Financial Reports and Social Capital

CSR is a measure of how socially responsible a firm is—it is not a measure of the intrinsic norms of the managers to behave honestly and the propensity to honor obligations—social capital measures these qualities. Also, higher CSR can sometimes be a cover for misbehavior (Hemingway and Maclagan 2004; Prior et al. 2008; Petrovits 2006).

These are some of the recent studies (e.g., Grullon et al. 2010; Hilary and Hui 2009; McGuire et al. 2012; Jha and Chen 2015; Jha and Cox 2015) that examine the role of either religiosity or social capital in financial decisions. In each of these studies the culture of where the firm is headquartered is used as measure for the culture of the managers. These studies provide detailed discussions on how the culture of the place of domicile and that of a firm’s manager are congruent. For brevity, I refer to these papers and do not present that argument in this paper.

I check the inflation factor (VIF) for multicollinearity among the independent variables in each OLS test on the main hypotheses. The largest VIF, 3.96, is much <10. This VIF shows that multicollinearity is not a problem.

I also verify that the results are robust when standard errors are double clustered in the following ways: clustering by firm and county, firm and industry, firm and year, county and year, and county and industry.

The fact that the GAO, AA, SCAC, and the AAERs omit a large faction of events they purport to capture and double count some of them creates type 1 and type 2 errors—this problem might compromise the validity of any tests using these data sources to examine the propensity to commit fraud.

The data and descriptions on how social capital indexes are constructed are available at: http://​aese.​psu.​edu/​nercrd/​community/​tools/​social-capital.

The results continue to hold if I remove religious organization.

The eigenvalues of the first components for these years are 1.80, 2.06, 1.94, and 1.80 respectively. The eigenvalues of the other components are less than one except in 2009 when the second component has an eigenvalue of 1.03.

The industries that I remove have the following Fama–French 48-digit Industry classifications: 3 (Soda: Candy and Beer), 5 (Smoke: Tobacco Products), 26 (Guns: Defense), and 29 (Coal).

Although untabulated, I also verify that the results continue to hold when I conduct a county-year regression. That is, I calculate the median of all variables at the county-year level and test if the results are robust.

In the main model, I do not control for ethnic diversity, literacy, and income inequality from the county where the firm is headquartered because these variables are highly correlated with social capital and are also its possible determinants. Whether these variables have a direct effect on financial reporting is also unclear, but quite likely they affect social capital, which in turn affects financial reporting. Therefore, the inclusion of these variables can likely take away from what might be the effect of social capital. Regardless, even when I control for these variables, I continue to find that, ceteris paribus, higher social capital has an association with better quality financial reports. However, as expected the significance levels are slightly lower. For brevity, I do not report these results.

Yu (2008) finds a negative association between analysts following and discretionary accruals, and Jiang et al. (2010) find a positive association between cash flow volatility and discretionary accruals.

Matching based on propensity scores constructed using these firm level variables produces a matched sample with the least bias—that is, the treated and the matched sample are the most alike. However, the results are robust when I construct propensity scores based on the control variable specified in Eq. 1.

These dates are available in the FSR database compiled by Karpoff et al. (2012).

Consider the following example: in the presence of a strict teacher all students are likely to behave. But in the absence of a strict teacher, naughty children are more likely to misbehave. Put differently, the effects of naughtiness (intrinsic nature) are more salient when the disciplinarian (external monitoring) is weak.

I also examine if the effect of social capital is stronger for firms that are further away from SEC. The idea is that the SEC’s enforcement might be weaker for firms headquartered further away from its office. To do so, I split the sample into two groups based on the median distance from the SEC. The results are qualitatively consistent with the idea that when external monitoring is weak, the effect of social capital is stronger. I find that the coefficient for social capital is negative and significant for firms that are further away, but nonsignificant for those close. However, there is no statistical difference. For brevity, I do not report the results.

I also find that qualitatively the effect of social capital is stronger for firms that have a higher g-index (i.e., those that are poorly monitored by their board of directors). I conduct this analysis by dividing the sample into two groups based on the median level of the g-index. I find that for the group with a high g-index the effect of social capital is large, negative, and significant but for those with a low value for the g-index, the coefficient is not significant. For brevity, I do not report the results.

The results are also similar when I split the sample based on the median level of analysts. For firms that have no analysts following, the effect of social capital on financial fraud is much stronger, as expected. It is nonsignificant for the group that has analysts following.

They conduct a similar analysis to examine if the social capital reduces the price of bank loan contracting for firms that move to a higher social capital region, compared to those that do not.

The results are similar when I use the performance adjusted discretionary accruals as suggested in Kothari et al. (2005).

Jha and Chen (2015) conduct a similar exercise with audit fees and find that in a multivariate framework, the effect of social capital hardly changes when religiosity is removed as one of the control variables in a regression where the dependent variable is the natural logarithm of the audit fees.

Column 1 presents the coefficients when both SocialCapital and Religiosity are included (it is the same as Column 1 of Panel C in Table 1). I present it here again to make it easier to compare.

Neither a likelihood ratio test as in Allison (1999) nor a test using an interaction term for a dummy of the high social capital and religiosity in a LPM shows that the difference is statistically significant.

The control variables include SocialCapital. But the results are similar if I exclude SocialCapital.