Do Managers Pay CSR for Private Motivation? A Dividend Tax Cut Case in Korea

A CEO who has an opportunity to pursue his interest may sacrifice investors with inefficient investments such as overinvestment in corporate social responsibility (CSR). As prior researchers have suggested a possibility to detect the perk portion of CSR investment using the dividend tax cut event, we tested whether managers decreased CSR spending while accelerating dividend payouts during the Korean dividend tax cut of 2015. Consistent with the prior studies on the dividend tax cut, we discovered a pattern of incremental dividend increase for the companies of agency conflict measured by extreme CEO ownership. However, we failed to find any statistically significant simultaneous reduction in donations after 2015. This study does not provide evidence that investments in CSR of Korean firms are not due to CEOs’ personal interest-seeking. Instead, we showed that the dividend tax cut event may not work as a universally applicable quasi-experimental setting to detect management overinvestments in CSR.


Introduction
Researchers have tested the relationship between corporate social responsibility (CSR) and firm financial performance and often found a positive relationship [1][2][3][4][5]. If CSR spending improves a firm's future performance, CSR is similar to other capital investments such as machinery purchase or Research and Development (R&D) spending in that a positive net present value is expected. If CSR is a form of investment, some CEOs may have a greater tendency of suffering inefficient investment including overinvestment in CSR. For example, as a slack resource is an important predecessor of CSR [6][7][8], a manager may increase CSR investment as she has more available resources. Just as free cash flow encourages overinvestment [9], investing in CSR can also lead to overinvestment as a CEO has more discretionary resources. Prior studies have suggested that monitoring devices such as debt [10][11][12] or activists [9] reduce the agency costs of free cash flow, implying that some forms of governance structures influence management discretions on investment.
Investments in CSR have both value-creating and value-destroying effects, depending on firms' agency conflicts [13]. Recent studies on CSR have supported not only the value-creating effects of CSR [14][15][16][17][18][19], but also the possibility of an agency problem in which managers overinvest in CSR for personal interests such as reputation building [20][21][22][23][24][25]. For example, Masulis and Reza [24] reported a value-destroying effect of donations and it was greater with poor monitoring. Under the agency theory, management pursues personal goals when it is possible [10,26,27]; therefore, a manager may misuse CSR spending as a vehicle to gain private reputation and social capital at the expense of shareholder wealth if she can freely use her discretionary power.
To show that CSR contains a portion of overinvestment motivated by a manager's self-interests, Cheng et al. [28] adopted an approach of putting their tests on a quasi-experimental setup of the US dividend tax cut in 2003. Studies have shown that as insider ownership increases, dividend tax cut facilitates dividend payouts [24,[29][30][31][32][33] while simultaneously reducing CSR [24,28]. Billings et al. [34] showed that the 2003 dividend tax cut resulted in a decrease in R&D spending, especially for companies in low R&D intensity industries, which implies that there might be some unnecessary investments that were reduced after the tax cut.
The idea of Cheng et al. [28] is that after a dividend tax cut, managers with higher ownership may stop their perk projects such as unnecessary investments and shift the resources to dividend payout because this way they can maximize private benefits. If CSR is reduced after the tax cut while the company increases dividend payments, the marginal spending in CSR canceled after tax cut can be interpreted as a management perk. Cheng et al. [28] reported that for Standard and Poor's 500 index (S&P 500) firms, CSR scores decreased while dividend increased after the tax cut, and this finding was supported by Masulis and Reza [24] who discovered a decrease in corporate giving after 2003 for the high-managerial-ownership firms.
In this paper, we tested whether dividend and corporate giving after the Korean dividend tax cut started in 2015 showed the simultaneous movements as reported in the prior studies. One year before 2015, the Korean government announced a dividend tax cut for qualifying firms, which gives us a natural experimental environment similar to the US case. As many prior studies on the dividend tax cuts of the US and some other countries have reported patterns of increase in dividend payout [30,32,33,[35][36][37], we expect an increase in dividend after the Korean dividend tax cut in 2015. We also tested whether CSR investment was reduced after the tax cut as reported in the US case studies [24,28]. The simultaneous increase in dividend and reduction in donations will support a hypothesis that claims the possibility that some of the investments in CSR of Korean firms are due to the managers' private interests, especially when the management has a higher level of discretionary power or resources to manipulate. Following Nam et al. [29], Chetty and Saez [30], and Brown et al. [31] for the US case and Wang and Guo [33] for the China tax cut, we considered management inner ownership as a sign of agency conflict. We also include bank debt ratio into our consideration following Barnea and Rubin [20] and Krüger [13] because debt can control the agency problem of free cash flow [10].
We applied ordinary least square (OLS) regressions to test our research question. Test results are as follows. First, we found an increase in dividend payout and a decrease in donations after 2015 among the qualifiers of the Korean dividend tax cut; however, the changes were not statistically significant. Second, test results showed that for the firms with an agency problem in terms of management ownership, the change in dividend was statistically significant. However, we failed to find any statistically significant test results for donations even for companies with agency problem.
To confirm the robustness of our main findings, we employed fixed and random effect models and found test results that partially support the dividend increase for both extreme CEO ownership and extreme bank debt ratio groups. However, when it comes to donations, we could not find any statistically significant results that support our hypothesis. When we limited our data to a subgroup that increased dividend after 2015, we did not find any significant reductions of donations. When we limit our data to the companies with declining donations after 2015, we also failed to find any significant change in dividend. The results imply that simultaneous changes in dividend and donations were not the common behavior of CEOs around the 2015 tax cut in Korea.
We replaced donations with a CSR report issuance dummy variable and the ESG scores and found a decrease in ESG scores after 2015, but no significant results were found for the companies of agency conflicts. Finally, when we employed alternative governance measures instead of CEO ownership and bank debt ratio, we found some partial results that showed the decrease in donations; however, in most models, we could not find any significant influences of governance on the simultaneous movements of dividend and donations.
As we failed to find consistent and complete evidence of management behaviors following our predictions, we conclude that the idea of Cheng et al. [28] and Masulis and Reza [24] is not suitable for the 2015 Korean dividend tax cut case. We suggest that the quasi-experimental design of dividend tax cut may not be universally applicable to prove managerial pursuits of self-interests via CSR overinvestment.

CSR as Investment
Studies on CSR have a long history of discussions, from the normative view of Bowen [38] to the stakeholder theory of Freeman [39] and finally to the creating shared value (CSV) of Porter and Kramer [40], with a gradual increase in the emphasis on the strategic view that focuses on what a company can get from the social involvements. From Carroll [41], the economic consequences have become one of the basic considerations of CSR studies. As a result, CSR researchers in recent decades have focused on the economic impacts of CSR mostly for the bigger organizations [42], but yet to reach a universal conclusion. For example, Margolis and Walsh [2] showed that less than half (54) of the 127 studies from 1972 to 2002 had reported some positive CSR-performance relationships.
This approach of CSR, often called "doing well by doing good," assumes that CSR may give a positive net present value to the firm. In this regard, some strategic investments in CSR are equivalent to corporate investments such as R&D by achieving competitive advantage through innovation or product differentiation [44][45][46]. Husted [47] suggested that CSR is a form of a real option in that by such an investment, a corporation can call benefits in the future when it provides a favorable outcome.
As it is an investment similar to other investments, we may infer two characteristics of CSR. First, investments in CSR will be limited by its resource availability as long as it is an activity that physically consumes corporate resources. Waddock and Graves [6] showed that financial performance has a positive impact on the next period's CSR. Other empirical studies on the determinants of CSR also support the idea that economic resource allows organizational slack to prepare CSR [7,8,[48][49][50]. The resource-based view explains the valuable, rare, and nonreplicable resources as the core of sustainable competitive advantage [51][52][53][54], and CSR activities are considered as one of the resources [55]. The interdependence between CSR and performance may create an endogeneity issue, which means that we have to control a firm's simultaneous performance carefully [56][57][58].
Second, just as not all R&D investments have positive returns, only some investments in CSR will increase shareholder value. The free cash flow hypothesis under the agency cost model [10,27] suggests that, as there are more free resources available and as a CEO has discretionary access to the resource, investments in the projects of low Net Present Value (NPV) increase. Richardson [9] showed that over investment was severe for firms of high free cash flow and the monitoring mechanism mitigated the relationship. Some studies have supported this finding by theoretically or empirically showing the negative relation between governance and overinvestment of free cash flow [59][60][61]. Officer [62] reported that firms with poor investment opportunities and high free cash flows had higher dividend announcement returns, implying that the market buys the decreased possibility of management overinvestment. Higher reporting quality also decreases overinvestment because reporting quality reduces agency costs [63].

CSR Overinvestment and Dividend Tax Cut
If CSR is a form of investment, as not all investment can have positive returns, we conjecture that the basic idea of the free cash flow hypothesis can apply to CSR. Managers with enough free resources and discretionary power may overinvest in CSR as well as other projects, resulting in poor returns. Some studies provided test results that can support "the agency cost view" of CSR as they discovered some negative or neutral relations between CSR and firm value depending on their conditions applied to each research [24,[64][65][66][67]. More importantly, there have been a group of studies that implied the prevalence of CSR overinvestment as good corporate governance was negatively related to CSR [20][21][22][23]25]. In a study by Jiraporn and Chintrakarn [22], as managerial power grows CSR investment increased, but after a certain point, powerful managers did not consider additional investments in CSR, a result consistent with the entrenched management view. Breuer et al.'s findings [25] were similar except that they showed a linearly positive relationship between power and CSR, and that the institutional discretion intensified the effect. Borghesi et al. [23] reported that institutional ownership decreased CSR investment.
Some studies used the US dividend tax cut event in 2003 to discover the proof of managerial self-interest seeking via CSR overinvestment. Although there are conflicting results on whether investment increased as intended after the US dividend tax cut [34,68] or even whether dividend had increased or not [30,35,69,70], studies have generally agreed that companies with higher managerial ownership issued more dividends after the tax cut [24,[29][30][31][32][33]. However, few studies have focused on the simultaneous change in CSR, except Cheng et al. [28] and Masulis and Reza [24].
The basic logic of this test is that, as R&D spending had fallen as a result of the dividend increase after the tax cut [34], CSR costs might also be sacrificed, which suggests that the reduced amounts, if any, might not be the necessary investments of positive returns. If it can be canceled at any time to convert the resource into other investment choices and the alternative is selected because it better serves managerial self-interest, then there is a strong possibility of overinvestment due mainly to the agency problem. Traditional donation models provided two explanations for its motivations: Profit maximization or management utility maximization [71,72]. The latter implies that excess spending in contribution can sacrifice firm profits while maximizing management utility. Boatsman and Gupta [73] supported this view by showing a negative relationship between marginal tax rate and donations.
Therefore, we need to check two points: First, the increase in dividend and a simultaneous decrease in CSR spending after the tax cut. The reduction in CSR spending is due to reasons other than revenue decline, as companies with bad earnings are less likely to increase dividends [24]. Second, it must happen in a more pronounced way where the agency conflict is severe because it is the premise of this argument that managers pursue their interest when they are free from owners' monitoring. Only if both conditions met, we might conjecture that the canceled portion of the CSR investment reveals the overinvestment by managerial preferences.
Cheng et al. [28] were one of the first researchers to provide this approach by testing the value-destroying activities of management, which are seemingly altruistic, through the natural experimental setting of the dividend tax cut. In their data, KLD Research and Analytics' scores (KLD scores) were decreased after the tax cut, and the drop was greater among companies where insider ownership is medium. Masulis and Reza [24] pointed out that because of the difficulty of figuring out the exact amount of returns from a CSR investment, a CEO has room for arbitrary investments in the direction of his preferences. They discovered that many donations are made to the institutions that have special relationships with the CEOs. They also found a $1 million reduction of corporate giving on average following a $6.4 million increase in dividend payout after the dividend tax cut in 2003. Finally, they showed that this tendency became severe as CEO ownership rises.

Hypothesis
Following the suggestions of Cheng et al. [28] and Masulis and Reza [24], our first research question was whether the Korean dividend tax cut in 2015 affected the behaviors of management dividend payout and charitable giving as reported in the US case. Some studies on the US dividend tax cut in 2003 have reported an increase in dividend [30,32,35,37]. Chetty and Saez [30] reported that dividend increased by 20% after the tax cut. However, according to Brav et al. [35], this increase was temporary because of the increase in share repurchase following the dividend payout. Brav et al. [35] and Brav et al. [69] also showed that the tax cut event was not the primary factor in managers' decision on dividend payout. Moreover, Edgerton [70] even argued that, considering the profit increase around the tax cut and the repurchase following dividend increase, companies did not actually pay more dividends because of the tax event.
Studies on the dividend tax cuts of other regimes have also reported not always consistent results on the dividend increase [33,36,74]. Kari et al.'s [36] The Finnish dividend tax cut in 2005 and Wang and Guo's [33] Chinese dividend tax cut in 2005 showed a significant increase in dividend payout. However, Bird's [74] study on the dividend tax reform of 2006 in Canada found little effect on dividend increase. Generally speaking, as relatively more studies including the US and some other countries showed a general pattern of increase in the dividend, we expect that dividends will increase after the Korean dividend tax cut in 2015. Based on the two US papers, which showed a decrease in CSR, we expect that the Korean tax cut also leads to a reduction in CSR investment.

Hypothesis 1.
After the Korean dividend tax cut in 2015, dividend payout increased and donations decreased.
Because we use dividend tax cut event as a tool to detect a portion of ineffective managerial investments in CSR, mainly due to motivations other than shareholder value maximization, we predict that this phenomenon will be better observed in agency conflicts. The first condition widely considered is management ownership. Nam et al. [29] reported that the CEO and other top management ownerships intensified the likelihood and level of dividends increase after the 2003 tax cut. In Chetty and Saez [30]'s study, companies with the highest executive ownership showed the most frequent issuance of new dividends after the 2003 tax cut. They explained that the benefit of decreased tax burden after the tax cut motivated managers with large stocks of their company to initiate dividend payout. Brown et al.'s [31] paper also showed a positive relationship between the top 5 executive ownership and the likelihood of dividend increase after the 2003 tax cut. Other studies such as Blouin et al. [32], Wang and Guo [33], and Masulis and Reza [24] discovered similar results. Therefore, we used management ownership to identify a subsample of the agency problem.
The second condition we considered is debt. Jensen's [10] agency model is about the managerial tendency not to distribute earnings to shareholders to grow his company bigger than its optimum. He suggested that by owning debt instead of issuing stock, free cash flows have to be paid in the future; therefore, managers will stop investing in negative NPV projects. Following his control hypothesis of debt, many studies also considered debt as a governance device [11,12,59,60]. To investigate management overinvestment in CSR, Barnea and Rubin [20] employed debt ratio as well as management ownership to find where the agency cost matters. Krüger [13] used leverage and liquidity because high leverage and low liquidity hinder managers to invest in negative NPV projects. Following prior studies, we selected debt as our second condition of agency conflict.

Hypothesis 2.
The effect of the dividend tax cut in 2015 was greater for firms of agency problem measured by inside ownership and debt.

Model and Data
We tested the year effect of dividend payout and donations based on the following models.
The models show the incremental changes in the level of dependent variables after the introduction of the dividend tax cut, with controls of the factors influencing dependent variables. We expect that for dividend model β 1 of the first model or β 3 of the second model have a positive value and for donations model, β 1 of the first model or β 3 of the second model show a negative sign. Unlike some prior studies [20,23,25,28] we did not use CSR ratings in our main tests because, first, CSR ratings indirectly represent investments in CSR, and second, greenwashing may distort the image of the firm and result in higher ratings [75]. Studies have shown inconsistent results on the CSR ratings' predictability of CSR activities [76,77]. Instead, we used donations because it is more directly related to CSR spending. The two moderating variables for the agency costs are the CEO ownership and debt-to-asset ratio.
Because dividend and donations are possibly the mechanical results of firm characteristics such as size and profitability, we included control variables based on previous studies. We collected variables regardless of the determinants of which. Firm size (size), market capitalization (cap), and profitability (roa) are the primary determinants of dividend [78,79] and donations [49,71,80]. We used the number of employees (emp) as more labor-intensive firms donate more [81]. As roa was not statistically significant in our preliminary tests, we additionally employed operating cash flows (ocf ) because excess cash is an important factor influencing dividend [82,83]. We included change in sales revenue (csales), market-to-book ratio (mb), and R&D expenditures (rnd) because not only size and profitability, but also growth opportunity influences dividend policy [84]. As Navarro [72] and Adams and Hardwick [49] showed a negative relationship between debt holdings and donations, we controlled leverage. We included advertising intensity (adv) because studies have suggested a positive relationship between advertising and CSR [72,[85][86][87]. Foreign ownership (for) was included in our model because dividend is favored by institutional investors, which bear lower taxes [88], and in the Korean stock market, foreign investors, who are mostly institutional investors, have such influences on dividend policy [89]. Bartkus et al. [90] showed that firms with higher institutional investors had a lower level of donations. Volume turnover rate (volume) was selected because market uncertainty influences dividend policy [91,92]. As Navarro [72] and Brammer and Millington [81] implied some systemic relations between dividend and donations, we controlled donations on dividend model and dividend on donations model. Finally, we controlled industries as Useem [93] suggested that corporate giving depends on industries. Our test models were verified mostly by OLS regression analysis except for some robustness tests, which employed fixed and random effect models. We used SAS 9.4 and Stata 13 together to perform all the processes of our empirical tests.
As the 2015 dividend tax cut had eligibility requirements (the conditions are as follows: (1) (a) Combined payout ratio for the recent three years, which is 1.2 times greater than the three-year market average; (b) combined dividend yield for the recent three years, which is 1.2 times greater than the three-year market average; (c) dividend growth 1.3 times than the greater of last year or the last three average; (2) (d) combined payout ratio for the recent three years, which is 0.5 times greater than the three-year market average; (e) combined dividend yield for the recent three years, which is 0.5 times greater than the three-year market average; (f) dividend growth 1.3 times than the greater of last year or the last three average; all companies must meet one of the two conditions, (a) + (b) + (c) or (d) + (e) + (f)), we classified firms depending on whether the condition is met and conducted our research based on the qualifiers. Data include all public companies in the Korean stock markets between 2013 and 2016. Mainly due to the missing data on the two dependent variables, the full sample size was reduced to 2553 firm-years and the qualifiers were 727 firm-years. All variables are winsorized by 1% to control outliers.

Descriptive Statistics
Descriptive statistics for the variables are reported in Table 1. Following Amato and Amato [94], we standardized our dependent variables with sales revenue, to reflect the possible correlation between firm performance and donations suggested by the slack resource theory. We separately tested the unscaled version for our regression models and did not report the results because test results were the same as those of our scaled version. The mean values of the two dependent variables, dividend and donations, are 0.017 times of sales and 0.001 times of sales. Some variables including size, cap, and emp are logarithmized values to control size, skewness, and kurtosis. Variables such as lev, mb, rnd, adv, roa, proa, ocf, and volume are the scaled values by total asset, book value, sales, or the number of issued stocks; calses are the percentage of change value; therefore, we did not modify such variables except outlier control. Qualifier, CEO ownership, and bank debt ratio are used to classify companies. The latter two are used to specify the subsamples of agency conflicts. The mean values show that the CEOs own 11% of the shares and 19% of the company assets comes from the banks. Independence, frequency, diversity, CEO duality, and CEO-owner are employed to replace our main governance variables CEO ownership and bank debt ratio. The mean percentage of independent directors on the boards is 40.7%. The average number of board meetings is 12.65 and the logged mean value used in this study is 2.336. Diversity shows that during our test period, only 1.7% of Korean listed companies employed female registered directors, and an unreported test shows that the percentage will not be improved when we limit our data to some big companies in size. In the sample, more than 70% of the CEOs have board chairmanship and more than half are owner-managers. The mean values of dividend increase dummy and donations decrease dummy show that almost half of companies have increased dividends and decreased donations since 2015. However, simultaneous change dummy shows that only 20% of companies experienced two phenomena simultaneously. The unreported tests show that for qualifiers the statistics are not quite different: For qualifiers, 57% increased dividend, 44% decreased donations, and 23% simultaneously increased dividend and decreased donations after 2015.
We did not report our correlation matrix table due to the space limit. The unreported test results show that dividend and donations are positively correlated with each other, and they are positively correlated with firm profitability and cash availability (roa, proa, ocf ), which implies that firm resource availability is the prerequisite of the two dependent variables. They are positively correlated with cap, mb, rnd, adv, and for, and negatively related to lev and volume. However, for some other variables, two dependent variables are reacting differently. Companies with bigger assets and more employees tend to provide fewer dividends per sales while they donate more. As size and emp are positively related to for and bank debt ratio and negatively related to CEO ownership, we may confer that better monitoring reduces the necessity of the immediate distribution of firm earnings.
The unreported correlation matrix table also shows that CEO ownership is positively related to dividend and negatively related to donations. This raises the possibility that higher CEO ownership causes an increase in dividends and a decrease in donations. For the bank debt, bank debt ratio is negatively related to dividend and positively related to donations. However, only the relationships to donations are statistically significant.

Multivariable Regression Analysis
Test results for hypotheses are in Table 2. Panel A of Table 2 reports test results of dividend and donations change after 2015 and the incremental effects of high and low CEO ownership. As an unreported test shows a non-linearity between CEO ownership and dividend/donations, we included both highest and lowest ownership because the low managerial ownership is related to the agent-principal issue [26] and the high ownership causes tunneling [95]. Due to space availability, we did not report control variables.

Dividend Dividend Donations Donations Dividend Dividend Donations Donations
Year 2015  Unlike the expectation of our hypothesis 1, Panel A columns 1 and 3 show that the qualifiers' dividend and donation change are not statistically significant. However, in column 2, the interaction variable between 2015 and management ownership is statistically significant and consistent with the hypothesis 2, which suggests that only the qualifier with an agency problem increased dividends after the tax cut event, consistent with prior studies [24,[29][30][31][32][33]. Column 5 to 8 show that for the non-qualifier, the coefficients of interaction terms are negative and statistically insignificant. Panel B column 2 also shows that for the lowest debt holders, the incremental effects are also statistically significant, supporting the control hypothesis of Jensen [10]. In general, our test results show that for the subgroup of agency conflicts, the tax event significantly affected the dividend policy of qualifiers. However, Panel A and B of Table 2 do not provide any significant effects on donations. Korean companies did not reduce donations after the tax cut in 2015, implying that the CSR behaviors of Korean CEOs are not the same as Cheng et al. [28] and Masulis and Reza's suggestions [24]. A graphical presentation for the main test results is provided in the Appendix B.

Alternative Tests
In this section, we checked the robustness of our main test results with different conditions. First, we employed fixed and random effect models instead of pooled OLS used in Table 2. Test results in Table 3 column 2 show that only dividend models provide the statistically significant incremental effects that are consistent with the theory. For donations, we found a significant positive result in column 7 of which the direction of the coefficient is the opposite of our expectation. Therefore, fixed or random effect models support the OLS model by confirming that only dividend changed according to hypothesis 2. Unreported test results showed no significant results for the non-qualifiers. For the random effect models, Wald Chi-square values are reported instead of F values.
Next, we tested our model on each subsample of dividend increased and donations decreased after 2015 to check the moderating role of governance differently. Column 1 in Table 4 shows that the increase in dividend after 2015 is concentrated on the highest and lowest deciles of CEO ownership. Unlike qualifiers, non-qualifiers in column 3 have a significant and negative intersection. Column 2 shows that companies that increased dividend after 2015 did not decrease donations, consistent with our main tests. Similarly, Column 5 shows that companies with decreased donations after 2015 did not increase dividend. Therefore, our test idea of the simultaneous change of dividend and donations is not supported. We omitted test results for bank debts because no significant results were found. The unreported tests for the simultaneously changed subgroup showed that in most models, intersection variables are not statistically significant except qualifiers' dividend model moderated by CEO ownership, which shows a positive and significant effect at 10%. Note: t-statistics in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1. Variables are defined in the Appendix A. Note: t-statistics in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1. Variables are defined in the Appendix A.
Thirdly, because donations may cover only part of the total CSR activities, instead of donations we employed the issuance of sustainability reports dummy and the ESG scores provided by the Korea Corporate Governance Service. We believe that donations can more faithfully represent our test idea, investment in CSR, since it quantifies the input of CSR activities in terms of expenditure. Instead, as a CSR rating is based on the evaluation, it may reflect stakeholders' perception of a company's CSR activities, which is the output of a firm's CSR activities. Moreover, unlike donations account, which is reported by all related listed companies through financial disclosure, companies with CSR ratings or sustainability reports are generally large and highly visible, which creates a biased subsample. However, donations may not cover all aspects of CSR activities, as long as CSR is defined as the total responses to stakeholders throughout the company. Therefore, we adopted donation as our primary test variable and used the index and the issuance of sustainability report for our additional test. We did not report the test results because no significant results in the intersections were found. The test results were consistent with our other tests which used donations.
Finally, we replaced our main moderators (CEO ownership and bank debt ratio) with governance measures for board and CEO characteristics: Board independence and frequency, CEO duality, CEO-owner, and diversity. We used the test model that inversely relates corporate governance and the 2015 effect as below: Alternative governance measures = α + β 1 change dummies + β 2 size + β 3 cap + β 4 emp + β 5 lev + β 6 csales + β 7 mb + β 8 rnd + β 9 adv + β 10 roa + β 11 proa + β 12 ocf + β 13 for + β 14 volume + ε In this model, the change dummies are the variables that give 1 when dividend increased, or donations decreased, or two events occurred simultaneously after 2015. This model does not test whether the average changes surrounding the tax event were statistically significant. Instead, it tests whether those increased or decreased companies were different in terms of corporate governance. We allowed a different sample size for each test, as each governance variable has different numbers of missing values. Table 5 tests whether the level each governance variable of the changed group is different from that of the unchanged group. Test results show that only one model (column 1) for dividend increase, two models (column 8 and 14) for donations decrease, and one model (column 21) for a simultaneous increase in dividend and decrease in donations have the test results consistent with our hypothesis. As the unreported test results for the non-qualifiers contained 12 significant results out of 21 models, we may conjecture that there is a possibility that CEOs may act in line with the hypothesis; however, the reason is not because of the dividend tax cut as we suggested, because the only qualifiers are the companies subject to change their behaviors due to the tax cuts.
We did not report the other model that changes the position of dependent and independent variables because we found similar test results in that only one model (CEO ownership) provides statistically significant simultaneous change. Note: t-statistics in parentheses *** p < 0.01, ** p < 0.05, * p < 0.1. Variables are defined in the Appendix A.

Conclusion and Discussion
We tested the possibility of detecting management perk via CSR investment using the natural experimental setup of the dividend tax cut following Cheng et al. [28] and Masulis and Reza [24]. As prior studies on the US case generally reported an increase in dividend in a certain CEO ownership environment, and as Cheng et al. [28] and Masulis and Reza [24] reported a decrease in CSR after the US dividend tax cut in 2003, we also expected similar management behaviors after the Korean dividend tax cut in 2015. We employed CEO ownership and bank debt ratio to detect weaker governance where the CEO has discretionary power and available slack resources to fulfill his interests at the cost of investment efficiency. Test results provide evidence of a dividend increase for the highest and lowest CEO ownership subgroups; however, we failed to find consistent results for simultaneous donations decrease. We concluded that for the Korean dividend tax cut case, CEOs who can achieve benefits from the tax cut did not transfer their investments in CSR to the dividend payout, unlike Cheng et al. [28] and Masulis and Reza's [24] the US study. As studies on dividend tax cut around the world have not provided consistent results, we believe that as a detecting tool of management overinvestment in CSR, the dividend tax cut event may not be universally useful.
One possible reason for the lack of consistent statistically significant changes in donations in this study is cultural or institutional. Utz [96] explained that companies in the Asia-Pacific regions tend to overinvest in CSR, unlike companies in the US, Europe, and Japan. His finding is that the higher the CSR of Asian companies, the greater the risk of the stock price crash, possibly due to the lower corporate governance. Lee [97] analyzed that even each Asian country has different inner and outer pressures, which makes the characteristics of CSR activities of Korea, Japan, China, and Taiwan different from each other. According to Lee [97], Korea has the highest tendency to keep the level of CSR investment, mainly because of higher pressure from both inside and outside. Therefore, it is possible that Korean CEOs have higher tendency to keep their current level of overinvestment in CSR because of the governance problem or some other cultural reasons such as maintaining social prestige, which is more critical in the East Asian culture.
However, we believe that it is still possible to apply universal theories and models to regional data because nowadays countries encounter similar inner and outer pressure due to the globalization of capital markets and industries, which reduces differences in each region and increases commonality. Moreover, considering that the evidence from the literature has shown inconsistent conclusions for the effect of dividend tax cut not only globally but also within the US, we may suggest a possibility that the difference is not from the locality but from the general limitation of the research setting.
This study contributes to the literature in that it is the first study on non-US companies that tests the management perks in terms of CSR investment using the natural experimental settings of the dividend tax cut. Prior studies have investigated the effect of dividend tax cut on dividend policy [29][30][31]35,69,70] or investment policy [34,68]. Among the literature, only a few studies tested data outside the US [33,36,74]. Moreover, no studies have focused on the effect on CSR policy outside the US. This study may add a doubtful view on the possibility of the dividend tax cut as a universal tool to detect managerial overinvestment in CSR, not because CEOs are altruistic, but because a dividend tax cut may not be the primary thing that managers consider in their decision making [35,69].
The limitations of this study are as follows. First, we did not consider the year 2017 for our reported tests because from late 2016 to 2017 the side effects of this tax cut was widely criticized in politics to make the 2017 tax benefit smaller and to completely abolish after 2018, which may change the effects of tax cut in 2017 incomparably with the former two years. As a result, the data for our main tests cover the two years before and two years after the tax cut. Unreported tests that included the year 2017 did not produce any significant results that support our hypothesis. Second, as this is not a study on the cultural or institutional differences among the tax regimes, we did not empirically explore the reasons for this difference. Further studies may be required to discover the origin of this variance in tax elasticity of corporate behaviors. Appendix A. Variables Definitions dividend = cash dividend divided by sales; donations = donations divided by sales; y2015 = a dummy variable 1 for the firm-years after 2015, 0 otherwise; size = natural log of total asset; cap = natural log of market capitalization; emp = natural log of number of employees; lev = leverage, total liability divided by total asset; csales = change in sales revenue; mb = market to book ratio; rnd = R&D dividend by total assets; adv = advertising intensity, advertising expense divided by total asset; roa = return on assets; proa = last period's return of assets; ocf = operating cash flow divided by total assets; for = foreign ownership; volume = volume turnover rate, year-end day transaction volume multiplied by 300 and divided by the number of issued stock; qualifier = a dummy variable 1 for the qualifier of dividend tax cut; CEO ownership = common stock shares owned by CEO dividend by total issued shares; bank debt ratio = debts owed by banks divided by total assets; independence = the ratio of outside directors on the board; frequency = natural log of the number of board meetings; diversity = 1 if a firm has female registered directors; CEO duality dummy = 1 if CEO is the chairman of the board; CEO-owner dummy = 1 if CEO is the largest shareholder or a related party of the largest shareholder; dividend increase dummy = 1 if dividend was increased after 2015; donations decrease dummy = 1 if donations were decreased after 2015; simultaneous change dummy = 1 if dividend was increased and donations was decreased after 2015;

Appendix B. Graphical Presentation
The following graphical presentation summarizes our main test results. All additional tests are basically the robustness checks for the relationships displayed below. We found an increase in dividend payout for the qualifiers of the tax cut plus for the managers of extreme ownership or debt holding as expected; however, we failed to find the significant and consistent results for the decrease in donations after the tax cut. Our conclusion is that Korean managers did not simultaneously increase dividend and decrease donations, which denies the possibility that they increased dividend with the money they have gained by reducing CSR spending.