The relationship between corporate social responsibility and earnings management : accounting for endogeneity ”

ARTICLE INFO Hyunjung Choi, Byungchul Choi and Jungyoon Byun (2018). The relationship between corporate social responsibility and earnings management: accounting for endogeneity. Investment Management and Financial Innovations, 15(4), 6984. doi:10.21511/imfi.15(4).2018.06 DOI http://dx.doi.org/10.21511/imfi.15(4).2018.06 RELEASED ON Tuesday, 23 October 2018 RECEIVED ON Wednesday, 19 September 2018 ACCEPTED ON Friday, 12 October 2018


INTRODUCTION
Corporate social responsibility (hereafter, CSR) has attracted a great deal of attention in the academia and business world (e.g., Kim, Park, & Wier, 2012;Porter & Kramer, 2006). The International Federation of Accountants has issued a sustainability framework (IFAC, 2011) and different consultation documents (IFAC, 2006a(IFAC, , 2006b to highlight the important roles that professional accountants play in facilitating the sustainable development of their organizations. In this regard, IFAC refers to CSR as sustainability, which originates from accounting transparency. Similarly, the American Institute of Certified Public Accountants (AICPA, 2010) also defines CSR as sustainability and recommends publicly announcing economic viability, social responsibility, and environmental responsibility, as well as corporate financial performance, for a great transparency about all aspects of business.
A number of previous works have examined the association between CSR and accounting transparency. Some researchers argue that CSR reflects a firm's value and culture, which, in turn, affect managers' decisions and reporting behaviors (Aguilera et al., 2007;Kim et al., 2012). The argument relies on the idea that managers of good corporate citizens have less incentive to manipulate accounting numbers through earnings management, since these managers are more likely to pursue ethical expectations by stakeholders rather than achieve their self-interest. However, other researchers argue that managers may use CSR for their self-interest to advance their careers and reputation or to cover up unethical practices such as earnings management. The assumption here is that these managers may want to perform CSR activities in order to avoid the scrutiny from third parties (Prior et al., 2008).
The aim of this paper is to examine whether the results found in previous CSR earnings management relation studies suffer from endogeneity issues of CSR. This study asserts that management's decision to engage in CSR activities is correlated with unobserved firm specific variables such as firm's value and culture and management's self-motivation. Ignoring this self-selection is the likely source of endogeneity bias. Accordingly, this paper revisits the association between CSR and earnings management and shows that endogeneity bias produces inconsistent estimates of the relation between firm's CSR commitment and its accounting transparency.
Using a sample of Korean listed companies between 2002 and 2010 and the Korean Economic Justice Institute (hereafter, KEJI) index as a proxy for conducting CSR activities, this study finds that CSR companies engage in less earnings management through both accruals and real activities manipulations in comparison with non-CSR companies when this study uses ordinary least squares (hereafter, OLS) estimation 1 . Before correcting for endogeneity, the results apparently suggest that the CSR engagement is negatively related to both discretionary accruals and real activities manipulation, consistent with the findings in previous studies. The majority of prior studies (Choi & Moon, 2013;Prior et al., 2008) show that companies with strong commitment to CSR are less likely to manipulate earnings through discretionary accruals. In a recent study, Kim et al. (2012) provide further evidence that CSR engagement tends to reduce real activities manipulation, as well as accruals manipulation.
However, once this study corrects for endogeneity of CSR commitment using the Heckman (1979) procedure and differencing specifications, some of empirical results differ from those in previous studies. The relation between CSR commitment and discretionary accruals reported in prior studies becomes insignificant, while the negative relation between CSR commitment and real activities manipulation remains significant even after controlling for the endogeneity. These results imply that proactive CSR engagement leads managers to behave in a more strategic manner and restrict real activities manipulation rather than constrain accruals manipulation. Considering that real activities manipulation decreases future firm performance due to its direct impact on cash flows, managers of CSR companies may end up restricting real activities manipulation, as it provides more benefits for the companies than accruals management, which does not affect direct cash disbursements.
Prior studies on the CSR-earnings management association use CSR proxies as exogenous variables when CSR engagement may be determined endogenously (e.g., Chih  ous research on the CSR-earnings management association and addresses endogeneity issues. Section 2 describes the methodology used in empirical analysis. Section 3 introduces the data and descriptive statistics. Section 4 reports the empirical results and discusses sensitivity analyses. Final section concludes with implications of findings.   conduct a comprehensive study of the relation between CSR and earnings management, using not only discretionary accruals, but also real activities manipulation and whether the firm has been subject to SEC investigation, as different proxies for earnings management. They provide evidence that CSR plays an important role in constraining earnings management.

Endogeneity concerns
This paper argues that previous results on the CSR-earnings management association might be subject to endogeneity issues. Indeed, prior literature has mainly used CSR proxies as exogenous variables (e.g., Chen  This study identifies and addresses two sources of endogeneity. The first source of endogeneity relates to the consideration that managers make CSR engagement decisions not randomly and, thus, CSR engagement might be determined by managers' overall policies or other internal factors. When the These authors also report that CSR companies display less interest in avoiding earnings losses, but they are still more aggressive in accruals management. However, the results in Chih et al. (2008) should be interpreted with caution since they could be explained by country differences (e.g., different accounting standards, earnings management practices, etc.) rather than differences in CSR activities (Kim et Figure 2 illustrates two endogeneity concerns related to CSR engagement.
In order to address the above two endogeneity issues, this study applies the Heckman 2-stage specification to correct the selection bias caused by CSR engagement decision. Further, this study uses a differencing equation to eliminate the simultaneity problem.

RESEARCH METHODOLOGY
In order to test the hypothesis that earnings management is an increasing or decreasing function of CSR engagement, the next subsection describes an OLS specification in detail. Then, the following subsection presents a selection model and a differencing model to address potential endogeneity concerns. Consequently, this study compares the results after accounting for endogeneity with those based on the OLS regression.

OLS model specification
This study uses two main proxies for earnings management: discretionary accruals and real activities manipulations 4 . Discretionary accruals are the practice of using tricks to misrepresent or reduce transparency of the financial reports without involving the changes of cash flows in the future. 4 Most of prior studies (e.g., Chen  These discretionary accruals can be reversed in the future. On the other hand, real activities manipulations are the management actions that deviate from normal business practice, undertaken with the primary objective of meeting certain earnings thresholds (Roychowdhury, 2006). To avoid earnings disappointments, companies can utilize three real activities manipulation methods: boosting sales volumes temporarily through increased price discounts or more lenient credit terms, which lead to decreasing cash flows, decreasing other operating expenses through reductions in discretionary expenditures, which include advertising expense, research and development, and SG&A expenses, and reducing the reported cost of goods sold through overproduction (Roychowdhury, 2006).
CSR commitment (DCSR) is a main variable of interest. We predict that 1 β in equations (1) and (2) is negative, indicating that the companies with CSR commitment are less likely to manage earnings than the companies without CSR commitment.
Following previous studies, this study includes the following independent variables that could affect financial reporting behavior.

Selection model estimation
One problem with standard OLS estimations in equations (1) and (2) is that they assume that error term ( ) ε is uncorrelated with the explanatory variable of interest (DCSR). A potential concern is that companies that choose to engage in CSR are not a random sample of population. Selection bias arises when an independent variable included in the model is potentially a choice variable, correlated with unobservable relegated to the error term.
When the correlation between ε and DCSR is To control for selection bias, Heckman (1979) proposes a two-stage estimation procedure, commonly known as a treatment effect model when the dependent variable is observed for all observations in the data. In the first stage, a regression for observing a positive outcome of the dependent variable is modeled with a probit model. The estimated parameters are used to calculate the inverse Mill's ratio, which is then included as an additional explanatory variable in the second stage (Lennox et al., 2012). However, to implement Heckman's twostage estimation, we need to find the variables, which are included in the first stage, but are excluded in the second stage. If such variables are available, the estimation using the treatment effect model would yield an unbiased estimator of the coefficient on DCSR.
The choice of CSR commitment in the first stage is estimated as follows:   We also include the level of cash and cash flows from operations to proxy for firm performance, which enables or gives rise to the external demand for CSR investment (Campbell, 2007).
Further, corporate governance is associated with the scope and effectiveness of CSR investment (Kim et al., 2012). Especially in Korea, it is important to take into account the influence of business groups: chaebols 5 . Previous research suggests that chaebols have viewed CSR activities as a way to restore credibility after reputational damage due to accounting fraud and the creation of illegal slush funds ( After running the probity model in the first stage and estimating inverse Mill's ratio from equation (3), this study controls for selection bias by adding inverse Mill's ratio to equations (1) and (2).

Differencing model specification
To rule out the existence of the simultaneity problem, this study implements a fixed effect specification (Garcia-Castro et al., 2010). Firm fixed effects are incorporated into the model by either including a set of firm indicator variables or differencing equations (1) and (2). Accordingly, this study uses the following differencing equations to mitigate the simultaneity problem by focusing on the companies that are ranked in the KEJI index for the first time:  In equations (4) and (5), this study assumes that being first included in the list of companies with the best CSR activities is exogenous. This research design allows us to investigate the change in earnings management following being listed as CSR companies.

Sample selection
This study first collects the information on CSR.

OLS estimation
Previous research has mainly used the OLS estimation for the CSR-earnings management association. Thus, this study first reproduces previous results using pooled cross-sectional OLS to identify in our sample the relation between CSR commitment and earnings management. Table  4 presents the regression results for equations (1) and (2). This study first reports the results using the absolute value of discretionary accruals (ABS_ DA) as the dependent variable in the first column. The coefficient on DCSR is negative (-0.004) and significant (t = -2.35), indicating that discretiontary accruals are significantly lower for CSR firms than non-CSR firms. 8 In an unablated result, this study observes similar results for the regression of ACFO, ADISEXP, and APROD as the dependent variable. The coefficient on DCSR is 0.015 (t = 5.49), 0.007 (t = 3.42), and -0.022 (t = -5.75), respectively. Given that higher levels of abnormal operating cash flows and discretionary expenses and lower levels of abnormal production costs indicate more conservative operating decisions, these results suggest that CSR companies engage in earnings management less by real activities manipulations than other companies.
As a control variable, we include the combined proxy for real activities manipulation (COM_ RAM). The estimated coefficient on COM_RAM is negative and significant; it is -0.001 (t = -3.59).
These results are consistent with Kim et al. (2012), suggesting that firms choosing earnings management through real activities manipulation are less likely to engage in accruals manipulation. Discretionary accruals are also negatively associated with SIZE, AD_INT and AGE, suggesting that larger and older clients spending more advertising expenses are less likely to engage in accrual-based earnings management. Further, discretionary accruals are positively associated with MB, LEV, and EO, indicating that firms with higher growth opportunity, more leverage, and subsequent equity offering are more likely to engage in accrual-based earnings management.
The second column presents the regression results for equation (2), using COM_RAM as the dependent variable. The coefficient on DCSR is also negative (-0.334) and significant ( t= -5.64), indicatcing that combined real earnings management is significantly lower for CSR companies than non-CSR companies 8 . These results support the notion that CSR companies manage their earnings less using real activities manipulation than non-CSR companies and are more transparent in their financial reporting.  Notes: Variables are defined as in Appendix A. *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.
Taken together, these results from OLS estimation in Table 4 indicate that CSR companies engage in earnings management less by manipulating accruals and real operating activities. In other words, CSR commitment is more likely to constrain earnings management and lead companies to make responsible operating decisions, thereby maintaining transparency in financial reporting.  Notes: Variables are defined as in Appendix A. *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

Heckman estimation
As discussed in the methodology section, potential endogeneity of CSR engagement may lead to a bias in DCSR coefficient estimates. To control for selection bias, this study uses a Heckman 2-stage estimation procedure to obtain consistent and efficient estimates for DCSR. Specifically, we model DCSR using probit in the first stage and then estimate IMR. In the second stage, we add IMR to Equations (1) and (2) to control for selection bias. Table 5 shows the second stage result of testing the CSR engagement-earnings management association after including IMR. The IMR coefficients in the first and second columns are positive and significant, suggesting the presence of selection bias.
More importantly, this study finds that some results differ from those in Table 4. For the regressions using ABS_DA and COM_RAM as the dependent variables, the estimated coefficients on DCSR are 0.001 (t = 0.06) and -0.124 (t= -2.14), respectively. After controlling for the selection bias of CSR engagement, the coefficient on DCSR for the ABS_DA regression becomes insignificant, while the coefficient on DCSR for the COM_RAM regression remains significant. Thus, these results suggest that a firm's CSR commitment significantly affects earnings management through real operating activities manipulation, but not through accruals manipulation.
After correcting for endogeneity of CSR activities, the results provide the evidence that CSR commitment leads managers to be more responsible in operational decisions. Discretionary accruals just make the manipulation of book income by managers' decision without considering the changes in cash flows and can be reversed in the future. In contrast, the real activities manipulations are the management actions that deviate from normal business practice and cannot be reversed in the future (Roychowdhury, 2006). Even though these activities enable managers to meet short-run earnings targets, they are likely to decrease firm value in the long run. Consequently, real activities manipulations often sacrifice long-term firm performance and are costlier to the firm than discretionary accruals. Thus, managers of CSR companies tend to restrict real activities manipulation rather than constrain accruals management to provide more benefits for their companies and foster future relationship with stakeholders.

Differencing estimation
In Table 6, this study uses the differencing specification to control for the simultaneity problem of CSR. This differencing specification can also address the concern of reverse causality in the sense that unethical practices such as earnings management affect a firm's decision regarding whether to engage in CSR activities. If changes in a firm's earnings management between the current and prior years are related to its inclusion in the KEJI index made by an independent CSR institution, it is hard to argue that earnings management causes CSR engagement. This study examines changes in earnings management from year t -1 to year t when a firm is ranked for the first time in the KEJI index in year t. Thus, we define an indicator variable RATING for a company first included in the KEJI index during the period from 2002 to 2010. Table 6 presents results for the two differencing models outlined in equations (4) and (5). The results are very similar to those in Table 5. This study finds that the coefficient on RATING is -0.003 (t = -0.94) and -0.205 ( t= -2.00), respectively, using ∆ABS_ DA and ∆COM_RAM as the dependent variable. The results suggest that firms newly ranked in the KEJI index are more likely to decrease real activities manipulations, but not discretionary accruals.
Thus, the findings in Tables 5 and 6 suggest that the proactive involvement in CSR activities leads managers to report accounting information more transparently by restricting real activities manipulations, and not by constraining accruals. The findings indicate that some of the results found in the previous research change when concerns about the potential endogeneity of CSR engagement are addressed in the analysis.

Industry dummy Included Included
Year dummy Included Included Notes: RATING is an indicator variable that takes a value of one if the firm is included in the KEJI index in year t for the first time and zero otherwise and ∆ indicates that variables are measured as the difference between t -1 and t. Other variables are defined as in Appendix A. *, **, *** denote statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

CONCLUSION
This study examines the relationship between CSR engagement and accounting transparency through discretionary accruals and real activities manipulations. Specifically, we examine the role played by endogeneity of CSR commitment in this association for a sample of Korean listed companies. In this paper, we argue that endogeneity problems might plague previous research studying the CSR-earnings management association. While endogeneity problems are not unique to the CSR research, few studies have properly considered endogeneity in CSR-earnings management association. This study contributes to the literature by controlling for the endogeneity problem using the Heckman procedure and differencing specifications.
This study empirically shows that ignoring the endogeneity of CSR activities biases the estimated association between CSR engagement and earnings management. More specifically, this study finds that the negative relation between CSR and discretionary accruals reported in most of the previous studies becomes insignificant when endogeneity of CSR is properly taken into account. However, the prior findings on a negative and significant relation between CSR and real activities manipulation hold even when the endogeneity is corrected. These results provide evidence that CSR commitment is more likely to lead companies to make responsible operating decisions rather than accruals management decisions.
Finally, this study acknowledges the limitation of making generalization for other countries based on the results from Korean listed companies and the KEJI index. Differences in legal, institutional, accounting system, and CSR measurement could lead to differences in the relationship between CSR and earnings management. Due to such differences, it would be interesting to examine the relationship between CSR and earnings management using KLD, a representative measure of CSR, or across countries. In addition, due to changes in accounting standards, the period after 2011 is not analyzed in this study. Since changing accounting standards will have a significant impact on corporate accounting, it will be the subject of further research to study the relationship between CSR and earnings management before and after changes in accounting standards. This study leaves these important questions for future researches.