*3.3. Control Variables*

To control for company characteristics that are expected to have a direct or an indirect impact on the cost of capital and for firm risk profile, our review of the cost of capital literature suggests that five factors are most likely to affect the cost of capital. The following factors are used in our study.

The first is financial leverage (LVRG) measured by "the ratio of long-term debt to total assets". It represents a source of funding that a corporation needs to ensure its continuity. We include leverage in the model for two reasons. *First*, increasing leverage is the result of additional scrutiny from financial institutions, which affects the cost of capital [73]. *Second*, leverage has a direct impact on the cost of capital since it relates to financial distress [68]. Therefore, high leverage is expected to relate to a higher cost of capital. Previous studies [21,33,44,74,75] have shown that leverage positively relates to the cost of capital.

The second is firm size (SIZE) measured by the "logarithm of total assets". Previous studies [14,33,44,72] have shown that firm size has a negative and significant impact on COK; small firms are perceived as riskier than their larger peers. Large firms attract more media and analyst coverage [76], which could mitigate information asymmetry risk since they dispose of more information to disclose than their smaller peers [77].

The third is firm profitability (PROFI), which is considered a key determinant of future investment. Then, higher expected profitability will lessen frictions the firm faces in the market. We include return on assets approximated by the ratio "operating income before depreciation divided by total assets". Firms with a high return on assets enjoy a low cost of capital [78].

The fourth is information asymmetry and agency conflicts. Information asymmetry (INF\_ASY) is approximated by the market-to-book ratio measured by the market value of equity/the book value of equity. This ratio depends on the extent to which a firm's returns on existing assets and expected future investments exceed its required rate of return on equity [79]. We argue that the larger the MB ratio is, the larger information asymmetry between the market and the firm [79]. Firms with a high market-to-book ratio have a higher cost of capital [14]. Agency conflicts are measured by free cash flows (FCF), approximated by the ratio "[operating income before depreciation − interest expense − total taxes − dividends]/total assets" [79,80]. As indicated in the literature, a higher proportion of free cash flows can lead to agency conflicts between managers and shareholders [79–82]. We expect a positive relationship between firms' free cash flows and their costs of capital.

Finally, firm age (FIRM\_AGE) was also added to the research model because, as predicted by firm lifecycle theory, the cost of capital tends to fall for older firms [83]. We expect a negative relationship between the firm age and the cost of capital

Table 1 summarizes the measurements of variables, their definitions, and their expected signs.


**Table 1.** Variables' definitions.

### *3.4. Empirical Methodology*

Our empirical methodology aims at checking the above hypotheses and whether CSR disclosure leads to cheaper capital access or not over time. Thus, we examine the effect of each of the three main dimensions of ESG disclosure on the cost of capital. We run the following regressions:

$$\text{COK}\_{\text{i,t}} = \beta\_0 + \beta\_1 \text{ENV\\_SC}\_{\text{i,t}} + \text{CONTROL VALBLES} + \text{YEARS} + \text{INDUSTRIS} + \varepsilon\_{\text{i,t}}.\tag{1}$$

$$\text{COK}\_{\text{i,t}} = \beta\_0 + \beta\_1 \text{SCC\\_SC}\_{\text{i,t}} + \text{CONTROL VALBLES} + \text{YEARS} + \text{INDUSTRIES} + \varepsilon\_{\text{i,t}}.\tag{2}$$

$$\text{COK}\_{i,t} = \beta\_0 + \beta\_1 \text{GOV\\_SC}\_{i,t} + \text{CONTROL VALES} + \text{YEARS} + \text{INDUSTRIS} + \varepsilon\_{i,t} \tag{3}$$

We use panel data methodology in our analysis. Indeed, unlike pooled regression, which neglects the time dimension and treats the data as cross-sectional by pooling across years [84], panel data models test group (individual-specific) effects, time effects, or both to deal with heterogeneity or individual effects that may or may be unobserved [85,86].

Furthermore, the panel data approach has several advantages over the analysis of individual time series or cross-sectional data. It gives more information with less collinearity among the variables, more degrees of freedom, and more efficiency, and it can control for individual heterogeneity [87,88]. Both fixed and random effects estimators were applied and distinguished on the basis of the Hausman test, which suggested that the random effects specification was more appropriate.

#### **4. Empirical Results**

#### *4.1. Descriptive Statistics*

Table 2 reports the descriptive statistics. First, the mean of the cost of capital (COK) is 0.076. The results of the ESG scores show that American firms seem to be more focused on environmental disclosure (61.3% of firms exceed the industry median ENV\_SC), followed by governance disclosure (half of the firms exceed the industry median GOV\_SC) and then social disclosure (31.3% of firms exceed the industry median SOC\_SC).



For the control variables, shows that the average firm year in our sample has a financial leverage (LVRG) of 0.264, a firm size (SIZE) of 7.616, a return on assets (PROFI) of 7.6%, a market-to-book ratio (INF\_ASY) of 1.931, a free cash flow (FCF) of 0.107, and a firm age (Firm\_AGE) of 81.03.

#### *4.2. Multivariate Analysis: Impact of ESG Disclosure on the Cost of Capital*

We examine the impact of the individual dimension of CSR disclosure on COK. Panels (A), (B), and (C) of Table 3 summarize the effect of ENV\_SC, SOC\_SC, and GOV\_SC on COK, respectively. For the effect of each dimension of ESG disclosure, the effect of the social dimension is the most important effect. The increase in the cost of capital is mostly explained by this dimension. By attributing a rank to the effect of each of the CSR disclosure dimensions on the cost of capital, the social dimension would be in the lead, followed by the governance dimension and then the environmental dimension, which has little influence.

**Table 3.** Regression results of ESG disclosure and cost of capital.



**Table 3.** *Cont.*

\*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1.

Table 3 Panel A reports the main results of the OLS regression of the COK-ENV\_SC relationship. We note that only in the first four columns of Table 3 (Panel A), the coefficient of environmental disclosure is negative. However, in the subsequent columns, the coefficient becomes statistically insignificant. This leads us to reject hypotheses H1a and H1b. Our estimations are very interesting since they put in evidence that the effect of environmental disclosure on firm risk changes over time. Legitimacy theory explains our results for the first four windows of years. There is a huge appeal from stakeholders and a positive reaction from the financial market for green products and environmental practices. This ecological commitment would then be translated into a competitive advantage since managers would use environmental disclosure to convey a proactive environmental image to maintain or increase the legitimacy of their firms. Therefore, given the growing stakeholder interest and media focus on environmental activities, it represents the perfect opportunity for CEOs to build a strong positive picture for stakeholders and society. The CEO, therefore, incites the firm into engaging in environmental disclosure. Such disclosure is a good way of increasing the legitimacy of the firm and consequently decreasing firm risk. Moreover, in the last decade, the request of investors to disclose environmental information (climate change, green environment, etc.) has become more pronounced. This awareness has been translated by a new environmental regulatory disclosure framework in the United States. More precisely, in 2010 the SEC issued guidance to help firms to assess the effect of their mandatory disclosure on climate change. Moreover, in 2011 the Environmental Protection Agency (EPA) issued requirements for disclosing some environmental activities. Consequently, firms have complied with this requirement and thus have gained the trust of stakeholders [50], especially equity capital providers and debtholders, leading to a decrease in the cost of capital. Thus, the compliance of most firms with these requirements could neutralize the effect of environmental disclosure on firm risk during the last two windows of years. Finally, with the improvement of consciousness on environmental issues and disclosure for government, companies, and investors, we can conclude that environmental

disclosure effect on firm risk changes over time. For the control variables, a positive effect is observed for financial leverage in almost all sub-periods except 2011–2017. However, a negative effect is observed over time for PROFI and firm size. Information asymmetry (INF\_ASY) is significant only for the 2011–2014 period. Free cash flow exerts a positive effect on the cost of capital during the 2011–2014, 2011–2015, and 2011–2018 periods. Finally, a negative effect is observed for firm age during the 2011–2017, 2011–2018, and 2011–2019 periods.

Hypothesis 2a (2b) predicts that the social disclosure coefficient will initially be positive (negative) and increasingly positive (negative). Results are reported in Table 3 Panel B. We note a significant effect on firm risk for sub-periods starting with the first period, 2011–2014, with a more significant effect in the two last periods. Over time, social disclosure becomes increasingly more favorable for increasing firm risk, which lends support to H2a. Although in the US there is a limited number of mandated requirements for corporate social disclosure, the California Transparency in Supply Chains Act of 2010 (CTSCA) is one among these requirements intended for manufacturing and retail firms doing business in California (with worldwide sales over \$100 million) and having to address issues related to slavery and human trafficking issues in corporate supply chains. This act took effect on 1 January 2012. Although this act is specific for some firms, it has generated a great deal of awareness about social disclosure among investors and the stakeholders of other sectors. Even though this wave of awareness is increasing, the costs–benefits trade-off of disclosing social information explains our results. Thus, some firms' huge recourse to bank financing somehow calls off the role of social disclosure and therefore increases firm risk. Our results are in line with the assumptions of agency theory. This leads us to conclude that the more the companies improve their social disclosure, the less likely they attract debtholders and equity capital providers, thus increasing the cost of the capital over time. From 2018, we observe the strengthening effect of social disclosure on the cost of capital; a more significant relationship is observed. The main reason that explains this result is that social disclosure has evolved over time. To reduce the cost of capital, firms have to go beyond social disclosure by focusing on the glocal drivers of CSR as presented in the theoretical framework. For the control variables, financial leverage has a positive effect on firm risk during the 2011–2015, 2011–2016 and 2011–2017 periods. The effect of firm size is negative and more pronounced over time. Our results show also that there is a negative and significant relationship between the firm's accounting performance (PROFI) and the cost of capital, but in this case, the effect remains stable over time. The coefficient of the market-to-book ratio is significant only for the first four windows of years. However, the free cash flow variable is not significant for all columns of Panel B of Table 3. The effect of firm age remains almost the same as in the regression of environmental disclosure: negative and significant for the last three windows of years.

Turning now to the governance disclosure effect, our results (Table 3 Panel C) show a negative and significant effect on firm risk for the first two windows of years. These results are significantly different from those of the subsequent sub-periods. We see that until 2015, the governance disclosure coefficient is still negative, and the negative coefficient in the first two sub-periods becomes positive starting from the 2011–2016 period, which supports. H3a More precisely, until 2015, governance disclosure decreases firm risk. From 2016, we observe an opposite effect: the firm risk increases. Our estimations are very interesting since they put in evidence that the effect of governance disclosure on firm risk changes over time. Following the SEC disclosure rules on diversity and other governance matters, firms' governance practices converge quickly to such requirements. Thus, firms move from the first level of compliance to a higher level in few years. As observed in our study, until 2015, the enhancement of governance disclosure decreased the cost of capital. Consequently, governance disclosure is a relevant issue for the key stakeholders. Beyond this period, an opposite effect is observed. This result is mainly explained by downturns in governance reforms during the last years. Thus, as firms comply with the previous requirement, the key stakeholders (debtholders and equity capital providers)

are not sensitive to the enhancement of governance disclosure when negotiating the cost of their funds. Overall, our results highlight the dynamicity of governance disclosure over time. For the control variables, our results show that the coefficient of the financial LVRG variable is significant only during the two first windows of years: there is a positive relationship between the company's leverage and cost of capital during the 2011–2014 and 2011–2015 periods. The effect of firm size is negative and more pronounced over time. Our results show that there is a negative and significant relationship between the firm's accounting performance (PROFI) and the cost of capital, and this effect becomes more favorable over time. The coefficients of the market-to-book ratio are not significant over time. The negative coefficient of the free cash flow (FCF) variable over time denies the assumption that firms having a high proportion of FCF would be exposed to more conflicts of interest and hence higher incentives for managers to engage in opportunistic behavior, thus increasing cost of capital. Finally, the coefficient of firm age is negative and significant during the 2011–2017 and 2011–2018 periods.

We can conclude that over time, internally oriented CSR activities such as social and governance disclosure exerts a more acute undesirable effect on firm risk than externally oriented CSR activities such as environmental disclosure.

#### *4.3. Robustness Checks*

To ensure the robustness of our primary findings, several robustness tests are conducted. One of the most prominent issues in related papers that deal with CSR disclosure is endogeneity bias. *First*, for reverse causality in our regression equations, we examine the impact of ESG score on the future cost of capital (Columns (1) to (3) of Table 4) by taking the dependent variables at time (t + 1). Our findings based on regression reaffirm our main findings. (We report results for the full sample period. In unreported results, we have also rerun our specifications using the sub-periods.)

**Table 4.** Robustness check results: ESG disclosure and cost of capital.


\*\*\* *p* < 0.01, \*\* *p* < 0.05, \* *p* < 0.1.

*Second*, to control the potential concerns regarding endogeneities, simultaneities, and firm-specific heterogeneities in our main regressions, we employ the system Generalized Method of Moments (GMM) in re-estimating our results. We report the results in Table 4 (Columns (4) to (6)). Our results meet the threshold of the standard tests for the system GMM that AR (2) tests for second-order autocorrelation and Hansen tests for instrumental validity. Furthermore, our findings in these models corroborate the main findings and highlight that our results are robust to potential spurious correlations that may arise from heterogeneities or endogeneities.
