Sustainability is a complex concept, and presents differing views on theoretical frameworks. The CSR concept is a way for companies to commit with all stakeholders inside and outside the organization (stakeholders), without focusing only on partners and shareholders (stockholders). Carroll (1979) explains that businesses’ social responsibility is comprised of the economic, legal, ethical, and discretionary perspectives that society presents to the organization. By this definition, companies should aim to profit according to legal requirements, and act as ethical and appropriate corporate citizens.

Within this theoretical framework - which provides a direction to and foundation for the existence of CSR activities and the dissemination of economic, social, and environmental information - Garriga and Melé (2004) incorporate various theories to classify CSR into four groups: ethical, political, integrative, and instrumental theories.

The first set of theories explains that companies should accept social responsibility as an ethical obligation (Carroll, 1979; Donaldson & Preston, 1995; Jones, 1995) and promote what is ethically acceptable for society. Regarding political theories, companies must seek a formal alternative to publicize the improvements they desire to insert in the community (Kim, Park, & Wier, 2012). Business success under integrative theories is a condition that involves by integrating the business’ social demands (Carroll, 1979; Swanson, 1995). Finally, in considering the economic objectives arising from instrumental theories, for example Friedman (1970), it is observed that CSR activities are a means by which stockholders receive added wealth. In this context, within the CSR theories of groups, the legitimacy and agency theories and the theory of stakeholders exist at the individual level.

The theory of legitimacy perceives the company and society as parts of a social contract, in that companies operate satisfactorily within the limits that society deems as socially acceptable (O’Donovan, 2002). The theory’s underlying assumption is that a two-way influence exists between the authorities and the society in which they operate (Deegan, 2002). From the theory of legitimacy perspective, CSR disclosures can be strategic by providing legitimacy for the company (Patten, 1990; Woodward, Edwards, & Birkin, 1996) and extrapolating economic goals to add to these social and environmental objectives.

The stakeholder theory considers the company as an organization comprised of both interdependent parts and divergent interests, and suggests that the wealth companies generate is established by their relationships with specific parts of society, rather than the society at large (Carroll, 1979; Donaldson & Preston, 1995; Jones, 1995).

This theory aims to align the company’s and stakeholders’ interests and the disclosure of sustainable information as a means to meet the demands for specific parts by seeking acceptance of their strategies (Gray, Kouhy, & Lavers, 1995).

Nevertheless, the problem caused by information asymmetry, which allows for the expropriation of wealth from the company’s minority shareholders, is the core of agency theory. The agency problem essentially states that managers with opportunistic financial reports tend to disclose CSR activities to compensate for poor-quality gains (Hobson & Kachelmeier, 2005; Krüger, 2015).

Financial reporting is not only a tool for decision-making, but also an informative mechanism for both managers and shareholders (Moneva & Llena, 2000). Reports that incorrectly depict the business economy increase the probability that shareholders will not make optimal decisions (Prior et al., 2008). In this context, the EM may be understood as an agency cost, as the EM reduces the quality FRQ (Kinney, Palmrose, & Scholz, 2004).

From this we can deduce that companies with transparent and reliable financial reports have incentives to disclose as much information as possible, such as information related to CSR activities (Martínez-Ferrero, Garcia-Sanchez, & Cuadrado-Ballesteros, 2015). On the other hand, opportunistic managers that publish low-quality financial reports have incentives to disclose CSR information, because the disclosure of CSR is a way to externalize the agency problem (Hobson & Kachelmeier, 2005; Krüger, 2015).

The association between CSR and financial reporting behavior is an empirical question, as recent academic literature has demonstrated. The above arguments indicate that two aspects in this research link CSR to FRQ: the possibility of transparent financial reporting, and the possibility of retreatment.

Chih et al. (2008), Kim et al. (2012), and Martínez-Ferrero et al. (2015) directly discuss these conflicting hypotheses. After analyzing 1,653 companies in 46 countries, Chih et al. (2008) have provided inconsistent results for different proxies of EM. Martínez-Ferrero et al. (2015) in multi-national research have used measures of DA, accounting conservatism, and the quality of accruals as proxies for FRQ. The authors demonstrated that conservative companies with low levels of EM and/or high-quality accruals tend to present high-quality CSR information.

Several authors used different measurements of the variables and econometric estimators, with different results. For example, Chih et al. (2008) used the Global FTSE4 Good index, while Martínez-Ferrero et al. (2015) used reports’ content to identify the level of adherence to the Global Reporting Initiative (GRI) standards, with an ordinal variable valued between zero and three.

Limiting the search to the US market, Kim et al. (2012) investigated whether financial statements differ in companies that are considered socially responsible. This study indicated that CSR companies are less likely to manage results through discretionary accruals and the manipulation of operational activities.

Recent literature has tested various hypotheses involving transparent and retreatment reports in a theoretical framework or, separately, empirically, and concentrated on causal relations in opposite directions: either the FRQ proxies as explanatory of socially responsible enterprises (Choi, Lee, & Park, 2013; Gao & Zhang, 2015; Prior et al., 2008); or CSR as a cause of FRQ (Bozzolan, Fabrizi, Mallin, & Michelon, 2015; Hong & Andersen, 2011; Muttakin, Khan, & Azim, 2015; Pyo & Lee, 2013).

Prior et al. (2008) examined whether companies use CSR to strategically hide EM by applying a proxy CSR from the scores collected by the Sustainable Investment Research International Company. The authors captured EM by calculating discretionary accruals and smoothing the results from a multi-country sample of 593 companies between the years 2002 to 2004; these indicated a positive relationship between EM and CSR for regulated companies. However, this result was not statistically significant for non-regulated businesses.

Choi et al. (2013) tested whether corporate governance’s size affects the intention to promote CSR managers. To operationalize the research, discretionary accruals were adopted based on the Jones model as modified by Dechow, Sloan, and Sweeney (1995), which used CSR proxies and corporate governance measures. The results indicated a negative relationship between the EM and CSR.

Gao and Zhang (2015) revealed that CSR is desirable in companies and adds a quality dimension. Their empirical data indicated that smoothing companies with high CSR results demonstrate incremental improvement in their yield-return ratios, and have significant, positive relationships with performance measures.

Hong and Andersen (2011) studied whether CSR correlates with FRQ by measuring CSR based on scores from the Kinder, Lydenburg, and Domini index in the North American stock market between 1995 and 2005. The authors noted that socially responsible companies are less likely to manage results.

Pyo and Lee (2013) Korean stock market research adopted the following CSR measures: i) a sustainability reporting disclosure in the standards issued by the GRI, and ii) expenses that include company donations. A regression against these accruals - captured by the discretionary model modified by Dechow et al. (1995) and Kothari, Leone, and Wasley (2005) - and two accounting conservatism measures demonstrated that companies with high donations have less discretionary increase and greater accounting conservatism. Additionally, voluntary reporting on GRI models exhibited a moderating relationship.

Nevertheless, the results appear to be sensitive to shapes of the constructs’ measurement proxies. For example, Bozzolan et al. (2015) used an EIRIS-specialized agency’s CSR data to collect a sample of companies in 24 countries to demonstrate that CSR activities constrain real manipulation gains. Already, Muttakin et al. (2015), measured a CSR index with twenty categories for a sample of 135 non-financial companies in Bangladesh that managers have low-quality reports when CSR provides further disclosure.

Following the findings in literature and as investigated by Chih et al. (2008), Kim et al. (2012), and Martínez-Ferrero et al. (2015), the hypotheses to be tested in this research pertain to transparent financial reporting (H1a) and retreatment (H1b):

H1a: A negative relationship exists between CSR and EM.

H1b: A positive relationship exists between CSR and EM.

Dependent variable measurement: FRQ. Based on previous literature relating CSR and the FRQ, we adopted DA as a proxy for the DA-FRQ. We use Jones (1991) model to capture the DA in our main tests and adopted the modified version by Dechow et al. (1995) and Kothari et al. (2005) for robust tests. The Jones (1991) model is defined as follows:

TotAcc r t /T A t−1 = α 0 + α 1 ( 1/T A t−1 )+ α 2 ( Sale s t /T A t−1 )+ α 3 ( PP E t /T A t−1 ) (1)

Considering that TotAccr _t is the total operating accumulation, ΔSales _t this is the change in revenues in year (t - 1) to t; PPE _t is the account property, plant, and equipment gross, and TA _t-1 is the total assets at the end of the year (t - 1).

The total accumulations were calculated by the formula:

TotAccr _{t =} [(ΔCA _{t -} ΔCASH _{t) - (} ΔCL _{t -} ΔRLTP _{t) -} DA _t ] , where ΔCA _t is the change in current assets, ΔCASH _t is the variation in cash, ΔCL _t is the change in current liabilities, ΔRLTP _t is the change in financing and short-term loans, and DA _t is the total depreciation and amortization.

Fixed Assets is adopted as a variable to replace the property, plant, and equipment variable according to Martinez (2001), given the nomenclature accepted in the accounting classification of the research environment.

We estimated the model by a cross-section per year, with the DA comprised of the estimates’ residuals. The relationship between CSR and FRQ may depend on the measures adopted (Chih et al., 2008). Thus, we re-estimated Jones (1991) model with the modifications suggested by Dechow et al. (1995), and we remove the portion of the accounts receivables’ range of variation of revenue. We also applied a suggestion from Kothari et al. (2005) and extracted the performance effect in the calculation.

Variable of interest: D.GRI. We believe that companies are socially responsible according to the standardized level of information disclosed in accordance with the GRI’s international guidelines. We followed the approach as posited by Martínez-Ferrero et al. (2015), Nikolaeva and Bicho (2011), and Pyo and Lee (2013).

The variable of interest is a binary variable that assumes a value of one if the companies provide information contained in their sustainability reports - consonant with the highest levels of sustainability reporting guidelines, or the GRI’s version G4 - and a value of zero if the company does not exhibit this feature.

Control variables. Profitability: ROA and ROE. Dechow et al. (1995) note that profitability can correlate with DA. We use accounting measures of financial performance to capture profitability: the return on assets (ROA) and return on equity (ROE). The results regarding yield are mixed, as Choi et al. (2013) and Martínez-Ferrero et al. (2015) found a positive association between CSR and profitability, while Calegari, Chotigeat, and Harjoto (2010) found a positive correlation between yield and EM. Thus, the result per variable is not determinate. We calculate the ROA as the ratio between the variation of net income and total assets, but the ROE is calculated as the ratio between net income and the average change in equity.

Leverage: LEV. Companies with loan restrictions and a greater presence of capital providers can encourage managers to manipulate results. Nevertheless, companies with high debt may have to persuade capital providers by disclosing increasingly more social and environmental information (Solomon & Lewis, 2002)a standard designed to enable corporate ethical performance to be judged. If such ethical performance is interpreted as social and environmental behaviour our view is that it should lead to more social and environmental disclosure. Yet the willingness to communicate corporate social and environmental performance still seems to be limited. Of the social and environmental disclosures that exist many appear to be rather fragmented and disparate. In this paper we explore the apparent paradox between concern for the environment and the limited amount of corporate environmental disclosure (CED. Previous studies indicate that the leverage related to FRQ can be positive (Bozzolan et al., 2015; Calegari et al., 2010; Cho & Chun, 2015) or negative (Chih et al., 2008; Litt, Sharma, & Sharma, 2013). These mixed results do not lead us to anticipate an expected signal. We take the ratio between the current and non-current liabilities and total assets as a debt level proxy.

Growth opportunity: MTB. The opportunity for growth can affect EM (Roychowdhury, 2006), however, we do not anticipate one direction, as previous studies have reported mixed results. The operationalization of opportunity for growth follows the studies Bozzolan et al. (2015), Chih et al. (2008), Cho and Chun (2015), Litt et al. (2013), Martínez-Ferrero et al. (2015) and Muttakin et al. (2015). We take market-to-book ratio as a measure of opportunity for growth.

Firm size: SIZE. The companies’ size may be reflected in its financial statements, as larger companies may have more complex statements, and may be a possible source of managerial expropriation (Hochberg, 2012). Alternatively, increased monitoring by shareholders and analysts (Watts & Zimmerman, 1986) may cause minor EM in larger companies. Thus, the expected sign for this variable is ambiguous. According to Choi et al. (2013), Pyo and Lee (2013), and Scholtens and Kang (2013), we measure size using the natural logarithm of total assets.

We capture the relationship between CSR and FRQ as established in the literature and previous research by proposing the following link to test our hypotheses:

FRQ=f( GRI,ControlVariables ) (2)

This relationship is empirically tested by the following econometric model:

D A it  ( or POS.D A it , or NEG.D A it )=  β 0 + β 1 D.GR I it + β 2 RO A it + β 3 RO E it + β 4 LE V it +  β 5 MT B it + β 6 SIZ E it +  ε it (3)

Initially, following Prior et al. (2008) and Pyo and Lee (2013), we estimated the Model 3 with measures of discretionary accruals (DA) as surrogates for earnings management. Previously, because EM may involve either income-increasing or income-decreasing accruals (Klein, 2002; Warfield, Wild, & Wild, 1995), we re-estimated the Model 3 with the DA’s positive (POS.DA) and negative (NEG.DA), as dependents variables. This procedure aims to capture the effect of CSR toward the DA. According Kim et al. (2012), a negative sign between D.GRI and POS.DA will indicate that CSR firms engage less in income-increasing EM; conversely, we interpret a positive relationship between D.GRI and NEG.DA as indicative that CSR firms manage earnings less through accruals.

The hypothesis of transparent financial reporting (H1a) is supported if β ₁ it is negative for models estimated with the dependent variable DA or POS.DA, and positive when the model is estimated with NEG.DA as the dependent variable. The hypothesis retreatment (H1b) is supported if β ₁ is positive, when the model is estimated having a dependent variable DA or POS.DA; and negative when the dependent variable is NEG.DA.

Given the characteristics of the sample chosen to estimate the parameters of Model 3’s specifications using the technical panel data, since the longitudinal regression models provide an increased amount of information, and a greater data variability, and a greater efficiency estimation of the parameters. The estimators for this technique have been defined according to their relevant tests. Due to the characteristic of the sample we re-estimate Equation 3 with Tobit models, the non-tabulated results of which didn’t present statistically significant differences from the estimated panel.

Panel A in Table 2 illustrates the sample’s descriptive statistics (see Appendix A for the variables’ definition). The DA had negative average values for all specifications, suggesting that companies are more conservative when performing discretionary EM. This corroborates the assumption in the entire sample of other statistics. The frequency of positive DA is greater than its negative occurrence of pairs to the accrual model as captured by Jones (1991), and this is the re-estimation suggested by Dechow et al. (1995). However, the negative accumulations values are on average larger than the positive. The negative DA exhibits higher concentration values in the lowest quartile compared to the upper quartile’s positive DA values. Thus, the positive DA are less than the negative DA even with a higher prevalence, causing the average of discretionary accruals to be negative regardless of management’s direction; this indicates that managers exhibit more conservative EM practices in the analyzed companies.

Table 2

	n	Mean	Median	Std.	25th	75th	Min	Max
Dependent Variables
DA.1991	1,181	-0.009	0.005	0.119	-0.061	0.056	-0.341	0.256
DA.1995	1,181	-0.009	0.002	0.123	-0.061	0.056	-0.357	0.275
DA.2005	1,181	-0.013	0.000	0.124	-0.064	0.049	-0.377	0.274
POS.DA.1991	616	0.079	0.053	0.087	0.026	0.098	0.003	0.418
POS.DA.1995	602	0.083	0.053	0.090	0.028	0.103	0.003	0.429
POS.DA.2005	589	0.080	0.049	0.091	0.025	0.099	0.002	0.428
NEG.DA.1991	565	-0.102	-0.065	0.105	-0.140	-0.026	-0.415	-0.003
NEG.DA.1995	579	-0.102	-0.063	0.109	-0.138	-0.023	-0.442	-0.002
NEG.DA.2005	592	-0.103	-0.064	0.108	-0.140	-0.026	-0.443	-0.002
Control Variables
ROA	1,181	-0.004	0.000	0.081	-0.026	0.018	-0.235	0.253
ROE	1,181	0.073	0.076	0.378	-0.013	0.182	-1.034	1.385
LEV	1,181	0.658	0.594	0.373	0.462	0.753	0.144	2.183
MTB	1,181	1.826	0.990	2.548	0.318	2.138	-0.231	12.069
SIZE	1,181	21.764	21.888	1.605	20.616	22.854	18.331	24.973

Note. *, **, *** indicates statistical significance at levels of 0.10, 0.05 and 0.01, respectively. Variables are defined in Appendix A.

The measure of profitability ROA presents negative average, which may reduce the disclosure of transparent financial reporting. However, ROE has positive mean values. This result is evidenced when analyzing the last quartile for this variable. The inverse values between the variables do not allow to identify a clear causal direction of the relationship between profitability and EM. The average debt level is 0.658, indicating the trend to obtain financing through debt rather than self-financing, which may reflect in greater monitoring of the company by creditors. The MTB index presents concentrated values in the lower part of the sample, which shows a positive asymmetry (median less than the mean: 0.990 < 1.826). The positive average result is associated with the size of the maximum values that are larger than their minimum peers. The control variable SIZE presented statistical dispersion of data and extreme values in acceptable parameters, indicating homogeneity characteristic and symmetrical distribution.

Table 3 compares the descriptive statistics of firms classified as either socially responsible (CSR) or non-CSR. We define socially responsible firms as those that disclose CSR information as noted in the GRI guidelines, G4 version. This information is captured by the variable D.GRI which assumes the value 1 (one) if the firm has CSR information aligned with the GRI guidelines, G4 version and 0 (zero) otherwise.

Table 3

	n	Mean	Median	n	Mean	Median	z-test	Mann-Whitney Test
Dependent Variables
DA.1991	236	0.002	0.011	945	-0.012	0.001	-1.865	0.062 *
DA.1995	236	0.002	0.012	945	-0.012	-0.001	-1.766	0.078 *
DA.2005	236	-0.003	0.005	945	-0.016	-0.003	-1.573	0.116
POS.DA.1991	140	0.056	0.044	476	0.086	0.058	3.321	0.001 ***
POS.DA.1995	136	0.058	0.044	466	0.090	0.058	3.539	0.000 ***
POS.DA.2005	132	0.054	0.040	457	0.088	0.052	3.827	0.000 ***
NEG.DA.1991	96	-0.077	-0.041	469	-0.107	-0.072	-3.380	0.001 ***
NEG.DA.1995	100	-0.075	-0.035	479	-0.107	-0.072	-3.492	0.000 ***
NEG.DA.2005	104	-0.075	-0.042	488	-0.108	-0.070	-3.494	0.000 ***
Control Variables
ROA	236	-0.004	-0.002	945	-0.004	0.000	0.267	0.789
ROE	236	0.088	0.074	945	0.069	0.076	-0.333	0.739
LEV	236	0.626	0.619	945	0.666	0.589	-1.280	0.201
MTB	236	2.501	1.233	945	1.658	0.902	-5.134	0.000***
SIZE	236	22.899	22.853	945	21.480	21.552	-12.490	0.000***

Note. *, **, *** indicates statistical significance at levels of 0.10, 0.05 and 0.01, respectively. Variables are defined in Appendix A. CSR firms are identified by a dummy variable that takes a value of 1 (one) if the firm has CSR information aligned with the GRI guidelines, G4 version. Non-CSR firms assumed values 0 (zero) for absence of this characteristic. The p values for Mann-Whitney statistical mean differences test is one-tailed.

The final sample is composed of 252 different firms, resulted in an unbalanced panel of 1,181 firm-years (the 252 firms are not present in all analyzed years). We identify that the dummy variable D.GRI is present in 88 different firms which corresponds to 35% of the sample. In panel data analysis, it is possible to identify that 236 firm-years are considered CSR versus 945 firm-years non-CSR. The absence of mandatory CSR information justifies this evidence.

In average the positive accumulations (POS.DA) and negative accumulations (NEG.DA) are larger in non CSR firms (Mann-Whitney Test, p < 0.01). This indicates that EM, regardless of direction, may be greater in firms non-CSR.

Based on the Pearson correlation matrix illustrated in Appendix B, to make the document manageable, neither the dependent nor independent variables’ coefficients, without considering the details of the various EM estimates, and those of the independent variables are high, indicating that no multicollinearity problems exist. In detail, POS.DA negatively correlate with the variations in D.GRI (p < 0.01). This suggests that opportunistic managers can manipulate earnings to increase profits, accentuating the conflicts of interest.

Table 4 lists the results of the estimations from Model 3, including the dependent variables with discretionary accruals in total values (DA), and the positive- (POS.DA) and negative-derived (NEG.DA) Jones (1991) models.

Table 4

Panel A: Model 3 estimates
	DA.1991	POS.DA.1991	NEG.DA.1991
D.GRI	-0.003 (0.008) 0.730	-0.002 (0.008) 0.803	0.007 (0.019) 0.716
ROA	0.0750 (0.055) 0.170	0.117 (0.076) 0.126	-0.115 (0.104) 0.272
ROE	-0.002 (0.012) 0.896	0.009 (0.019) 0.658	-0.008 (0.015) 0.583
LEV	-0.057 (0.011) 0.000***	0.012 (0.020) 0.548	-0.006 (0.068) 0.931
MTB	0.003 (0.001) 0.034**	0.0003 (0.001) 0.850	-0.004 (0.005) 0.436
SIZE	0.005 (0.002) 0.029**	-0.007 (0.003) 0.009***	0.037 (0.041) 0.374
Cons	-0.075 (0.048) 0.121	0.240 (0.057) 0.000***	-0.863 (0.913) 0.346
Year dummy	Included	Included	Included
n	1,181	616	565
R2	0.087	0.138	0.259
R2 adjusted	0.079	0.124	0.246
R-sq
whithin			0.286
between			0.108
overall			0.137
Panel B: Tests for choosing the estimator
BP	0.00 [1.000]	1.60 [0.103]	6.20 [0.006]
F Chow			1.32 [0.011]**
Hausman’			24.51 [0.006]***
Hausman’’			8.16 [0.000]***
SH			18.42 [0.048]**
Estimator chosen	Pooled robust	Pooled robust	Clustered fixed effect

Note. *, **, *** indicates statistical significance at levels of 0.10, 0.05 and 0.01, respectively. Variables are defined in Appendix A. Below the coefficients and standard errors, these in parentheses, the p values are presented. BP is the Lagrange multiplier of Breusch-Pagan test testing H0: POLS and H1 model random effects. F Chow test confirms the BP test. Hausman’ is the Hausman test (H0: random effects; H1: fixed effects). Hausman” is the robust test Hausman described by Hoechle (2007) and Cameron and Trivedi (2010). SH is the Sargan test proposed by Schaffer and Stillmam (2016), which considers the robust standard errors. This test confirms the Hausman’.

Although the expected direction of causality between CSR and proxies for FRQ, following ethical, political, integrative, and instrumental theories, would support the formulation of the hypotheses of transparent financial reporting and retreatment, we find no relationship between CSR disclosures and manipulation proxies. This suggests that sustainable companies not explain lower or higher levels of DA. In this scenario, we argued that there does not exist an ex-ante incentive on the part of executives to use CSR to conceal or reduce EM in companies. Coffee Jr. (2003)approximately 10% of all listed companies in the United States announced at least one financial statement restatement. The stock prices of restating companies declined 10% on average on the announcement of these restatements, with restating firms losing over $100 billion in market capitalization over a short three day trading window surrounding these restatements. Such generalized financial irregularity requires a more generic causal explanation than can be found in the facts of Enron, WorldCom or other specific case histories. Several different explanations are plausible, each focusing on a different actor (but none giving primary attention to the board of directors discusses the possible lack of significance in this relationship, since it opposes traditionally researched approaches. The author advocates that it is “perverse” incentives, not a decline in ethics, that cause scandals. Thus, factors related to fraud and professional misconduct would be related to EM. Chih et al. (2008) present arguments and identify them as an institutional hypothesis, where factors unrelated to business ethics interfere with the direction and quantity of DA, proxy for EM. This alternative hypothesis shows measures of financial and variable characteristics at the firm level can explain the extent to which companies conduct EM. In part, this was captured with the control variables inserted in the models tested in this research.

Explanations at the operational level can elucidate the lack of significance between CSR and EM. Initially, the lack of consensus in the definition of sustainability is reflected in its operationalization. For example, several authors used different measurements of the variables and econometric estimators presenting different results. Second, the relationship between CSR and EM can be driven by other factors such as political costs suggested by Yip et al. (2011). Third, the underlying idea of our model is to measure managerial opportunism in the use of CSR disclosures. Nonetheless, we admit the possibility that this effect will be reduced in companies with strong mechanisms of corporate governance. This measure, neglected in this study, could reduce the extent of the use of CSR disclosure in EM.

Despite the limitations and lack of empirical support of the postulated hypotheses, our results resemble recent research. Kolsi and Attayah (2018) and Laksmi and Kamila (2018) found no statistically significant relationship between CSR and EM when analyzing firms from the United Arab Emirates and Indonesia, respectively.

Regarding the control variables, the negative relationship is observed between leverage and DA (coefficient -0.057, p < 0.01). One explanation for this is that the highest external monitoring, for example due to the greater presence of capital providers, can inhibit EM practices. This relationship conforms to Chih et al. (2008) and Litt et al. (2013).

Growth’s opportunity marginal effect, MTB (coefficient 0.003, p < 0.05), indicates that highest market-to-book ratio can manipulate the accumulations, regardless of their direction. This evidence partially aligns with results from Cho and Chun (2015) and Kim et al. (2012).

The SIZE variable positively correlates with DA (coefficient 0.005, p < 0.05) and negatively with positive discretionary accruals, POS.DA (coefficient -0.007, p < 0.01). Thus, larger companies are likely to practice EM, both to increase and to reduce current earnings. This is consistent with results from Bozzolan et al. (2015), Kim et al. (2012), and Muttakin et al. (2015).

In set, the empirical results do not support the hypotheses raised. However, they are consistent with recent research (e.g. Kolsi & Attayah, 2018; Laksmi & Kamila, 2018). Theoretical explanations regarding the sustainability concept, and operational can justify the lack of statistical significance of the main interest variable.

Measurements modified from the Jones model (1991). Literature disagrees on a single measurement for FRQ (Dechow, Ge, & Schrand, 2010). Thus, we re-estimated this by applying Model 3 with Jones modifications, as suggested by Dechow et al. (1995) and Kothari et al. (2005). Table 5 illustrates the results.

Table 5

Panel A: Model 3 re-estimated
	DA.	DA.	POS.DA.	POS.DA.	NEG.DA.	NEG.DA.
	1995	2005	1995	2005	1995	2005
D.GRI	-0.004 (0.008) 0.592	-0.004 (0.008) 0.653	0.002 (0.009) 0.823	0.004 (0.009) 0.675	0.004 (0.010) 0.705	0.008 (0.020) 0.694
ROA	0.089 (0.060) 0.136	0.092 (0.062) 0.135	0.122 (0.085) 0.151	0.141 (0.092) 0.124	-0.011 (0.068) 0.875	-0.051 (0.102) 0.622
ROE	-0.001 (0.013) 0.959	-0.001 (0.013) 0.952	0.005 (0.020) 0.794	0.006 (0.021) 0.772	-0.004 (0.008) 0.602	-0.013 (0.015) 0.385
LEV	-0.056 (0.011) 0.000***	-0.054 (0.011) 0.000***	0.021 (0.023) 0.372	0.021 (0.023) 0.375	-0.055 (0.016) 0.001***	-0.006 (0.070) 0.932
MTB	0.004 (0.002) 0.009***	0.004 (0.001) 0.015**	-0.0003 (0.001) 0.823	-0.0005 (0.001) 0.751	0.002 (0.002) 0.197	-0.002 (0.005) 0.679
SIZE	0.005 (0.002) 0.017**	0.006 (0.002) 0.016**	-0.009 (0.003) 0.002***	-0.009 (0.003) 0.003***	0.009 (0.003) 0.001***	0.042 (0.041) 0.305
Cons	-0.088 (0.051) 0.083*	-0.091 (0.050) 0.069*	0.290 (0.066) 0.000***	0.287 (0.067) 0.000***	-0.255 (0.060) 0.000***	-0.984 (0.911) 0.281
Year dummy	Included	Included	Included	Included	Included	Included
n	1,181	1,181	602	589	579	592
R2	0.087	0.072	0.141	0.159	0.253	0.245
R2 adjusted	0.080	0.064	0.127	0.145	0.240	0.232
R-sq
whithin			0.136	0.168	0.265	0.250
between			0.129	0.131	0.256	0.125
overall			0.140	0.158	0.253	0.137
Panel B: Tests for choosing the estimator
BP	0.00 [1.000]	0.00 [1.000]	3.18[0.037]**	2.580[0.054]*	7.49[0.003]***	6.84[0.005]***
F Chow			1.48[0.000]***	1.47[0.001]***	1.34[0.007]***	1.33[0.009]***
Hausman’			11.50[0.320]	13.72[0.186]	14.54[0.150]	20.55[0.025]**
Hausman’’			3.351[0.000]***	3.59[0.000]***	6.72[0.000]***	8.35[0.000]***
SH			9.05[0.528]	9.97[0.443]	10.12[0.430]	20.06[0.003]***
	Pooled robust	Pooled robust	Clustered random effect	Clustered random effect	Clustered random effect	Clustered fixed effect

Note. *, **, *** indicates statistical significance at levels of 0.10, 0.05 and 0.01, respectively. Below the coefficients and standard errors, these in parentheses, the p values are presented. Variables are defined in Appendix A. BP is the Lagrange multiplier of Breusch-Pagan test testing H0: POLS and H1 model random effects. F Chow test confirms the BP test. Hausman’ is the Hausman test (H0: random effects; H1: fixed effects). Hausman” is the robust test Hausman described by Hoechle (2007) and Cameron and Trivedi (2010). SH is the Sargan test proposed by Schaffer and Stillmam (2016), which considers the robust standard errors. This test confirms the Hausman’.

The re-estimates’ of Model 3 didn’t alter the interpretations of the results. The main interest variable (D.GRI) didn’t present statistical significance, preventing the acceptance to the hypotheses raised. As discussed in the section 4.2., there appears to be a signal that EM is influenced by firm-level financial and institutional variables.

The relationship between CSR and Real Activities Manipulation (RAM). Chih et al. (2008) report that the relationship between CSR and the accounting information’s quality depends on the extent of the EM adopted. Cho and Chun (2015) and Kim et al. (2012) investigated EM motivations through actual activities, with CSR as the variable of interest. To examine whether our results are robust to this alternative measure of FRQ, we re-estimated Model 3, replacing DA, for the magnitude of manipulations of operating activities, calculated according to Roychowdhury (2006).

The results follow those presented in Tables 4 and 5. Thus, it was not possible to validate the hypothesis raised in the research, as justified in sections 4.2.

Analysis of comparable sample. Kim et al. (2012) employ additional essays using a comparable sample. This research adopts a similar technique, by including a sample of comparable companies that integrate the Corporate Sustainability Index (ISE). Calculated since 2005 by B3, ISEs aims to align sustainable development investments and stimulate corporate ethical responsibility to compel companies to provide better-quality sustainable information. We used a control group to form the comparable sample, which enables a comparison of the degrees of FRQ while strengthening the primary sample’s findings.

The comparable sample is then constructed as a branch from the main sample, and we selected and analyzed the samples’ respective comparable companies by considering the same companies that integrated the ISE in the period. These companies had similar values of total assets and belong to the same economic sector, with similar financial activities. We applied samples proportional to the types of EM to determine the results are robust.

The results presented in Appendix C re-affirm the previous findings, where there is lack of statistical significance for variable of main interest (D.GRI).