CSR Disclosure Breadth and Earnings Management: Evidence Consistent with Opportunistic Reporting in an Emerging Market

Abstract

This study examines the contemporaneous association between earnings management (EM) and corporate social responsibility (CSR) disclosure breadth in an emerging-market setting. Using a balanced panel of 136 manufacturing firms listed on the Tehran Stock Exchange across eight annual panel periods (2017–2024; 1088 firm-year observations), CSR disclosure is measured with a 33-item binary content-analysis index with explicit coding boundaries. The primary Kothari-style performance-adjusted signed abnormal-accrual measure is positively associated with disclosure breadth (β = 0.198, p = 0.031). A secondary modified Dechow–Dichev-style signed accrual–cash-flow residual sensitivity is positive but marginal (β = 0.285, p = 0.060), and real earnings management is positive but marginal in the beta-regression specification (β = 0.254, p = 0.071). In firm- and year-fixed-effects OLS, AEM-K remains positive and significant, AEM-DD remains marginal, and REM is not statistically significant. The evidence is associational rather than causal and is consistent with—but does not establish—an opportunistic-reporting interpretation in which financial-reporting opportunism can coexist with broader CSR disclosure.

1. Introduction

Earnings management (EM) arises when managers use accounting judgments or operating decisions to influence reported earnings, potentially obscuring the firm’s underlying economic performance. Such practices can impair the usefulness of financial reports, complicate investors’ assessments, and weaken confidence in reported information (Alhadab et al., 2015; Graham et al., 2005). EM has also been linked to subsequent stock-price crash risk and debt-contracting outcomes (Francis et al., 2016; Y. S. Kim et al., 2021), underscoring consequences beyond the reported earnings number itself.

The reputational consequences of EM create incentives for managers to shape stakeholders’ perceptions through other communication channels. One possible response is to expand the visibility and breadth of CSR-related disclosure when the credibility of financial reporting is under pressure (Prior et al., 2008). The present study examines whether EM and the extent of CSR disclosure are associated within the same firm-year.

The distinction between CSR activity and CSR disclosure is central to this question. Disclosure may reflect substantive social responsibility, but it may also function as strategic communication. Firms can use charitable, environmental, or stakeholder-oriented reporting to shape reputation and stakeholder perceptions (Agarwal et al., 2021; Hossain & Alam, 2016; Shiu & Yang, 2017). Accordingly, a higher disclosure score cannot, by itself, be interpreted as stronger CSR performance.

Prior evidence is mixed. Some studies report a positive association between EM and CSR-related communication, consistent with the view that managers use CSR disclosure to offset reputational concerns (Habbash & Haddad, 2019; Prior et al., 2008). Others find no association (Hickman et al., 2020), whereas a further strand documents a negative relationship and interprets socially responsible conduct as incompatible with opportunistic reporting (Ajina et al., 2019; Dimitropoulos, 2022; Gonçalves et al., 2021; Gras-Gil et al., 2016; Y. Kim et al., 2012; Tran et al., 2022; Velte, 2020). These conflicting results suggest that the association may depend on managerial incentives, measurement choices, and institutional context.

Iran provides a useful emerging-market setting because CSR-related information is embedded in corporate narrative reporting, making communication choices especially relevant to a disclosure-based construct. Recent evidence from Tehran Stock Exchange firms links abnormal tone in board reports to CSR-related outcomes (Mahmoudi Yekebaghi et al., 2025), underscoring the relevance of narrative reporting features in this setting. Evidence from Iran can therefore complement findings from other emerging and developed markets.

This study makes three contributions. First, it tests an explicitly opportunistic cross-channel reporting mechanism in an emerging-market setting. Second, it uses transparent item-level coding and an explicitly bounded 33-item disclosure instrument to separate stakeholder-oriented disclosure from broad commercial activity, raw workforce counts, and duplicate content. Third, it examines whether the primary association is directionally consistent across a modified signed-residual sensitivity, a real-activities proxy, and firm- and year-fixed-effects OLS, while distinguishing directional consistency from statistical significance and treating all estimates as contemporaneous associations rather than causal effects.

The remainder of the paper develops the opportunistic reporting argument and its principal alternative interpretation, describes the sample and measurement procedures, reports the main and robustness results, discusses the findings in relation to prior studies, and then presents implications, conclusions, limitations, and directions for future research.

2. Literature and Hypothesis Development

2.1. Opportunistic Reporting Perspective

The central theoretical question is whether managerial discretion in financial reporting can extend to non-financial communication. The opportunistic perspective predicts that managers facing reputational costs from earnings management may increase the breadth or visibility of CSR disclosure to protect legitimacy, shape stakeholder perceptions, or redirect attention. The study therefore treats opportunistic reporting as the primary explanatory framework and uses ethical/stakeholder reasoning as an important competing interpretation rather than presenting all theories as equally determinative.

2.1.1. Agency and Legitimacy Management

From an agency perspective, earnings management and selective CSR communication can arise from the same underlying managerial discretion. Managers who use accounting choices or operating decisions to influence reported earnings may also have incentives to emphasize stakeholder-oriented information when financial-reporting credibility is under pressure (Jensen & Meckling, 1976; Prior et al., 2008). CSR disclosure can therefore serve a reputational-buffering or legitimacy-preserving function without implying that the underlying CSR performance changed in the same direction.

Legitimacy management gives this mechanism a stakeholder-facing interpretation. A firm exposed to reputational concerns can respond by increasing visible environmental, employee, customer, or community disclosure, thereby demonstrating conformity with socially valued expectations (Benites-Lazaro & Mello-Théry, 2017; Dowling & Pfeffer, 1975; Y. Hong & Andersen, 2011; Suchman, 1995). This does not make every disclosure opportunistic; rather, it identifies a theoretically coherent reason why higher EM and greater CSR disclosure may coexist within the same firm-year.

2.1.2. Selective Signaling and the Information Environment

Selective-signaling reasoning provides a complementary communication channel. CSR reporting can affect the information environment and perceived transparency, while managers retain discretion over which non-financial attributes to emphasize (Cespa & Cestone, 2007; Dhaliwal et al., 2011; Spence, 1978). When financial-reporting credibility is under pressure, broader CSR disclosure can therefore become strategically salient. The empirical design cannot observe managerial intent directly; it tests the predicted co-movement and treats motive as an interpretation rather than an observed variable.

2.1.3. Ethical and Stakeholder-Based Alternative

A competing ethical interpretation predicts the opposite relationship. Stakeholder theory holds that managers should account for parties affected by the firm, including employees, customers, investors, communities, and regulators (Freeman, 1984; Gray et al., 1997; Mitchell et al., 1997). Under this view, responsible corporate conduct and high-quality financial reporting are complementary, so firms with stronger stakeholder accountability should have weaker incentives to manage earnings.

Social-norm reasoning reaches a similar counterprediction: shared expectations concerning honesty, accountability, and responsible conduct can discourage opportunistic financial reporting (Akerlof, 1980; H. Hong & Kacperczyk, 2009; Romer, 1984). Prior empirical studies reporting negative or null EM–CSR relationships are therefore theoretically meaningful alternatives to the opportunistic prediction (Ajina et al., 2019; Dimitropoulos, 2022; Gonçalves et al., 2021; Gras-Gil et al., 2016; Hickman et al., 2020; Y. Kim et al., 2012; Tran et al., 2022; Velte, 2020).

The study nevertheless advances the positive hypothesis because its focal research question concerns opportunistic cross-channel reporting: whether firms exhibiting greater earnings-management behavior also disclose more CSR information in the same period. Agency, legitimacy-management, information-transfer, and selective-signaling mechanisms jointly provide the direct basis for that directional prediction; the ethical perspective remains a competing explanation against which the results are interpreted.

H1. 

Higher income-increasing abnormal accruals, as captured by the primary AEM-K measure, are positively associated with broader CSR disclosure.

3. Methodology

3.1. Sample and Study Period

The empirical dataset comprises eight annual panel periods from 2017 through 2024 and contains complete observations for the 136 sampled firms. EM and CSR are aligned within the same firm-year throughout the analysis; the annual panel periods are therefore not interpreted as a one-period lead structure.

The sample comprises manufacturing firms listed on the Tehran Stock Exchange (TSE) that met the following criteria throughout the study period:

• the firm was listed on the TSE before 2017;
• the firm used the common fiscal year-end applied in the sample;
• the firm did not change its principal activity or fiscal year during the study period;
• the firm was not a bank, financial institution, or investment company; and
• the required financial and disclosure data were available.

Applying these criteria produced a balanced panel of 136 firms and 1088 firm-year observations.

3.2. Data

The dataset was manually assembled from publicly available corporate disclosures and the authors’ archived research files. Financial variables were extracted from audited financial statements and related Tehran Stock Exchange/CODAL disclosures. CSR disclosure is computed directly from the item-level binary coding matrix used for the study. This item-level construction provides a transparent link between each disclosure criterion and the firm-year CSR score.

3.3. Research Model and Variable Measurement

The baseline specification used to test H1 is presented in Model (1).

log(μ_i,_t/(1 − μ_i,_t)) = α + β₁ EM_i,_t + Σβ_k Control_k,_i,_t + δ_t

(1)

In the beta regressions, let μ_i,_t = E(CSR_i,_t|X_i,_t). Because all observed CSR33 values lie strictly between 0 and 1, the conditional mean is modeled with a logit link. EM and CSR are aligned contemporaneously within each firm-year. The specifications include the stated firm-level controls and year fixed effects, with standard errors clustered by firm (136 clusters). Statistical estimation was performed in Python 3.13.5 using statsmodels 0.14.6.

3.3.1. Independent Variable: EM

EM is measured primarily through a Kothari-style performance-adjusted abnormal-accrual specification (Kothari et al., 2005), with real-activities earnings management used as an alternative behavioral proxy. Performance adjustment is implemented within Model (2) through the current-performance term. The signed Kothari residual is retained, so higher AEM-K values are interpreted as more income-increasing abnormal accruals rather than as the absolute magnitude of earnings management. The Dechow and Dichev (2002) implementation is retained only as a secondary modified signed accrual–cash-flow residual sensitivity; because it uses total accruals and firm-year signed residuals, it is not treated as the canonical Dechow–Dichev accrual-quality score or as a directional measure of opportunistic manipulation.

Accrual-Based Earnings Management (AEM)

The primary accrual-based proxy is AEM-K, a Kothari-style performance-adjusted signed abnormal-accrual residual. AEM-DD is reported only as a secondary modified Dechow–Dichev-style signed residual sensitivity rather than as a co-equal directional proxy for opportunistic earnings management.

• Kothari et al. (2005) Model (AEM-K)

Model (2) estimates expected total accruals by industry-year with current performance included in the specification. The signed regression residual is used as AEM-K; higher values represent more income-increasing abnormal accruals relative to the performance-adjusted expectation.

$${\displaystyle \frac{{\mathrm{TA}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{it}-1}}}={\mathsf{\beta }}_1{\displaystyle \frac{1}{{\mathrm{A}}_{\mathrm{it}-1}}}+{\mathsf{\beta }}_2{\displaystyle \frac{{\Delta \mathrm{REV}}_{\mathrm{it}}-{\Delta \mathrm{REC}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{it}-1}}}+{\mathsf{\beta }}_3{\displaystyle \frac{{\mathrm{PPE}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{it}-1}}}+{\mathsf{\beta }}_4{\displaystyle \frac{{\mathrm{NI}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{it}-1}}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(2)

In Model (2), TA is the difference between operating earnings and operating cash flow; ΔREV is the annual change in revenue; ΔREC is the annual change in receivables; PPE is net property, plant, and equipment; the current-performance term is NIit/Ai,t − 1 (the ROA adjustment used in the proxy specification); A denotes total assets; and ε is the regression residual.

2.

Modified Dechow and Dichev (2002) Signed Residual Sensitivity (AEM-DD)

$${\displaystyle \frac{{\mathrm{TA}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{it}-1}}}={\mathsf{\beta }}_0+{\mathsf{\beta }}_1{\displaystyle \frac{{\mathrm{CFO}}_{\mathrm{it}-1}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_2{\displaystyle \frac{{\mathrm{CFO}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_3{\displaystyle \frac{{\mathrm{CFO}}_{\mathrm{it}+1}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(3)

Model (3) maps scaled total accruals to operating cash flows at t − 1, t, and t + 1. Because this implementation uses total accruals and retains the signed firm-year residual, AEM-DD is reported only as a modified Dechow–Dichev-style residual sensitivity. It is not interpreted as the canonical residual-variability measure of accrual quality or as a directional measure of opportunistic manipulation.

Firms may also manage reported earnings through real operating decisions. REM changes actual activities—such as sales timing, production, or discretionary expenditure—rather than relying solely on accounting choices. It is introduced below as an alternative EM measure in the robustness analysis.

Dependent Variable: CSR Disclosure Breadth

CSR disclosure is treated as an observable reporting construct rather than as a direct measure of underlying CSR performance. Content analysis is widely used to transform narrative corporate reports into systematic disclosure measures, but the literature also shows that CSR measurement can vary in scope, dimensionality, and whether it captures disclosure extent, disclosure quality, or substantive practice (Crowther & Lauesen, 2017; Ehsan et al., 2018; Shukla et al., 2022; Soobaroyen et al., 2023). Ehsan et al. (2018) illustrate a multi-method CSR measurement approach in a developing-economy setting, while Shukla et al. (2022) review alternative CSR measures and underscore the importance of matching the instrument to the construct. Soobaroyen et al. (2023) further distinguish the extent and quality of CSR disclosure. Consistent with these distinctions, the present binary index measures disclosure breadth or extent: whether defined information is disclosed. It does not measure disclosure quality, assurance, credibility, or underlying CSR performance.

The study uses an unweighted 33-item binary content-analysis index organized across environmental issues, product responsibility, human resources, customers, community responsibilities, and energy. Each item receives 1 when the relevant disclosure is explicitly present and 0 otherwise. The item boundaries are deliberately conservative: ordinary product-development and commercial-quality activities, routine production shutdowns, standalone workforce counts, and duplicate health/safety content are not counted as separate CSR disclosures. This approach follows the broader content-analysis literature while adapting the operational categories to the reporting context (Crowther & Lauesen, 2017; Ehsan et al., 2018; Shukla et al., 2022; Soobaroyen et al., 2023). GRI 401 is used as a construct benchmark for employment-related boundaries rather than as a claim that the study instrument is GRI-compliant (Global Reporting Initiative, 2016).

CSR33_i,_t = (Σ_j=₁³³ d_j,_i,_t)/33, where d_j,_i,_t ∈ {0,1}

(4)

CSR33_i,_t denotes the CSR disclosure breadth index for firm i in annual panel observation t.

The numerator is the number of the 33 predefined disclosure items coded 1 for the firm-year report.

The denominator is fixed at 33 items, so the index ranges from 0 to 1. Higher values indicate broader disclosure across the defined checklist; they should not be interpreted as evidence of higher-quality CSR performance. The CSR Disclosure Checklist used in the content analysis is presented in Table 1. Appendix A reports the item-level criteria so that the measurement boundaries are transparent and reproducible.

Table 1. CSR Disclosure Checklist Used in the Content Analysis.

Retained Disclosure Items	Dimension
Pollution control; prevention of environmental damage; recycling/waste prevention; conservation of natural resources; environmental R&D; environmental policy/management; environmental projects; other environmental issues.	Environmental Issues (8)
Product safety and health.	Product Responsibility (1)
Salary/bonus and benefits; employee shares; retirement/termination benefits; occupational health and safety; training/development; sports/welfare; loans/insurance; morale/communication; other employee responsibilities.	Human Resources (9)
Complaints/satisfaction; responsible payment/credit support; after-sales facilities/services; responsiveness to customer needs; other customer responsibilities.	Customers (5)
Social investment; community activities; charitable donations/services; legal action/litigation responsibility; religious/cultural activities; other social responsibilities.	Community Responsibilities (6)
Energy conservation; development/exploration of new resources; use of new resources; other energy issues.	Energy (4)

Appendix A reports the item-level coding criteria and examples. The instrument is interpreted throughout as a measure of disclosure breadth, not as a score of substantive CSR performance or disclosure quality.

Inter-coder consistency was assessed on a validation subsample of 30 firm-year reports. Two human coders applied the same final 33-item codebook to the same reports, yielding 990 paired binary item decisions before reconciliation. Pre-reconciliation raw agreement was 80.4%, Krippendorff’s alpha for nominal binary coding was 0.581, and Cohen’s kappa was 0.581. Disagreements were then reviewed against the underlying source reports under the fixed item definitions; the reliability statistics reported here remain pre-reconciliation. These statistics provide a direct reproducibility check of the operational coding rules for the final CSR33 instrument.

Control Variables

Model (1) includes profitability (ROA), financial leverage (LEV), firm size (Size), sales growth (Growth), operating cash flow (CFO), the current ratio (Current), and market-to-book ratio (M/B) as controls. Year fixed effects are included in the beta specifications, and standard errors are clustered by firm. The fixed-effects OLS robustness models include both firm and year fixed effects with firm-clustered standard errors.

Table 2 presents the descriptive statistics for the variables used in the analysis. Robustness and sensitivity are assessed in three ways: (i) AEM-DD is retained as a secondary modified signed accrual–cash-flow residual sensitivity, (ii) REM is used as an alternative real-activities EM proxy, and (iii) firm- and year-fixed-effects OLS is estimated as an alternative panel specification.

Models were estimated in Python 3.13.5 using statsmodels 0.14.6. Firm-clustered covariance uses 136 firm clusters, and Supplementary File S1 provides the complete coefficient vectors and model diagnostics.

4. Findings

4.1. Descriptive Statistics

The descriptive statistics of the variables are presented in Table 2.

Table 2. Descriptive Statistics of the Variables.

Variable Name	Symbol	Mean	Median	Min	Max	Q1	Q3	SD
Performance-adjusted signed abnormal accruals (Kothari-style)	AEM-K	0.069	0.077	−0.487	0.565	−0.004	0.156	0.150
Modified DD-style signed residual sensitivity	AEM-DD	0.039	0.041	−0.491	0.475	−0.013	0.092	0.107
Real Earnings Management	REM	0.076	0.092	−0.573	0.533	0.011	0.167	0.161
CSR Disclosure Breadth Index (33 items)	CSR	0.498	0.485	0.121	0.879	0.455	0.545	0.111
Profitability	ROA	0.154	0.132	−0.400	0.499	0.043	0.254	0.149
Financial Leverage	LEV	0.519	0.512	0.031	1.537	0.359	0.664	0.218
Firm Size	Size	15.370	15.054	11.198	21.572	14.155	16.301	1.783
Sales Growth	Growth	0.357	0.374	−0.999	4.264	0.069	0.678	0.668
Cash Flow Ratio	CFO	0.120	0.106	−0.488	0.478	0.040	0.203	0.143
Current Ratio	Current	1.615	1.437	0.204	4.581	1.085	1.971	0.795
Market-to-Book	M/B	3.087	2.703	0.020	7.974	1.706	4.213	1.762

Table 2. Descriptive Statistics of the Variables.

Variable Name	Symbol	Mean	Median	Min	Max	Q1	Q3	SD
Performance-adjusted signed abnormal accruals (Kothari-style)	AEM-K	0.069	0.077	−0.487	0.565	−0.004	0.156	0.150
Modified DD-style signed residual sensitivity	AEM-DD	0.039	0.041	−0.491	0.475	−0.013	0.092	0.107
Real Earnings Management	REM	0.076	0.092	−0.573	0.533	0.011	0.167	0.161
CSR Disclosure Breadth Index (33 items)	CSR	0.498	0.485	0.121	0.879	0.455	0.545	0.111
Profitability	ROA	0.154	0.132	−0.400	0.499	0.043	0.254	0.149
Financial Leverage	LEV	0.519	0.512	0.031	1.537	0.359	0.664	0.218
Firm Size	Size	15.370	15.054	11.198	21.572	14.155	16.301	1.783
Sales Growth	Growth	0.357	0.374	−0.999	4.264	0.069	0.678	0.668
Cash Flow Ratio	CFO	0.120	0.106	−0.488	0.478	0.040	0.203	0.143
Current Ratio	Current	1.615	1.437	0.204	4.581	1.085	1.971	0.795
Market-to-Book	M/B	3.087	2.703	0.020	7.974	1.706	4.213	1.762

Table 2 summarizes the study variables. The 33-item CSR disclosure index has a mean of 0.498 (SD = 0.111), indicating substantial cross-sectional and temporal variation within the balanced panel. The descriptive statistics refer to the same 1088 firm-year observations used in the primary analysis.

4.2. Hypothesis Tests

Baseline Association

Table 3, Table 4 and Table 5 report coefficients for all seven controls together with firm-clustered uncertainty statistics and model-level fit diagnostics; fixed-effect inclusion is shown explicitly, and the complete coefficient vectors are provided in Supplementary File S1.

Table 3. Beta-Regression Results for AEM-K and the Modified AEM-DD Sensitivity.

Variable	Symbol	Primary AEM-K Model	Modified DD-Style Sensitivity
Focal measure	—	0.198 (0.092); p = 0.031	0.285 (0.152); p = 0.060
Profitability	ROA	−0.316 (0.123); p = 0.010	−0.311 (0.121); p = 0.010
Financial leverage	LEV	−0.110 (0.127); p = 0.384	−0.101 (0.126); p = 0.423
Firm size	Size	0.005 (0.009); p = 0.581	0.006 (0.010); p = 0.542
Sales growth	Growth	−0.016 (0.030); p = 0.590	−0.016 (0.030); p = 0.598
Market-to-book	M/B	0.002 (0.010); p = 0.806	0.002 (0.010); p = 0.822
Operating cash flow	CFO	0.161 (0.134); p = 0.229	0.178 (0.136); p = 0.189
Current ratio	Current	0.006 (0.029); p = 0.834	0.007 (0.029); p = 0.815
Intercept	Constant	−0.029 (0.182); p = 0.873	−0.039 (0.183); p = 0.832
Log precision parameter	log(φ)	2.903 (0.092); p < 0.001	2.903 (0.092); p < 0.001
Year fixed effects		Yes	Yes
Firm-clustered standard errors		Yes	Yes
Observations	N	1088	1088
Log likelihood		821.459	821.403
AIC		−1608.918	−1608.807
BIC		−1524.053	−1523.941
Cox–Snell pseudo-R2		0.0138	0.0137

Table 3 reports the beta-regression estimates for the primary AEM-K specification and the secondary modified Dechow–Dichev-style signed residual sensitivity, including all seven firm-level controls and key model diagnostics. The models include year fixed effects and firm-clustered standard errors; Supplementary File S1 provides the complete coefficient vectors, including the year effects. AEM-K is positive and significant at the 5% level (β = 0.198, p = 0.031), providing the primary empirical support for H1. AEM-DD is also positive but marginal at the 10% level (β = 0.285, p = 0.060); it is reported only as a secondary residual-based sensitivity result and is not treated as independent corroboration of H1 or as an accrual-quality score. Among the controls, ROA is negative and statistically significant in both models, while the remaining firm-level controls are not significant at the 5% level. The evidence therefore supports a positive contemporaneous association through the primary AEM-K specification, with secondary results interpreted more cautiously.

4.3. Robustness Analyses

The robustness analyses test a secondary modified signed accrual–cash-flow residual sensitivity, an alternative real-activities EM measure, and a fixed-effects panel estimator using the same 33-item CSR disclosure breadth outcome and the same firm-level control set.

4.3.1. Alternative EM Measure: REM

REM is used as an alternative proxy because managers can influence reported earnings through operating decisions as well as accrual choices (Cohen et al., 2008; Gunny, 2010; Liu et al., 2023; Zang, 2012). Following Roychowdhury (2006), the measure combines abnormal operating cash flows, abnormal production costs, and abnormal discretionary expenses estimated from industry-year models.

The REM proxy captures abnormal operating activity consistent with real-activities manipulation; it does not observe managerial intent directly. The component models estimate expected cash flows from operations, production costs, and discretionary expenses, and their residuals represent abnormal activity. These components are combined so that higher values indicate greater real-activities manipulation (Badertscher, 2011; Cohen et al., 2008; Roychowdhury, 2006; Zang, 2012).

The REM-CFO model estimates the normal level of operating cash flow, as shown in Model (5).

$${\displaystyle \frac{{\mathrm{CFO}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\displaystyle \frac{1}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_1{\displaystyle \frac{{\mathrm{Sales}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_2{\displaystyle \frac{∆{\mathrm{Sales}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(5)

CFO_it is operating cash flow for firm i in year t.

A_i,t−1 is total assets at the end of year t − 1.

Sales_it is net sales in year t.

ΔSales_it is the change in net sales from t − 1 to t.

The REM-PROD model estimates normal production costs, as shown in Model (6).

$${\displaystyle \frac{{\mathrm{PROD}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\displaystyle \frac{1}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_1{\displaystyle \frac{{\mathrm{Sales}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_2{\displaystyle \frac{∆{\mathrm{Sales}}_{\mathrm{it}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_3{\displaystyle \frac{∆{\mathrm{Sales}}_{\mathrm{i},\mathrm{t}-1}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\epsilon }}_{\mathrm{it}}$$

(6)

PROD_it is the sum of cost of goods sold and the change in inventory.

ΔSales_i,t−1 is the change in net sales from t − 2 to t − 1.

The REM-DISX model estimates normal discretionary expenditure, as shown in Model (7).

$${\displaystyle \frac{{\mathrm{DISX}}_{\mathrm{i},\mathrm{t}}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\displaystyle \frac{1}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\beta }}_1{\displaystyle \frac{{\mathrm{Sales}}_{\mathrm{i},\mathrm{t}-1}}{{\mathrm{A}}_{\mathrm{i},\mathrm{t}-1}}}+{\mathsf{\epsilon }}_{\mathrm{i},\mathrm{t}}$$

(7)

DISX_it denotes selling, general, and administrative expenditure.

For each component, the regression fitted value represents normal activity and the residual represents abnormal activity. Following Cohen et al. (2008), the aggregate REM measure adds abnormal production costs and subtracts abnormal operating cash flow and abnormal discretionary expenditure.

REM = (−1 × abnormal discretionary expenses) + abnormal production costs + (−1 ×
abnormal operating cash flow).

Table 4 reports the beta-regression estimates obtained when REM replaces accrual-based EM, retaining the same seven controls, year fixed effects, and firm-clustered standard errors. Complete coefficient vectors, including year effects, are provided in Supplementary File S1.

log(μ_i,_t/(1 − μ_i,_t)) = β₀ + β₁ REM_i,_t + Σβ_j Controls_j,_i,_t + δ_t

(8)

Table 4. Robustness Results Using Real Earnings Management (REM).

Variable	Coefficient	Cluster SE	Robust z	p-Value
Real earnings management	0.254	0.141	1.804	0.071
Profitability	−0.331	0.125	−2.645	0.008
Financial leverage	−0.106	0.127	−0.835	0.404
Firm size	0.006	0.009	0.600	0.549
Sales growth	−0.053	0.034	−1.559	0.119
Market-to-book	0.002	0.010	0.213	0.832
Operating cash flow	0.157	0.135	1.156	0.248
Current ratio	0.008	0.029	0.282	0.778
Intercept	−0.002	0.185	−0.009	0.993
Log precision parameter (log φ)	2.901	0.093	31.282	p < 0.001
Year fixed effects	Yes
Firm-clustered standard errors	Yes
Observations	1088
Log likelihood	820.677
AIC	−1607.354
BIC	−1522.489
Cox–Snell pseudo-R2	0.0124

Table 4. Robustness Results Using Real Earnings Management (REM).

Variable	Coefficient	Cluster SE	Robust z	p-Value
Real earnings management	0.254	0.141	1.804	0.071
Profitability	−0.331	0.125	−2.645	0.008
Financial leverage	−0.106	0.127	−0.835	0.404
Firm size	0.006	0.009	0.600	0.549
Sales growth	−0.053	0.034	−1.559	0.119
Market-to-book	0.002	0.010	0.213	0.832
Operating cash flow	0.157	0.135	1.156	0.248
Current ratio	0.008	0.029	0.282	0.778
Intercept	−0.002	0.185	−0.009	0.993
Log precision parameter (log φ)	2.901	0.093	31.282	p < 0.001
Year fixed effects	Yes
Firm-clustered standard errors	Yes
Observations	1088
Log likelihood	820.677
AIC	−1607.354
BIC	−1522.489
Cox–Snell pseudo-R2	0.0124

REM is positively associated with the CSR disclosure index (β = 0.254, p = 0.071). The estimate is marginal at the 10% level and is interpreted as secondary directional evidence rather than as a separate causal result. The complete control-variable coefficients and model-quality statistics are reported in Table 4.

4.3.2. Firm- and Year-Fixed-Effects OLS

Table 5 reports the firm- and year-fixed-effects OLS specifications with firm-clustered standard errors and all seven firm-level controls. AEM-K remains positive and significant at the 5% level (β = 0.052, p = 0.040), while AEM-DD is positive and marginal at the 10% level (β = 0.066, p = 0.057). The REM coefficient is positive but not statistically significant (β = 0.037, p = 0.324). Supplementary File S1 provides the complete coefficient vectors, including the firm and year effects. Across these specifications, the primary AEM-K result remains directionally stable, while the secondary sensitivities show weaker statistical support.

Table 5. Robustness Results Using Firm- and Year-Fixed-Effects OLS.

Variable	Symbol	AEM-K	AEM-DD	REM
Focal measure	—	0.052 (0.025); p = 0.040	0.066 (0.035); p = 0.057	0.037 (0.038); p = 0.324
Profitability	ROA	−0.102 (0.039); p = 0.009	−0.103 (0.039); p = 0.009	−0.106 (0.040); p = 0.007
Financial leverage	LEV	−0.058 (0.035); p = 0.102	−0.055 (0.035); p = 0.122	−0.055 (0.036); p = 0.122
Firm size	Size	0.019 (0.009); p = 0.037	0.020 (0.009); p = 0.027	0.018 (0.009); p = 0.053
Sales growth	Growth	−0.001 (0.007); p = 0.866	−0.001 (0.007); p = 0.893	−0.006 (0.008); p = 0.428
Market-to-book	M/B	−0.001 (0.002); p = 0.584	−0.001 (0.002); p = 0.561	−0.001 (0.002); p = 0.562
Operating cash flow	CFO	0.059 (0.029); p = 0.040	0.064 (0.029); p = 0.028	0.058 (0.030); p = 0.052
Current ratio	Current	−0.008 (0.007); p = 0.227	−0.008 (0.007); p = 0.238	−0.008 (0.007); p = 0.237
Intercept	Constant	0.298 (0.110); p = 0.007	0.281 (0.110); p = 0.011	0.308 (0.115); p = 0.007
Firm fixed effects		Yes	Yes	Yes
Year fixed effects		Yes	Yes	Yes
Firm-clustered standard errors		Yes	Yes	Yes
Observations	N	1088	1088	1088
R2		0.283	0.282	0.279
Adjusted R2		0.168	0.167	0.164
Model degrees of freedom		150	150	150
Residual degrees of freedom		937	937	937

Table 5. Robustness Results Using Firm- and Year-Fixed-Effects OLS.

Variable	Symbol	AEM-K	AEM-DD	REM
Focal measure	—	0.052 (0.025); p = 0.040	0.066 (0.035); p = 0.057	0.037 (0.038); p = 0.324
Profitability	ROA	−0.102 (0.039); p = 0.009	−0.103 (0.039); p = 0.009	−0.106 (0.040); p = 0.007
Financial leverage	LEV	−0.058 (0.035); p = 0.102	−0.055 (0.035); p = 0.122	−0.055 (0.036); p = 0.122
Firm size	Size	0.019 (0.009); p = 0.037	0.020 (0.009); p = 0.027	0.018 (0.009); p = 0.053
Sales growth	Growth	−0.001 (0.007); p = 0.866	−0.001 (0.007); p = 0.893	−0.006 (0.008); p = 0.428
Market-to-book	M/B	−0.001 (0.002); p = 0.584	−0.001 (0.002); p = 0.561	−0.001 (0.002); p = 0.562
Operating cash flow	CFO	0.059 (0.029); p = 0.040	0.064 (0.029); p = 0.028	0.058 (0.030); p = 0.052
Current ratio	Current	−0.008 (0.007); p = 0.227	−0.008 (0.007); p = 0.238	−0.008 (0.007); p = 0.237
Intercept	Constant	0.298 (0.110); p = 0.007	0.281 (0.110); p = 0.011	0.308 (0.115); p = 0.007
Firm fixed effects		Yes	Yes	Yes
Year fixed effects		Yes	Yes	Yes
Firm-clustered standard errors		Yes	Yes	Yes
Observations	N	1088	1088	1088
R2		0.283	0.282	0.279
Adjusted R2		0.168	0.167	0.164
Model degrees of freedom		150	150	150
Residual degrees of freedom		937	937	937

5. Discussion

The results show a positive contemporaneous association between the primary earnings-management measure and CSR disclosure breadth, while the strength of secondary evidence varies across specifications. AEM-K is positive and statistically significant at the 5% level. AEM-DD, reported only as a modified signed residual sensitivity, is positive and marginal at the 10% level, and REM is likewise positive and marginal in the beta-regression specification. In firm- and year-fixed-effects OLS, AEM-K retains the positive and significant pattern; AEM-DD remains marginal, whereas REM is not statistically significant. The evidence is therefore consistent with the opportunistic-reporting interpretation but does not establish managerial intent or show uniform significance across every estimator.

The positive AEM-K association is consistent with Habbash and Haddad (2019) and Prior et al. (2008), which report positive EM–CSR relationships, but contrasts with the significant negative corporate-environmental-disclosure–EM relationship reported by Gerged et al. (2020) in Kuwait. Differences in construct measurement, institutional setting, and reporting incentives should therefore be considered rather than treating prior evidence as uniform. The present pattern remains compatible with agency and legitimacy-management mechanisms in which managers expand visible non-financial communication when financial-reporting credibility is under pressure.

At the same time, the findings should be read against the substantial contrary literature. Hickman et al. (2020) report no association in their setting, while several studies document a negative EM–CSR relationship and interpret responsible conduct as inconsistent with opportunistic reporting (Ajina et al., 2019; Dimitropoulos, 2022; Gonçalves et al., 2021; Gras-Gil et al., 2016; Y. Kim et al., 2012; Tran et al., 2022; Velte, 2020). Differences in institutional oversight, disclosure regulation, construct measurement, and the distinction between CSR performance and disclosure breadth can therefore help explain why empirical signs differ across studies. The present results do not resolve those competing mechanisms universally; they identify the positive association in the sampled emerging-market context.

6. Practical and Policy Implications

The Iranian setting is practically relevant because the CSR construct analyzed here is derived from narrative corporate reporting rather than from an audited measure of CSR performance. Regulators and market institutions can improve decision usefulness by defining clearer disclosure categories, encouraging quantitative and externally verifiable indicators, and distinguishing disclosure breadth from evidence of actual CSR performance.

Investors, auditors, and other stakeholders should evaluate extensive CSR disclosure jointly with earnings quality rather than treating disclosure volume as independent evidence of responsible conduct. The results also imply that governance and assurance mechanisms aimed at financial-reporting integrity can be complemented by stronger verification of non-financial reporting.

7. Conclusions

This study finds a positive contemporaneous association between the primary Kothari-style performance-adjusted signed abnormal-accrual measure and CSR disclosure breadth in 136 Tehran Stock Exchange manufacturing firms. The modified signed-residual and real-activities beta specifications are also positive but weaker, and only AEM-K remains significant at the 5% level in firm- and year-fixed-effects OLS. The results are therefore consistent with opportunistic reporting as one plausible interpretation, while remaining associational and non-causal.

8. Limitations and Future Research

Several limitations define the scope of the conclusions. First, the sample is restricted to listed Iranian manufacturing firms, limiting generalization to other sectors and institutional settings. Second, the binary index measures disclosure breadth—whether predefined information is reported—not disclosure quality, assurance, credibility, or underlying CSR performance. Third, the study is observational and contemporaneous; it does not establish temporal precedence, managerial intent, or causality. Fourth, the inter-coder reliability assessment was conducted on a 30-report validation subsample rather than by double coding the entire historical panel. Pre-reconciliation agreement was 80.4%, with Krippendorff’s alpha = 0.581 and Cohen’s kappa = 0.581, indicating moderate coder consistency. Future research can extend full-sample dual coding, graded measures of disclosure quality and assurance, replication in other markets, and research designs better suited to causal identification.