do not necessarily reflect the views of UKDiss.com.
The association between Corporate Social Responsibility (CSR) activities And Earnings Quality: Evidence from Mozambican extractive industry
The purpose of this article is to examine the relationship between Corporate Social Responsibility (CSR) and Earnings Quality, by using 368 firm-year observations covering the 2010-2017 period. We provide strong evidence that Mozambican extractive industry are using CSR activity to engage in earnings management (poorer earnings quality). Furthermore, we find that both positive and negative CSR activity reduce earnings quality. Moreover, the influence of positive CSR indicator is much stronger than that negative CSR indicator in reducing earnings quality. These findings are consistent with the idea that the opportunistic managers use CSR activity to reach their particular interest. Our results are robust to alternative proxy measures of CSR activity and earnings quality.
Keywords: Corporate Social Responsibility (CSR); Earnings Quality (EQ); Mozambican extractive industry.
The purpose of this study is to examine the relationship between CSR and earnings quality of Mozambican extractive industry. Specifically, the study examines whether socially responsible firms behavior differ from other firms in their financial reporting.
According to Hoffman , CSR is observed as how a company takes account of its environmental and social effects of its operations. For Choi et al.  CSR can be observed as key aspect in the successful and survival of a company. Companies may demonstrate ethical commitment via philanthropic contributions and via using resources and expertise to benefit society, for instance caring for the environment, reducing waste and employing minorities (Hoffman, and Moser) [15,26]. Following (Castelo, M., Lima, L ) we only consider the ethics and moral aspect about CSR 0activity such as the satisfaction of human resource, the relationship with the local community, environmental protection and the product. In this perspective, (Chun, R. Ethical )  argue that Ethical and moral companies will revel integrity by being socially responsible, sincere, honest and trustworthy. For (Atkins)  the ethics and moral aspect of CSR activity consist of the firms to be transparent in financial reporting. (Bozzolan ) , argue that the society requires that the actions of public entities be sustainable, efficient and transparent in its management. Socially responsible firms are likely to constrain earnings management, reduce the financial risk and win the corporate reputation providing to investors more transparent and reliable financial information according to the ethical and moral hypothesis.( Lev, B., ) , argue that firms may conduct CSR activities when pretends to increase sales and promote a good reputation in society. For (Pyo G., ) , the managers can have two motivations to practice SCR activity (the opportunistic or integrity). For the opportunistic motivations companies who engage in voluntary CSR actions may have lower earnings` quality. On the other hand, for the integrity motivations companies who engage in voluntary CSR disclosure can have an exemplary conduct providing ethic on the financial statement.
(Zang, ) , argue that CSR-behaved firms care about their capability to deliver more value to stakeholders, they might be less likely to engage in earnings management activity.
In this study, we shed light on the unexplored relationship between CSR activity disclosure and earnings quality. In particular, we address two research questions: (a) Does the link between CSR activity disclosure and earnings quality differ for firms with CSR practice? And, (b) how does CSR activity affect earnings quality? These questions improve our understanding of how CSR policy affects earnings quality. In the literature, we can find several theories that explains the link between CSR activity disclosure and EQ (ethical theory, agency theory or neo-classical theory, legitimacy theory, “trade-off theory”, Supply and demand theory, Stakeholder theory, and others). In this study, we applied ethical theory to solve the research questions, because in our point of view this theory better explains the link between these two variables.
While few prior study has documented a positive (negative) impact of CSR on earning quality (e.g., 30 Prior et al., 2008; 08 Chih et al., 2008; 16 Hong & Andersen, 2011; 39 Yip et al., 2011; 20 Kim et al., 2012; 35 Scholtens & Kang (2013); 33 Salewski and Zulch, 2014; 25 Litt, et al., 2014; 03 Bozzolan et al., 2015, 13 Gil et al., 2016) 11, 12, 13, 14, 15, 16, 17, 18 , it has never considered that both positive and negative indicator of CSR activity appear to provide information that reduces earning management activity and increase earnings quality. The interpretation of positive (negative) CSR indicator can be different for each investor thus, to avoid losing useful information about CSR activity we disggregate the CSR activity index into strengths (positive) and concerns (negative) categories and we adjust industry and year of effects.
In the nutshell, all these prior studies employ the U.S. data, the European data and the Asian data, particularly for developed economies and emerging country. But evidence from under-developed economies remains rare, while our study explores a new setting based on under-developed country.
Our empirical analyses consists all of the extractive firms operating in Mozambique according to the National Institute of Statistics – INE Mozambique 19, from year 2010 to 2017. Our regression results show that CSR activity disclosure is significantly and positively associated with earnings management, implying that firms with a CSR policy are more likely to be involved in earnings management (poorer earnings quality). We further find that both positive and negative indicators of CSR activity contribute to rising earnings management. One important finding is that the influence of positive CSR score is much stronger than that of negative CSR score in rising earnings management.
Moreover, our study tests how four different CSR activities affect earnings quality. We find that the relationship with the local community and the environmental protection are positively related to earnings management but for the satisfaction of human resource and the product are negatively related to earnings management. Our models did not show endogeneity problems, the heteroscedasticity problem, the multi-collinearity problems, even the problems of autocorrelation because we addressed the respective statistics test to solve it. The robust outcomes in this study provide novel evidence about the link between CSR activity and earnings quality.
These results are inconsistent with ethical and moral hypothesis, but consistent with the idea that the opportunistic managers use CSR activity to reach their particular interest. Also, this finding are in contrast with the expectation that CSR policy companies are less likely to engage in the earnings management activity. These results can be explained by the fact that Mozambique is an under-developed country and the pressure from stakeholder is not so strong, suggesting that the degree of development of a country is key fundamental to explain the link between CSR and earning quality.
However, this study differs from the previous studies, addressing the unexplored issue linking the CSR activity and earnings quality. Secondly, our study employs two different measurements of earnings management from one used by (Gil) 18 , more specifically, we apply Dechow et al. (1995) and Kotari et al. (2005) models 20, 21 [11,21]. Thirdly, we use three different measure of CSR score (i.e., the positive and negative indicators of CSR score and the aggregation of both CSR activity). Fourthly, this is a novelty about Mozambican Extractive sector (an example of an “under-developed country”). Fifthly, our study makes a great contribution to the literature, expanding the existing accounting and financial literature on CSR and earnings quality field showing that the ethical and moral behavior can drive the firm philosophy.Finally, the evidence from the study can help the regulators to understand firms’ business practices in under-developed country.
The rest of this paper is organized as follows. Section 2 provides the country background. Section 3 introduces the relevant related literature and develops the hypothesis of the study. Section 4 introduces the data, explains the sample and explains the research design. Section 5 summarizes the empirical results and discussion. Finally, Section 6 concludes.
2. Country background
The Republic of Mozambique is an African country located in the southeastern region of the African continent. Since the beginning of the last decade, Mozambique has been discovering several mineral resources leading the extractive industry to constitute one of the main driving force for Foreign Direct Investment (FDI). 22
Since the year 2000 the Mozambican Government signed several contracts with multinational companies in the areas of mineral resource exploration. One of the contractual terms among these multinational companies was compliance with corporate social responsibility obligations in the areas of human resources, community, environment, production. In 2014, the Mozambican Government approved a Corporate Social Responsibility Policy for the Extractive Industries and Mineral Resources 23 .
However, Mozambican society has raised several questions about the role of large multinational firms in relation to corporate social responsibility (CSR) of these firms, mainly in the socioeconomic and environmental area. On the other hand, few researchers argue that in recent years these multinational firm has been a failure complying with these contracts with regard to CSR in relation to the community and the environment (as illustrated by qualitative studies from CIP, 2012; Langa and Mandlate, 2013; Langa and Massingue, 2014 24, 25, 26 [22,23,44]) We found a convenient time to undertake an empirical study.
3. Relevant Related Literature and Hypotheses
Although there are several definitions of Corporate Social Responsibility (CSR) in the accounting literature, the definition offered by 27  is uncontested and widely accepted. For 27 (p 289) [05, pp 289], socially responsible companies must operate in four distinct areas, such as economic, ethical, legal, and philanthropic. The lower level of definitions requires companies to act with the minimum of business ethics and, on the other hand; the higher level includes sustainability, suggesting that companies should be proactive.
Likewise, the uncontested and widely accepted definition of earnings management offered by Healy & Wahlen (1999) 28 [ 14]. For 28 (p 368) [14 pp 368], earnings management occurs whenever managers use judgment in financial reporting and in structuring transactions to change financial reporting to mislead some stakeholders about the company’s financial performance or influence contractual outcomes that depend on informed accounting practices. Following previous research in accounting literature, Earnings Equity (EQ) is opposite of earnings management (Barth et al., 2008) 29 .
To develop our hypotheses about the relationship between CSR and EQ, we rely on previous studies, both theoretical as well as empirical evidence. As the previous studies document, this relationship depends on three links, as shown in figure 1.
Content analysis CR activity
Figure 1: Link between CSR activities and EQ.
Content analysis is subjective, but it is a reliable proxy for the measurement of CSR activities disclosure 30 .
In the literature, we can find several theories that explains the link between CSR activity disclosure and EQ (ethical theory, agency theory or neo-classical theory, legitimacy theory, “trade-off theory”, Supply and demand theory, Stakeholder theory, and others). In this study, we applied ethical theory to solve the questions, because in our point of view this theory better explains the link between these two variables.
For 27 ( p 289) [05 pp 289]Carrol (1979, pp. 289), ethical theory imply that socially responsible company must accept ethical obligation, adhering to highest behavior standard, being trustworthy and honest in their business process.
Recent studies show that CSR activities disclosure generate transparency because they are motivated by voluntary information (Kim et al., 2012; Bozzolan et al., 2015) 27, 15 [05, 20] Transparency and accountability are the main characteristic of the socially responsible company, allowing stakeholders to confront the company, Bozzolan et al. (2015)  27.
Kim et al. (2012) 15 , voluntary CSR activities disclosure also leads to less engagement in earnings management because CSR firms are inherently committed to transparent disclosure policies and creating more shareholder value, which implies a higher earnings quality. Zang (2012) 10 , argue that CSR-firms care about their capability to deliver more value to stakeholders, they might be less likely to engage in earnings management activity.
Overall previous research support the idea that CSR behave firm is negatively linked to earnings management and positively associated to earnings quality.
The literature on the relationship between CSR and earnings management is scarce and the results are mixed. However, these researches have examined CSR policy in a positive disclosure setting. In our study, we are interested in how both positive and negativeCSR activity affects earnings quality.
Empirically the study of Trébucq and Russ (2005) 31 , was considered one of the first studies that examined the relationship between corporate social responsibility and earnings management. Using data from the 587 U.S. firms as a sample. However, these results were not significant. In addition, the research of Chih et al. (2008)  12 uses cross-country data from 1653 corporations in 46 countries during the period of 1993-2002, to examine the association between CSR and earnings management. This provided inconsistent results across different earnings management proxies (smoothing, earnings aggressiveness and earnings losses and decreases avoidance). It also is shown that CSR firms are more aggressive in accruals management but less likely to engage in earnings loss avoidance and earnings smoothing. Then they calculated the results in multivariate regressions and found that the relationship was a greater commitment to CSR that leads to a mitigation of earnings smoothing reducing the avoidance of earnings losses. Yip et al. (2011)  14 have examined whether the relationship between earnings management and social corporate disclosure is affected by ethical consideration. Theytestedtheirhypothesesusingdatafromtwo differentU.S.industries,theoilandgasindustry (30 firms observed)and thefoodindustry (80 firms observed). It was found that for the food industry CSR disclosure is positively significant linked to earnings management indicating poor earning quality, but for oilandgasindustry the relationship between CSR disclosure and earnings management is negatively significant. This tow evidences support the political hypotheses point of view but not the ethical hypothesis.
Overall, the results show that the relationship between earnings management with CSR is unclear. This lead to the following hypothesis:
- H0:CSR firms has a significant impact on earnings quality.
Kim et al. (2012)  15 using a sample of US firms from 1991 to 2009 examined three different proxies for earnings management on the relationship with CSR: (1) discretionary accruals, (2) real activities of manipulation and (3) the incidence of Accounting and Auditing Enforcement Releases applying the Kothari et al. (2005)  21 model as a proxy for determinate discretionary accruals (earnings management). The outcome was a negative and significant relationship between the magnitude of earnings management and CSR – Score which supported the transparent financial reporting hypothesis, suggesting that the CSR firms behave more ethically compared to non-CSR firms. Pyo and Lee (2013)  09 using 4,198 firm-year as sample of non-financial firms listed on the Korea Stock Exchange (KSE) from 2004 to 2010, to examine the association between the level of CSR activities and earnings quality with the level of donation expenses and the voluntary issuance of CSR reports filed with the Global Reporting Initiative (GRI) as proxies for CSR activities. The outcome was a negative relationship between CSR activity and earnings management, more specifically a negative link between discretionary accruals and donation in Korean listed firms.
Gil et al. (2016)  18 investigated the relationship between corporate social responsibility and earnings management. Using 100 most reputed Spanish non-financial companies according to the Merco Index between 2005 and 2012, the result was a negative impact of CSR activity on earnings management. Park (2017)  32 examined whether CSR activity affects tax avoidance using a sample of 1,477 Listed Korean (KSE) firms-year observations from 2004 to 2009. It was found that there was a significant and negative relationship between CSR activity and earnings management, more specifically, the outcome was that companies with CSR behavior significantly related to lower tax avoidance. Hence, according to Hong and Andersen (2011)  13 more socially responsible company have higher quality accruals and fewer earnings management behavior. This suggested a negative association between CSR and earnings management. This lead to the following hypothesis:
- H1:Thereis a positiverelationship between CSR and earnings quality.
4.1 Data and Sample
Our initial sample consists of all the extractive firms operating in Mozambique according to the National Institute of Statistics (INE–Mozambique), over the period from 2010-2017. The financial reporting data was downloaded from the company’s website and it is supplement with data extracted from the INE–Mozambique website and Mozambique Stock Exchange (BVM) website 19, 33 [43, 45]. The final sample consists of 368 firm-years, after excluding firms with missing necessary data for our analysis and, we delete observations in the bottom and the top 1% of standard deviations in each tail to minimize the effect of the outliers in the regressions.
4.2 Estimation Methods
In this section, we present our research design. First, the measurement of CSR in this study adopted a similar disclosure-scoring methodology based on content analysis. Such technique consists of codifying qualitative information in order to quantify the corporate social responsibility level. It requires the definition of categories information and the choice of the unit of measurement. Items selected for inclusion based on their relevance to the Mozambican context, and these was classified into four categories of CSR namely: environmental performance, Community involvement, employee relations, and product. Each category has sub-categories disclosures that are adjusted based on whether the items are disclosed (see Table 1).
It should be noted that the adoption of this model arises because in Mozambique there are no rating organizations – trusted third-party institutions with the resources to collect the necessary information for quantitative CSR and, in accordance with to common practice in previous research (Branco and Rodrigues, 2006)  30 . In addition, we present the measures of earnings quality according to Dechow et al. (1995)  20 and Kothari et al. (2005) 21  models.
(Insert Table 1 here)
4.2.1 Corporate Social Responsibility (CSR)
Several studies used different methods to measure CSR. In line withBranco and Rodrigues (2006) 30 , we applied the content analysis method to select and sort firms into CSR firms and non-CSR firms. Thus, we counted and used the total number of sentences provided for each firm within each category of social information for both the coding and measurement elements, under the content analysis method. More specifically, in this study, we obtained the CSR disclosure from the firm annual reports over the eight years of our study and their websites. First, the level of corporate social disclosure (CSR_Score) = Number of sentences related to environmental information + Number of sentences related to local community involvement information + Number of sentences related to Human resources information + Number of sentences related to products information. Based on these 4 categories companies were scored from 0-4 deepened on how many categories they disclosed in their annual reports. For each requirement, a score of one and zero points was given to the annual report, Branco and Rodrigues (2006)  30 argue that it is a better way to classify the companies. According to Van Staden & Hooks (2007) 34 , firms use various media to make CSR disclosures, so we supplement these data searching the firms` websites to identify firms which have issued CSR in their reports. Firms were considered as disclosing firms if there was CSR report on their website and non-disclosing firms otherwise. CSR reports through their website as a stand-alone document for the October of 2016 to March of 217. This variable is labeled CSR_Web and is coded as 1 for disclosing firms and 0 otherwise.
4.2.2 Earnings Quality
Following previous research in accounting literature, earnings equity is opposite of earnings management (Barth et al., 2008)  29. According to previous studies discretionary accruals is a proxy for the earning quality, we defined discretionary accruals (abnormal) as total accruals minus estimated normal accruals (non-discretionary), where the estimated normal accruals are derived from a number of discretionary accruals (Dechowet al., 1995; Kotari et al., 2005 and Jones et al., 2008) 20, 21, 35 [11, 17, 21] . Gil et al. (2016)  18 argue that the manager to meet earnings targets can involve either income-decreasing accruals or income-increasing accruals, so the magnitude of absolute discretionary accruals is used to assess the degree of earnings management. So, the lower magnitude of absolute discretionary accruals corresponds to a lesser level of earnings management, or higher earnings quality, and vice-versa.
Where: DAC denote discretionary accruals for firm i in period t; TAi,t = Total Accruals for firm i in period t and NDACi,t denote non-discretionary accrual for firm i in period t.
To compute the accrual component we use the following equations:
TAi,t=(ΔTCAi,t – Δcashi,t) – (ΔTCLi,t – ΔLTDi,t) – Depi,t
Where all variables have already been defined,
ΔTCAi,t = Change in Total Current Assets between year t and in year t-1;
ΔCashi,t = Change in cash and cash equivalents between year t and in year t-1;
ΔTCLi,t = Change in Total Current Liabilities between year t and in year t-1;
ΔLTDi,t= Change in Long-Term Debt included in Current Liabilities between year t and in year t-1 and, Depi,t = Depreciation and Amortization Expenses.
Additional, we follow both the cross-sectional modified Jones (1991) model as suggested by Dechow e al. (1995)  20 and the performance-matches discretionary accruals model suggested by Kothari et al. (2005)  21 to compute the non-discretionary component of total accruals (TA) as shown in equation 3 and equation 4.
TAi,tASi,t-1=λ0 +λ11ASi,t-1+λ2ΔRevi,t- ΔARi,tASi,t-1+λ3PPEi,tASi,t-1+ Ɛi,t
TAi,tASi,t-1=λ0 +λ11ASi,t-1+λ2ΔRevi,t- ARi,tASi,t-1+λ3PPEi,tASi,t-1+λ2ROAi,t-1+ Ɛi,t
Where all variables have already been defined, ASi,t is total assets in year t-1;
ΔRevi,t the change in revenues between year t and in year t-1;
ΔARi,t is changes in accounts receivable between year t and in year t-1; PPE is the level of gross property, plant, and equipment scaled by lagged total assets (ASt−1) in order to avoid problems of heteroscedasticity; ROA is a return on assets in year t-1.
Using the coefficient in equation 3 and equation 4 we estimated the equation 5 and equation 6 by regressing financial statement values for each firm in the industry:
NDACi,tASi,t-1=β0 +β11ASi,t-1+β2ΔRevi,t- ΔARi,tASi,t-1+β3PPEi,tASi,t-1+ Ɛi,t
NDACi,tASi,t-1=β0+β11ASi,t-1+β2ΔRevi,t- ARi,tASi,t-1+β3PPEi,tASi,t-1+β2ROAi,t-1+ Ɛi,t
Where all variables have already been defined.
Finally, we compute discretionary accruals (DAC), our equation (1)
4.3 Basic Model
Consistent with prior study, we employ the absolute value of discretionary accruals (Abs_DAC) as our measure of earnings management (Jones, 1991; Kothari, 2005, Yip et al., 2011, Salewski and Zulch, 2014) [17, 21, 33, 39] 14, 17, 21, 36, . In order to evaluate the effect of CSR activity on earnings management, we regress the absolute value of discretionary accruals (Abs_DAC) on CSR_Fall and control variables. We expect that Corporate Social Responsibility (CSR) are associated with earnings management. The basic model is as follows:
Abs_DACi,t=λ0 + λ1CSR_Falli,t + λ2Levi,t + λ3Sizei,t + λ4RAOi,t + λ5Big4i,t + λ6Growthi,t + Year dummy + Industry dummy + εi,t
Abs_DACi,t donate absolute value of discretionary accruals in year t, scaled by lagged total assets computed using the Dechow et al. (1995) model 20  (model 1) or Kothari et al. (2005) model 21  (model 2); CSR_Falli,t is a dummy variable that equals 1 if the firm is classified as social responsible firm and zero otherwise (CSR_ Falli,t = CSR_Score + CSR_Web); Levi,t is long-term debt scaled by total asset; Sizei,t is the natural logarithm of total assets in year t; RAOi,t is net income by end-of-year divided by total asset; Big4i,t is a dummy variable that equals 1 if the firm is audited by Big4 and zero otherwise; Growthi,t is changes in sales between year t and in year t-1; Year and Industry are dummies representing two-digit INE industries and years and εi,t is white`s robust error regression coefficient.
Following prior research (e.g., Gil et al. 2016)  18 we employ the absolute value of discretionary accruals (Abs_DAC) as our earnings management`s proxy. If H0 holds true, we predict a negative association between CSR and the degree of earnings management. Therefore, for the control variables we predict that firms have greater leverage, small size, a lower growth degree, negative income, higher return are likely to earnings management through the absolute value of discretionary accruals and provide poorer earnings quality (DeFond and Subramanyam, 1998)  37.
We test if CSR orientation firms is positively related with earnings quality (H1) by estimating more three different measures of CSR score in place of the single one used in model (7.0):
Abs_DACi,t=λ0 + λ1CSR_FSTi,t + λ2Levi,t + λ3Sizei,t + λ4RAOi,t + λ5Big4i,t + λ6Growthi,t + Year dummy + Industry dummy + εi,t
Abs_DACi,t=λ0 + λ1CSR_FCOi,t + λ2Levi,t + λ3Sizei,t + λ4RAOi,t + λ5Big4i,t + λ6Growthi,t + Year dummy + Industry dummy + εi,t
Abs_DACi,t=λ0 + λ1CSR_Findi,t + λ2Levi,t + λ3Sizei,t + λ4RAOi,t + λ5Big4i,t + λ6Growthi,t + Year dummy + Industry dummy + εi,t
Where all variables have already been defined, CSR_FSTdenote positive index of CSR activity and CSR_FCOdenote negative index of CSR activity and CSR_Find represents the overall individual category of CSR activities.
4.3.1 Variable Explanation
18.104.22.168 Dependent variable
Abs_DACi,t donate absolute value of discretionary accruals in year t, scaled by lagged total assets computed using the Dechow et al. (1995) model  20 (model 1) or Kothari et al. (2005) model 21  (model 2), prior accounting literature (e.g, Jones, 1991; Kothari et al., 2005; Yip et al., 2011, Kim et at., 2012; Salewski and Zulch, 2014; Gil et al., 2016.) [13, 17, 21, 20, 33, 39], use the Abs_DAC as earnings management because it involves either income-decreasing or income-increasing accruals. If the result is consistent with the ethical and moral hypothesis, a negative relation between the absolute discretionary accruals and CSR activity is expected.
22.214.171.124 Independent Variable
CSR_Falli,t denote CSR activities index (the sum of positive and negative category of CSR activity), CSR_FSTi,t denote positive index of CSR activity and CSR_FCOdenote negative index of CSR activity and CSR_Findi,t represents the overall individual category of CSR activities (environmental protection, local community, employee relationship and products), to explore how each dimension interacts with earnings management activity. If H1 holds true, we predict a negative and significant coefficient on λ1 (higher earnings quality).
126.96.36.199 Control Variables
Following prior study, we added several control variables that could affect CSR activity and financial reporting behavior (e.g., Park 2017; Gil et al., 2016; Salewski and Zulch 2014, Kim et al., 2012; Yip et al., 2011) [13, 20, 27, 33, 39] 14, 17, 18, 32 . First, we control leverage (Lev) an indicator of the firm’s financial structure. There are two opposite empirical findings regarding the relationship between the leverage and earnings management. Press and Weintrop (1990)  38 argue that high leverage firms tend to manage earnings aggressively to avoid violating debt covenants, while Dechow et al. (1995) 20  suggest that high leverage may imply fewer earnings management. At the same time, leverage is also a frequently tested factor to analyze the influential elements on CSR (e.g., Lev et al., 2010)  08. It is expected that the coefficient on Lev variable to be positive through the absolute value of discretionary accruals. Second, we control firm size (Size) measured by the natural logarithm of total assets. Size has widely been recognized as a proxy for political costs hypothesis (Zimmerman 1983)  39. The larger the size the firm is enabled to make more contributions to the whole society. We predict that the relationship between firm size and discretionary accruals will be negative. Third, we control for firm performance. Following Dechow et al. (1995)  20 the discretionary accruals are correlated with firm profitability. We expect that the relationship between ROA and discretionary accruals will be positive. Fourth, we control Big4 as a proxy for the quality of auditing. The auditors, who are concerned with litigation risk, may decrease discretionary accruals or increase the level of conservatism (DeFond and Subramanyam, 1998)  37, a negative association with discretionary accruals is expected. Fifth, we control growth firms (Skinner and Sloan 2002) 40. Following Chen et al. (2008)  41 predicted the negative association between earnings management and firm’s growth, thus suggesting that firms with problems have more incentives to engage in earnings management. Finally, following Gil et al. (2016)  18, we include year and industry dummies to control industry`s effects on earnings management.
5. Results and Discussion
5.1 Statistics and Correlation Matrix
Table 2 summarizes the descriptive statistics of the main variables of interest for the object of the study. We can observe that the mean value of CSR_Fall is 5.071, these results represent a variation in the corporate social responsibility behave. Moreover, concerning earnings quality aspects the mean value of absolute discretionary accruals (Abs_DAC) moves around 0.051. This value is lower in our sample compared to that of Yip et al. (2011)  14 and Gil et al. (2016) 18  and is so close to the value presented in Choi et al. (2013) 02  and Prior et al. (2008) 11  studies.
(Insert Table 2 here)
The correlation matrix of the main variables used in our study is shown in Table 3. We can observe a positive and significant correlation coefficient between absolute discretionary accruals (Abs_DAC) and the ratio of the three measures of Social activity (CSR_Fall, CSR_FST and CSR_FCO), overall, the results suggest that our sample firms with low CSR activity are more likely to engage in earnings management.
The level of leverage (Lev) is significantly and positively correlated with Abs_DAC ( = 0.291), implying that in all the firms are leveraged and poorly performing which tends to manage earnings. Further, it can be observed that the firm size is negative and significantly correlate with Abs_DAC ( = -0.401), implying that large firms are more likely to engage in CSR activity. Likewise, for other control variables, we find correlations consistent with previous researches.
As the highest correlation is about 40% between variable Abs_DAC and size, the highest VIF – variance inflation factor has a maximum value of 1.414, showing that the multi-collinearity does not appear to be a problem in both models.
(Insert Table 3 here)
5.2 Regression Result
Table 4, presents the results after applying the regression model, the first 2 columns; display the estimations for equations using Dechow et al. (1995)  20 (model 1) and, the last 2 columns displays estimations for equations using Kothari et al. (2005)  21 (model 2). This model investigates the effects of the CSR activity on earnings management behave. We can observe that the coefficient λ1 for all three different measure of CSR activity is positively associated with all measure of earnings management at the conventional level (p < 5% or better).
In the nutshell, our model overall is statically significant (P-value <0.05) and the highest Adj. R-squared equal 45.8%; indicating the strength and explanatory of our model. This result, also displayed the maximum VIF – variance inflation factor as 2.241, suggesting that the multi-collinearity does not appear to be a problem. Similarly, both models did not show the endogeneity problems, the autocorrelation problem neither, even the problems of heteroscedasticity because we applied the respective statistics test to solve it.
Particular, in panel A, we find that the coefficient on CSR_Falli,t variable (i.e., the sum of positive and negative category of CSR activity) is significantly positive linked to both measure of earnings management (λ1 = 0.058, t-stat = 2.130, and λ1 = 0.072, t-stat = 3.011 respectively). This result suggests that CSR firms manage earnings more through discretionary accruals (poorer earnings quality). Specifically, this result is inconsistent which the ethical hypothesis which expects a negative relationship between CSR disclosure and earnings management. Also, the finding provides strong evidence for the effect of CSR activity in rising earnings management. Jointly, these findings do lend to support our null hypothesis H0 but do not lend support our alternate hypothesis H1. These findings are consistent with the results of Petrovits (2006)  42 and Prior et al. (2008) 11 , who argue that companies tend to pay more contributions to humanitarian foundations in order to realize earnings purposes. However, these results are in contrast with the findings of Chih et al. (2008)  12, who argue that companies with less CSR conduct greater earnings management.
Following prior study, we added several control variables that could affect CSR activity and financial reporting behavior. First, it can be observed that the coefficient λ2 on leverage is positive and statistically significant (p < 5% or better) for both the model 1 and model 2 ( = 0.411, t-stat = 3.020 and = 0.045, t-stat = 3.460), suggesting that absolute discretionary accruals are increasing in leverage. Consistent with the prior studies by (Watts and Zimmerman, 1978; Skinner and Sloan, 2002[34, 38]) 40, 43 who argue that earnings choice has been found to be associated with debt covenant violation. So managers can increase earnings to avoid debt covenants violations. Our findings are consistent with the prior researchers (Gil 2016; Pyo and Lee, 2013; Kim et al., 2012) [13, 20, 31] 09, 15, 18.
Additionally, we control the effects of firm size trough absolute discretionary accruals. As can be observed the coefficient λ3 on Sizei,t variable is significantly negative (p < 5% or better) for both the model 1 and model 2 ( = -0.003, t-stat = -1.512 and = -0.002, t-stat = -1.501) imply that the firm’s sizes have an important impact on the earnings management. More specifically, these findings imply that large firms manage earnings less. According to prior study (Yip et al., 2011), the political costs hypothesis explain why firm size is significantly related to earnings management. Following Zimmerman 39 (1983), large companies with great profits may try to manage earnings downwards. Therefore, we control the firm performance trough absolute discretionary accruals. It can be observed that the coefficient λ4 on ROAi,t variable is negative and significant for both the model 1 and model 2 ( = -0.008, t-stat = -2.997 and, = -1.009, t-stat = -2.086), suggesting that the firms have the incentive to practice earnings management, because the profitability is not high enough to satisfy managers and investors (Skinner and Sloan, 2002)  40.
In term of our control variable Big4, it can be observed that the coefficient λ5 on Big4i,t variable is negative but not statically significant (p > 10%) for the model 1 and model 2, respectively, and is inconsistent with the idea that auditing provides a constraint for earnings management practice (Park, 2017)  32. Thus, we do not have enough evidence that firms audited well-established audit firm’s (Big4) engage less in earnings management compared to the firms audited by the smaller auditors (non-Big4). We, also, control growth through absolute discretionary accruals. It can be observed that the coefficient λ6 on Growthi,t variable is negative and significant, as expected.
Interestingly, in panel B, we find a positive link between strength (CSRST) and earnings management at the conventional level (λ1 = 0.049 and t-stat = 4.022 and, λ1 = 0.054, t-stat = 3.988 respectively). Similarly, in panel C, we find that concern (CSRCO) is positively associated with all measure of earnings management (λ1 = 0.048, t-stat = 3.957 and, λ1 = 0.052, t-stat = 4.017 respectively). These results strongly suggest that both positive and negative indicator of CSR activities play a significant role in rising earnings management.
Regarding control variables, as observed in panel B and C, the results are similar to those documented in panel A.
(Insert Table 4 here)
Additionally, we carried out two additional analyses to address possible alternative justifications for our results. First, we re-run our model applying each individual category of voluntary CSR activity (i.e., environment, community, employee relations, and product) as metrics of total CSR activity score.
As can be observed in Table 5, two positive indicators of CSR activity, specifically, CSR_F1(the environmental protection) and CSR_F2 (the relationship whit the local community) are positively related with earnings management at the conventional level (λ1 = 0.0398, λ1 = 0.0357 and, λ1 = 0.0317, λ1 = 0.0229 respectively) suggesting that the firms use this two elements of social responsibility behavior as tools to engage earnings, it means that the CSR firms invest less in local community relations and the environmental protection activities.
For negative indicators of CSR activity, we observed one puzzling result on the CSR_F3 variable, which is negatively significant (p < 5% or better). Implying that employee relationship is negatively related to earnings management. In contrast, it can be observed that for CSR_F4 variable (Products) is insignificantly negative related to earnings management, at the conventional level. The findings suggest that this CSR element do not affect the earnings management activity. Jointly, our findings suggest that earnings management activity is being driven specially by community and environmental categories.
(Insert Table 5 here)
Additionally, we investigate whether a positive indicator of CSR activity (CSRST) differs from a negative indicator of CSR activity (CSRCO) in rising earning quality. Therefore, we construct a nested model with CSRCO and CSRST variables in the same regression. As can be observed in Table 6, the coefficient of CSRST is positive and significant for all measure of earnings management (λ1 = 0.051 and λ1 = 0.053, p < 5% or better). However, the coefficient of CSRCO is negative and insignificant at the conventional level for both models (λ1 = 0.026 and λ1 = 0.043, p > 10%). Likewise, we find that the difference in the two coefficients (CSRTS vs CSRCO) is positive and significant (F-test 27.13). This result suggests that a positive indicator of CSR activity is more impactful in terms of its effect on earnings management than a negative indicator.
These findings are consistent with hypothesis H1, still providing poorer earnings quality, after the inclusion of the positive and negative indicators of CSR activity in the same model which and jointly do not support the hypotheses H1 too.
(Insert Table 6 here)
Instrumental variables (2SLS) regression analysis
Followingprevious studies (e.g., Scholtens and Kang, 2013; Bozzolan et al., 2015) [35, 03] 07, 16, to avoid endogeneity problem we carried our 2SLS regression to validate the interpretation of the results.The findings shows that the coefficient of CSR_FST still positive and significant at the conventional level (λ1 = 1.081 and λ1 = 1.094, P < 5% or better), even after accounting for endogeneity. This findings implies that endogeneity does not appear to be a problem in our model. Similarly, we carried out the F-test across the two models and obtained 31.84 which is above the conventional level (minimum of 10). Likewise, we rejected the null hypothesis (p < 0.000).
Similarly, both models did not show the endogeneity problems, the autocorrelation problem even the problems of heteroscedasticity because we applied the respective statistics test to solve it.
(Insert Table 7 here)
6. Discussion and Conclusion
Starting from the hypothesis, an empirical study has been developed, we examine the relationship between Corporate Social Responsibility (CSR) and Earnings Quality using Mozambican extractive industry the sample for the period from 2010 to 2017. Applying the absolute value of discretionary accruals (Abs_DAC) as our measure of earnings management and an alternative measure of CSR (using content analysis of annual financial reports for obtaining quantitative data and supplementing these data searching the firms` websites reports about CSR discloser), our main findings can be summarized in these points. Firstly, the coefficient λ1 that captured the association between CSR_Falli,t and Abs_DACi,t was significantly consistent with previous literature (Gil et al., 2016; Choi et al., 2013; Yip et al., 2011 and Prior et al., 2008) [08, 13, 30, 39] 02, 11,14, 18, providing evidence to lend support our null hypothesis H0:Thereis a relationship between CSR and earnings quality. We further find that both positive and negative voluntary CSR activity are related do earnings quality.
Secondly, we expected but did not provide evidence to lend support our hypothesis H1: Thereis a positiverelationship between CSR and earnings quality. This implying that our sample does not support the ethical hypothesisthroughthe absolute value of discretionary accruals. Additionally, we show that positive indicator of CSR activity is more impactful in terms of its effect on earnings management than negative indicator of CSR activity.
In terms the four elements of corporate social activity, it obvious that earnings management activity is being driven specially by community and environmental categories. These findings suggests that managers shows less commitment to the corporate social responsibility and they are manipulating earnings to archive their particular interest. Consequently, these findings are consistent with the idea that the opportunistic managers use CSR activity to reach their particular interest. Our results are robust to alternative proxy measures of CSR activity and earnings quality.
Our results are consistent with the study of (Prior et al., 2008; Choi et al., 2013) 02, 11 [08, 30]. However, the results are inconsistent with the study of (Kim et al., 2012; Pyo and Lee, 2013; Gil et al., 2016; Park (2017) [13, 20, 27, 31]. 09, 15, 18, 32
In the scientific scope, we believe that this study makes a great contribution filling the gap in the literature, particularly assessing an unexplored link between voluntary CSR activity and earnings quality. Also, we show the important role of both positive and negative indicator of CSR activity, which has been overlooked in previous studies.
In practical terms, we believe that these results have important implications for investors, managers, regulators and, other stakeholders showing that CSR firms are less commitment to manipulating earnings.
This study is subject to at least two limitations: First, the fact that the study concentrates on one kind of industry may limit the possibility of generalizing the findings to other contexts. Future studies might address this issue in different sector of activity, in order to explore other institutional contexts. Second, the study was based on self-collected information from the annual reports, firms’ websites internet searches; perhaps if CSR data were collected from other source the result could be different.