm. The results are reported in Panel A of Table 5.
The reputation dummy, RPT5, is positively related to dividend yield, firm size, age and profitability. Following Zhang (2006) and Lu et al. (2010), this result supports the idea that mature firms with less information asymmetry have already established dividend reputations. The coefficients of ROA and CashFlow reveal that profitability from operations rather than cash flows affects reputation building. However, the coefficient of r 1Y shows that stock market performance has a marginally negative relationship with reputation building. The impact of past stock returns on reputation building through dividend payouts is likely to be relatively weak, noting the insignificance of the difference in r 1Y between the RPT5     and NR groups in Table 3. Contrary to stock performance, stock return volatility, s r1Y , has a significantly negative association with reputation building. Taking the significantly negative coefficient of earnings volatility, s ROA and the leverage ratio, LVRG, into account as well, less risky firms with less information asymmetry and firms with low debt are likely to pay non-decreasing dividends. Interestingly, results from the investment-related variables are mixed. GS has a negative coefficient that is significant at the 10% level, while the coefficient of MB is significantly positive with a t-statistic of about 5. Consistent with Lee and Kong (1994), Kim et al. (2010) and Flavin and O'Connor (2017), Korean firms with greater growth opportunities tend to seek to build dividend reputations, in contrast to the findings for firms in more developed markets. As they have larger investment opportunities, they may signal this information to potential investors in emerging markets with countrylevel institutional barriers to investment. For corporate governance, the significantly positive coefficient of LargeSh and significantly negative coefficient of MinorSh indicate firms with concentrated ownership tend to maintain or increase dividend payouts to establish a reputation, which suggests Korean firms' actions are not consistent with the agency theory-based explanation for dividend policies (La Porta et al., 2000;Brav et al., 2005) or dividend smoothing (Leary and Michaely, 2011). According to these prior studies, firms with relatively low largest holdings and high minority holdings, suggesting the existence of agency conflicts, are likely to pay large dividends and smooth dividends more. La Porta et al. (2000) document that by paying dividends, insiders return corporate earnings to investors and hence are no longer capable of using these earnings to benefit themselves. According to their outcome model, more dividends are paid because minority shareholders pressure corporate insiders to disgorge cash. Also, Leary and Michaely (2011) find that firms subject to agency conflicts smooth the most. However, this study's results do not support their views. Finally, the coefficient estimates of ForeignSh and FMsh in the logit regressions imply that foreign holdings have no significant association with corporate dividend reputations but the presence of foreign holdings of more than 5% are significantly related to building reputations through dividend streams. Literature such as Sul and Kim (2006) also finds that dividend policies are more related to the presence of foreign holdings of more than 5% than just foreign holdings. Sul and Kim (2006) argue that foreign investors as majority shareholders may affect firms to pay higher dividend payments. The evidence in this study suggests that foreign investors do not prefer decreases in dividends, and further, may use their voting power to demand non-decreasing dividend series. I also perform logit regressions of the reputation dummies with other reputation-building periods, RPT3 or RPT7, using the same explanatory and control variables. Overall, the results are robust across different reputation-building periods that is firms with higher dividend yields, larger firms, older firms, more profitable firms, less leveraged firms, firms with less volatile returns and firms with more concentrated ownership build dividend reputations. The marked differences across the reputation-building periods are the following: the longer the reputation-building period, the stronger the significance of GS; MB has a significantly positive relationship with the reputation dummies for three and five reputation-building years but does not have a significant relationship in the seven-year period; and the longer the reputation-building period, the lower the coefficient and significance of ROA and the higher the coefficient and significance of FMsh. In other words, when the reputation-building period is shorter than five years, the growth rate is not significantly related to reputation but MB is significantly related, although when the reputation period is longer, the growth rate is significantly and negatively related to reputation but MB is not significant. Also, the increases in the coefficients and significance of FMsh for the longer reputation-building period imply that foreign investors as majority Dividend reputation shareholders may impact corporate dividend policies to ensure companies do not reduce dividend payments.
In addition, I divide the full sample into two sub-periods, from 1998 to 2008 and from 2009 to 2019 and replicate the previous analyses for each sub-period, to compare the relations between dividend reputation and the firm characteristic variables before and after the 2008 financial crisis. The results are reported in Table 6. The overall results for the two sub-periods are very close to previous outcomes for the full sample reported in Table 5. However, the market-to-book ratio does not have a significant relation to reputation during 1998$2008, contrary to the positive relationship during 2009$2019. The evidence indicates that the growth opportunities are not related to building reputation before 2008, but firms with larger growth opportunities are likely to pay non-decreasing dividend payouts after 2008. For corporate governance, the relatively concentrated ownership of firms with a reputation is more significant after 2008 than before 2008. Also, the coefficients of FM share significant before 2008, but not after 2008.

Cross-sectional regressions
I examine cross-sectional differences in the determinants of reputation-building behavior using Fama and MacBeth's (1973) cross-sectional regressions to account for time-specific economy-wide shocks. Each year, I regress reputation dummies on lagged Log(DY) and other relevant variables used in the prior logit regressions. The regression coefficients are then averaged over time; standard errors and t-statistics are computed and reported in Table 7.
The dependent variables are RPT5 in Panel A, RPT3 in Panel B and RPT7 in Panel C. The evidence confirms the logit regression results: firms with higher dividend yields, larger   Dividend reputation firms, older firms, more profitable firms, less leveraged firms, firms with less volatile returns, firms with more concentrated ownership and firms held by more foreign investors tend to establish dividend reputations. Consistent with Table5, only when the reputationbuilding period is seven years, the GS is significantly and negatively related to the reputation dummy. Cross-sectionally, the foreign ownership effect is larger than in the logit regressions. Both ForeignSh and FMsh have larger coefficients and greater significance in the Fama-MacBeth regressions.
3.4 Reputation-established firms and portfolios formed on size and market-to-book ratio As the previous results indicate that firms with greater growth opportunities as captured by market-to-book equity unexpectedly tend to pay non-decreasing dividends, as do larger firms, as expected, I investigate which firms seek to build dividend reputations based on size and the market-to-book ratio. Each year, I sort the full sample of firms by size, Log(MV) and (independently) by market-to-book equity, MB. I construct 25 portfolios from the intersections of the size and MB quintiles and count the number of firms for which the reputation dummy, RPT5 or RPT3, equals one for each of 25 portfolios. I also analyze the equivalent portfolios formed by RPT7 and the findings are very close to the results with RPT5 and RPT3. Thus, to save space, I do not report these findings. Table 8 shows the number and proportion of reputation-established firms. For firms in the smaller size quintiles, the lowest market-to-book quintiles have the greatest number of reputationestablished firms. However, for firms in larger size quintiles, the higher market-to-book quintiles have the most reputation-established firms. For instance, among firms in the smallest quintile, the lowest market-to-book quintiles have 118 firms with reputations over Table 8. Number of reputationestablished firms for 25 portfolios formed on size and marketto-book ratio at least five years; the percentage in both the smallest and lowest market-to-book quintiles is 7.5%, which is the highest. In contrast, among the portfolios in the largest size quintile, the highest market-to-book quintile has the largest stocks, 312 and the second highest marketto-book quintile has the largest percentage of firms with reputations, 24.7%. These patterns are similar for firms whose RPT5reputation dummy equals one. The evidence implies that among small firms, firms with higher market-to-book ratios may not be able to pay nondecreasing dividends because they have large investment opportunities and should retain their earnings due to lack of capital. However, among big firms, firms with higher market-tobook ratios are likely to pay stable and non-decreasing dividends because they also have large investment opportunities and can convey their information to external capital providers through dividend payout policies.

Relationship between dividend yields and future stock returns based on dividend reputation
There are conflicting opinions in finance regarding the relationship between dividend yields and stock returns. Litzenberger and Ramaswamy (1979), Blume (1980), Keim (1985), Fama and French (1988), Hodrick (1992), McQueen et al. (1997), Naranjo et al. (1998), Lewellen (2004), Campbell and Yogo (2006) and Visscher and Filbeck (2003) show that dividend yields have a strong positive relationship with expected market returns. However, Miller and Scholes (1982), Goetzmann and Jorion (1993), Filbeck and Visscher (1979), Wolf (2000), Lanne (2002) and Welch and Goyal (2008) argue that there is no significant relationship between expected returns and dividend yields. Moreover, literature using Korean corporate data, such as , Chung and Kim (2010), Kim andSeo (2011) andChoi (2016), do not find evidence that dividend yields can forecast returns. This obscure association may be caused by the emerging market nature of the Korean financial and corporate environment, which investors perceive as a relatively risky and growing market where firms pay lower dividends than developed markets. Therefore, I expect dividend reputations to impact this association because even in this environment, reputations can provide potential investors with a prediction of the firm's future dividend payout policies or cash flows, thus affecting investor reactions.

Regressions for dividend reputation and no-reputation groups
In Section 2, all firms are divided into two groups based on their regular dividend streams; RPT3 or RPT5 include reputation-established firms over at least three or five years and NR includes no reputation firms over three or five years. To reveal the difference in the relationship between yields and returns for RPT and NR firms, I run firm-year fixed effects panel regressions for each group of one-year future excess returns on the variable for dividend yields, Log(DY). Excess returns are stock returns minus the risk-free rate measured by the CD91 rate. As control variables that influence stock returns, I include DDY, Log(MV), ROA, GS, MB, LVRG and r m . t-statistics are based on robust standard errors clustered by the firm and the results are reported in Table 9.
In Panel A, the reputation-building period is five years; that is, RPT5 includes reputation-established firms for at least five years. Columns (1) and (3) do not include DDY as a control variable; only Log(DY) is used as a dividend-related variable. As shown in Column (1), for RPT5, the dividend yield has a positive relationship with one-year future excess returns, whereas for NR in Column (3), the yield has no significant relationship with future excess returns. To determine whether the positive relationship for RPT5 is caused simply by an increase in dividend payments, I add DDY to the panel regressions in Columns (2)  includes reputation-established firms for at least three years and NR includes no-reputation firms over three years. The variables are defined in Table 1. t-statistics are based on standard errors adjusted for the firm-clustering effect and are in parentheses. *** , ** and * denote statistical significance at the 1, 5 and 10% levels, respectively. The sample period is from 1998 to 2019 Table 9. Relation between stock returns and dividend yields for reputation and no reputation groups: panel regressions JDQS 29,1 the group. For reputation-established firms, the degree of increase in dividend yield does not affect future returns, but for no-reputation firms, it has a positive relationship. The results shown in Panel B, where the reputation-building period is three years, are consistent with those in Panel A. In other words, dividend yield positively affects future returns but changes in yield do not significantly affect returns for the group of reputation-established firms, while yield has no significant relationship with future returns but changes in yield positively affect returns for the group of no-reputation firms.
Consequently, dividend yield and changes in dividend yield have different impacts on stock returns depending on whether firms build a dividend reputation. Further, the evidence in this study differs from that of Kang et al. (2019), who use US stocks. They find that all firms in their sample have a positive relationship between yields and returns, but the positive relationship is weaker for reputation-established firms than other firms. They interpret this finding as indicating firms with dividend reputations tend to have less risk compared to firms without dividend reputations, and thus produce relatively lower returns. However, in Korea, the positive relationship between yields and returns only for reputationestablished firms may be induced by the higher growth potential of mature firms, considering the positive coefficient estimates of size, age and the market-to-book equity ratio in Tables 5 and 7, consistent with Flavin and O'Connor (2017) argument. Additionally, changes in dividend yield do not affect future returns for reputation-established firms but positively affect future returns for no-reputation firms. This implies that investors are sensitive to yield fluctuations for firms without reputations, but yield fluctuations of firms with reputations are not important to investors; higher yields matter. Investors can expect to enjoy continuous high yields based on a non-decreasing dividend payout reputation.
The values within R 2 for reputation-established firms in Panels A and B are more than twice as high as the within R 2 for no-reputation firms. This evidence indicates that the effect of dividend yields for RPT as the determinant of stock returns is higher than for NR. In equivalent panel regressions (not reported) with only Log(DY) as the explanatory variable and a reputation period of five years, within R 2 is 11.05% for RPT5 and 0.35% for NR, a large difference.
No-reputation firms in NR have a positive association between ROA and future returns while RPT5 or RPT3 firms do not show a significant relationship. As reputation-established firms may already have a reputation for their payout policies that allow investors to expect continuous earnings and payouts, profitability is not the determinant of future returns for them. However, for firms without reputations, profitability affects investor reactions as suggested by the results from the change in yields, DDY. Also, NR firms do not have a significant relationship between MB and future returns, whereas RPT firms do have a significantly negative relationship, which implies that value firms with reputations tend to produce higher returns.
I examine cross-sectional differences in the relationship between returns and yields for the two groups using Fama-MacBeth cross-sectional regressions and the same variables in Table 9. The results are reported in Table 10 and the previous findings are robust to cross-sectional regressions. That is, for reputation-established firms, the dividend yield has a positive relationship with one-year future excess returns, whereas for no-reputation firms, the yield has no significant relationship with future excess returns. Also, for reputation-established firms, the degree of increase in dividend yield does not affect future returns, but for no-reputation firms based on the five reputation-building periods, it has a positive relationship.   Tables 9 and 10 and in Flavin and O'Connor (2017), I add interaction terms between the reputation dummies and relevant variables such as size, change in yield, market-to-book ratio, the largest shareholder ownership and foreign ownership because they are primary factors in reputation building. I use FMsh for the variable for foreign ownership, instead of ForeignSh because FMsh has more related to the firm's reputation than ForeignSh in previous tests. Moreover, I include ForeignSh in this investigation, but its coefficients are not significant, so I report only the results using FMsh. In particular, as Flavin and O'Connor (2017) document that mature firms with greater growth opportunities (higher market-to-book ratios) in Korea tend to pay larger dividends to build a reputation, I also add the interaction term between the reputation dummies, size and the market-to-book ratio. Panels A and B in Table 11 present the regression results with RPT5 and RPT3 as reputation dummies; t-statistics are based on robust standard errors clustered by firm.

Relationship between dividend reputation, yield and future returns
When I include only RPT Â Log(DY), RPT Â DDY and RPT Â Log(MV) as interaction terms in Columns (1) and (4), the coefficient of RPT is significantly negative, but when other interaction terms are included in Columns (2), (3), (5) and (6), it is no longer significant. Importantly, after including an extensive set of control variables, the interaction terms between the reputation dummy and dividend yield have a significantly positive association with future returns, whereas yields have no relationship with returns, consistent with the results in Tables 9 and 10. However, their significance declines as the reputation building periods shorten from five years to three years. The significantly negative coefficients of RPT Â DDY also confirm the results in Tables 9 and 10. That is, future returns of firms without reputations are likely to increase with changes in dividend yields.
In Columns (1) and (4), larger firms with reputations are likely to produce higher future returns, as shown by RPT Â Log(MV), although the coefficients of Log(MV) indicate the wellknown negative relationship between firm size and returns. Noting firms with reputations tend to be mature; investors positively value mature firms that also build dividend reputations. However, when the interaction term between the reputation dummies, size and the market-tobook ratio, RPT Â Log(MV) Â MB, is included in the regressions, the coefficients of RPT Â Log (MV) become insignificant in Columns (2), (3), (5) and (6). That is, the relatively higher future returns of mature firms with dividend reputations are caused by some other determinant.
The interaction term, RPT Â Log(MV) Â MB, positively affects future returns, which implies that when larger firms with higher growth potential have dividend reputations, future returns are higher. According to the previous section, firms building dividend reputations are likely to be larger and have greater growth opportunities. Moreover, the coefficient estimates reveal that markets also positively react to mature firms with greater growth opportunities that pay non-decreasing dividends as signals.
For ownership, the prior results indicate that firms with more concentrated ownership tend to build dividend reputations. However, the coefficient estimates of RPT Â LargeSh and LargeSh in Table 11 show that when firms with more concentrated ownership have dividend reputations, their future returns tend to be lower, while firms with more concentrated ownership without reputations are likely to have relatively high future returns. In other words, although firms with more concentrated ownership are likely to pay nondecreasing dividend streams (Table 5 and 7), investors may react negatively to these companies. Additionally, Tables 5 and 7 show that firms with foreign holdings of more than 5% tend to build dividend reputations, but the evidence in Table 11   are based on standard errors adjusted for the firm-clustering effect and are in parentheses. *** , ** and * denote statistical significance at the 1, 5 and 10% levels, respectively. The sample period is from 1998 to 2019 Table 11. Relation between reputation, stock returns and dividend yields JDQS 29,1 is no effect of the interaction between foreign holdings and dividend reputation on stock returns.

Conclusion
This study uses Korean stock market data to investigate the determinants of dividendreputation-building for firms and shows that the association between dividend yields and stock returns depends on dividend reputations. The evidence indicates that firms paying higher dividends, larger firms, older firms, more profitable firms, less leveraged firms and firms with less volatile returns are more likely to establish a dividend reputation by not reducing regular cash dividends over at least three, five or seven years. These results are consistent with the literature indicating that mature firms tend to increase dividend payments (Fama and French, 2001;Grullon et al., 2002). Interestingly, the results for growth opportunities measured by sales growth and the market-to-book ratio are mixed. Firms that do not reduce regular dividends over at least three years tend to have higher sales growth and market-to-book ratios, but when I expand the reputation-building period to more than five years, reputation-building firms have relatively lower sales growth and higher marketto-book ratios. The evidence indicates that firms with greater growth opportunities are likely to build dividend reputations, regardless of their past growth. However, firms that pay continuous non-decreasing dividend streams over a longer horizon experience a slowdown in sales growth. For corporate governance, firms with more concentrated ownership establish dividend reputations. This result does not support agency theory for corporate dividend policies (Jensen, 1986) or dividend smoothing (Leary and Michaely, 2011). Also, firms with foreign holdings of more than 5% tend to pay non-decreasing dividend streams. I explore whether the relationship between dividend yields and future stock returns depends on a firm's dividend reputation. The sample is divided into two groups based on whether the firm establishes a dividend reputation and the relationship is investigated using panel regressions and Fama-MacBeth cross-sectional regressions. After including control variables, the evidence indicates that reputation-established firms have a significantly positive relationship between dividend yields and future returns, whereas no-reputation firms show no significant relationship. Similarly, the results from the panel regressions and Fama-MacBeth cross-sectional regressions of the one-year future stock returns on the dividend reputation dummy variable, dividend yield and interaction terms between reputation dummies and some relevant variables, including dividend yields, provide evidence consistent with the prior regressions for each group separately. Although the yield coefficients are not significant, those of the interaction terms between the reputation dummy and yield are significantly positive. This indicates that, among firms with higher yields, only reputation-established firms are expected to produce higher future returns. In particular, when larger firms with higher market-to-book ratios and firms with less concentrated ownership have dividend reputations, future returns are higher. That is, investors positively react when mature firms with higher growth potential and firms with higher agency conflicts build dividend reputations.