Financial Reports and Social Capital

Abstract

I examine social capital’s impact on financial reports. Based on the social capital literature, I predict that the quality of the financial reports is higher when a firm is headquartered in a region with high social capital. Consistent with this prediction, I find that the firms that are headquartered in this type of region in the USA have a lower probability of committing fraud by misrepresenting financial information. Further, I find that the firms in regions with high social capital have lower levels of discretionary accruals and much more readable annual reports.

Appendices

Appendix 1

Dependent variables
FinclFraud	This is an indicator variable that equals one if it is included in the FSR database compiled by Karpoff et al. (2012). To be included in this database, the firm needs to be guilty of financial fraud. Source: Karpoff et al. (2012)
|DiscAccrual|	This variable is the absolute value of the discretionary accruals calculated by using the modified Jones (1991) model. To calculate the discretionary accruals, I first calculate the total accruals (ibc − (oancf − xidoc)) by using the cash flow approach as suggested by Hribar and Collins (2002). To remove the effect of the outliers, I remove the observations where the ratio of the total accruals to the assets is more than two standard deviations away. Following the steps of the modified Jones (1991) model, I use this regression for each industry year: $\frac{{TA_{it} }}{{Assets_{it - 1} }} = \beta_{0} \frac{1}{{Assets_{it - 1} }} + \beta_{1} \frac{{\Delta Sales_{it} }}{{Assets_{it - 1} }} + \beta_{2} \frac{{PPE_{it} }}{{Assets_{it - 1} }} + \varepsilon_{it}$
	The industry classification is based on the two-digit SIC code, and I require that there are at least 15 observations in each industry year. I also require that the firms be headquartered in the USA. The TA is the total accruals; Assets is the total assets (at); Sales is the total sales (sale); and the PPE is the gross plant, property, and equipment (ppegt). In the second step, I use the coefficient for the regression in the first step to calculate the discretionary accruals as follows: $\begin{aligned} & NDA_{it} = \left( {\hat{\beta }_{0} \frac{1}{{Assets_{it - 1} }} + \hat{\beta }_{1} \frac{{\Delta Sales_{it} - \Delta Receivables_{it} }}{{Assets_{it - 1} }} + \hat{\beta }_{2} \frac{{PPE_{it} }}{{Assets_{it - 1} }}} \right) \\ & DISC\_ACC_{it} = \frac{{TA_{it} }}{{Assets_{it - 1} }} - NDA_{it} \\ \end{aligned}$where the Receivables are the total receivables (rect). Source: COMPUSTAT
DechowDichevAQ	This variable is calculated following Francis et al. (2005). As in their study, I first calculate the residuals from the following regression for each industry year
	$\begin{aligned} TCA_{it} & = \hat{\beta }_{0} + \hat{\beta }_{1} \frac{{CFO_{it - 1} }}{{Assets_{it - 2} }} + \hat{\beta }_{2} \frac{{CFO_{it} }}{{Assets_{it - 1} }} + \hat{\beta }_{3} \frac{{CFO_{it + 1} }}{{Assets_{it} }} \\ &\quad + \hat{\beta }_{4} \frac{{\Delta Sales_{it} }}{{Assets_{it - 1} }} + \hat{\beta }_{4} \frac{{PPE_{it} }}{{Assets_{it - 1} }} + \varepsilon_{it} . \\ \end{aligned}$
	The TCA is the total accruals calculated as follows: ((act-L.act) − (lct-L.lct) − (che-L.che) + (dlc-L.dlc))/(L.at)). The CFO is calculated as follows: (ibc-TA), where TA is calculated as ((act-L.act) − (lct-L.lct) − (che-L.che) + (dlc-L.dlc) − (dp)). The Sales is the total sales (sale); the PPE is the gross plant, property, and equipment (ppegt); and the Assets is the total assets (at). Before running this regression and to remove the effect of the outliers, I remove the observations where the ratio of the total accruals to the assets is more than two standard deviations away. The industry classification is based on the two-digit SIC code, and I require that there be at least 15 observations in each industry-year. I also require that the firms be headquartered in the USA
	I then calculate the standard deviations in the residuals of the current year and the past 4 years. If this cannot be calculated due to missing values, then I replace the missing values with the standard deviations of the residuals of the current year and the past 3 years. Higher values represent higher accruals management. Source: COMPUSTAT
RealErngMgmt1	Following the literature (Cohen and Zarowin 2010), this variable is the sum of the abnormal discretionary expense (AB_DISEXP) and the abnormal level of production (AB_PROD). It is constructed so that the higher values represent greater real earnings management. The industry classification is based on the two-digit SIC code, and I require that there be at least 15 observations in each industry year. I also require that the firms be headquartered in the USA
	The AB_DISEXP is the residual (actual-predicted) of the following regression: $\frac{{DIS\_EXP_{it} }}{{Assets_{it - 1} }} = \beta_{0} \frac{1}{{Assets_{it - 1} }} + \beta_{1} \frac{{\Delta Sales_{it} }}{{Assets_{it - 1} }} + \varepsilon_{it} ,$ where DIS_EXP is the total discretionary expenses (xsga + xad + xrd). Consistent with the research (Cohen et al. 2008), the advertising (xad) and the R&D expense (xrd) are set to zero if the SG&A expense (xsga) is available. The Assets variable is the total assets (at), and the Sales is the total sales (sales). To remove the effect of the outliers, I remove the observations where the ratio of the discretionary expense to the assets is more than two standard deviations away. Following their studies, I standardize the residuals
	The AB_PROD is the standardized value of the abnormal level of production. I first estimate the residuals (actual-predicted) of the following regression: $\frac{{PROD_{it} }}{{Assets_{it - 1} }} = \beta_{0} \frac{1}{{Assets_{it - 1} }} + \beta_{1} \frac{{Sales_{it - 1} }}{{Assets_{it - 1} }} + \beta_{2} \frac{{\Delta Sales_{it} }}{{Assets_{it - 1} }} + \beta_{3} \frac{{\Delta Sales_{it - 1} }}{{Assets_{it - 1} }} + \varepsilon_{it}$where the PROD is the sum of the costs of the goods sold and the change in inventory (cogs + (invt − L.invt)), Assets is the total assets (at), and the Sales is the total sales (sales). To remove the effect of the outliers, I remove the observations where the ratio of the PROD to the assets is more than two standard deviations away. The residual is multiplied by −1 so that the higher values represent higher earnings management. I then standardize it. Source: COMPUSTAT
RealErngMgmt2	This variable is the sum of the abnormal discretionary expenses (AB_DISEXP) and the abnormal level of cash flows (AB_CASH). The higher values represent greater real earnings management calculated as in Cohen and Zarowin (2010). The industry classification is based on the two-digit SIC code, and I require that there be at least 15 observations in each industry year. I also require that the firms be headquartered in the USA
	The AB_DISEXP is calculated as described above. The AB_CASH is the residual (actual-predicted) of the following regression: $\frac{{CFO_{it} }}{{Assets_{it - 1} }} = \beta_{0} \frac{1}{{Assets_{it - 1} }} + \beta_{2} \frac{{Sales_{it - 1} }}{{Assets_{it - 1} }} + \beta_{3} \frac{{\Delta Sales_{it} }}{{Assets_{it - 1} }} + \beta_{4} \frac{{\Delta Sales_{it - 1} }}{{Assets_{it - 1} }} + \varepsilon_{it}$where the CFO is the cash flow from operations (oancf-xidoc), Assets is the total assets (at), and the Sales is the total sales (sales). To remove the effect of the outliers, I remove the observations where the ratio of the CFO to the assets is more than two standard deviations away. The residual is multiplied by −1 so that the higher values represent higher earnings management. I then standardize the residuals. Source: COMPUSTAT
FogIndex	This variable measures the readability of the annual reports as in Li (2008). It is constructed by Feng Li and available on his website. Source: http://webuser.bus.umich.edu/feng/
NumberOfWords	This variable measures the number of words in the annual report. It is constructed by Feng Li and available on his Web site. Source: http://webuser.bus.umich.edu/feng/
ln(SizeMegabytes)	The data is obtained from Bill McDonald’s Web site. Source: http://www3.nd.edu/~mcdonald/Word_Lists.html
Main variable of interest
SocialCapital	This measures the social capital at the county level and is constructed as in Rupasingha et al. (2008). Source: Northeast Regional Center for Rural Development (NERCRD), Rupasingha et al. (2008)
Demographic controls
Rrural	This is an indicator variable that takes the value of one if a firm is in a county that does not fall under the 100 most populated and zero otherwise. Core Based Statistical Area (CBSA). CBSA is a US geographic area defined by the Office of Management and Budget that consists of one or more counties (or equivalents) anchored by an urban center of at least 10,000 people. Source: Census Bureau
Religiosity	This variable is the percentage of religious adherents in the county. Source: Association of Religion Data Archive (ARDA)
ln(Population)	This variable is the natural log of the county’s population. Source: Census
IncomePerCapita	This variable is the income per capita in the county. Source: BEA
ln(DistancefromSEC)	This variable is the natural log of one plus the distance to the closest SEC branch. As in Kedia and Rajgopal (2011), the SEC branches considered are the SEC headquarters in Washington D.C and the regional offices located in New York City, NY; Miami, FL; Chicago, IL; Denver, CO; and Los Angeles, CA. As in their studies, the Haversine formula that uses the longitude and latitude of the two points is used to calculate the distance. Source: BEA
PopulationDensity	This variable is the population of the county divided by the land area of the county. Source: Census Bureau
Firm-level controls
ln(MarketValue)	This variable is the market value of equity calculated by multiplying the stock’s closing price in the calendar year and the number of common shares outstanding (prcc_c * csho). Source: COMPUSTAT
Analysts	This variable is the number of analysts for the latest consensus forecast (numest). If this number is not available for a firm, then the number of analysts following is assumed to be zero. Source: I/B/E/S unadjusted summary file.
ReturnOnAssets	This variable is the ratio of earnings before the interest tax depreciation and amortization to the total assets. Source: COMPUSTAT
DebtToAssets	This variable is the ratio of the total debt to the total assets (lt/at). Source: COMPUSTAT
Big4	This variable is a binary variable that equals one if the auditor is one of the Big4 and zero otherwise. Sources: COMPUSTAT
MarketToBook	This variable is the market-to-book value of the firm. It is the ratio of the common equity to the market value (ceq/(prcc_c * csho)). Source: COMPUSTAT
Loss	This is an indicator variable that equals one if the income before the extraordinary items (ibc) is less than zero for the current year or the last 2 years. Source: COMPUSTAT
VolatilityOfCashflow	This variable measures the volatility of the cash flows to the total assets (oancf/at) for the current year and the last 4 years. If the lag values for all 4 years are unavailable, I construct the volatility measure using the current year and the last 3 years. Source: COMPUSTAT
Benchmark	This variable is a binary variable that equals one if (1) the net income to the lag of assets (ni/at) is greater than or equal to zero but is <0.01 or (2) the change in the net income divided by the assets from the previous year is greater than or equal to zero but <0.01, and zero otherwise. Source: COMPUSTAT
ln(AuditorTenure)	This variable is the natural log of one plus the number of years the auditor has been with the firm. Sources: COMPUSTAT
ChangeInGDP	This variable is the percentage growth rate of the real GDP compared to the previous year. Source: BEA
Investment	This variable is the ratio of the capital expenditure in year t to the net property, plant, and equipment at the end of year t − 1 (capxv/L.ppent). Source: COMPUSTAT
NetOperatingAssets	This variable is the net operating assets. It is the sum of the stockholder’s equity less marketable securities and the total debt that is scaled by the total assets ((seq − che + lt)/at). Source: COMPUSTAT
ln(NumberOfSubs)	This variable is the natural logarithm of the total number of the subsidiaries. Source: COMPUSTAT
CEOReplaced	This is an indicator variable that equals one if the CEO is replaced in the following 2 years after the firm’s possible fraud was first revealed to the public. Source: Execucomp
FraudTriggerYear	This is an indicator variable that equals one for the firm year in which investors were told of possible fraud in the firm. Source: Execucomp
Post	Post is an indicator variable that equals one for the firm years after the firm relocates and zero otherwise. Source: Karpoff et al. (2012)
SocialCapital_Increasing_Move	This variable equals one for the firm year after the firm moved to a county with higher social capital, and zero otherwise. Source: Northeast Regional Center for Rural Development (NERCRD), Rupasingha et al. (2008).
SocialCapital(NoElection)	This variable measures the social capital without the influence of political activism. It is calculated in the same way as the SocialCapital except that one of the norm variables (the votes cast in the presidential elections divided by the population above 18) is omitted. Source: NERCRD
Electoral participation	This variable measures participation in a presidential election
Census response rate	This variable measures the percentage of people that return filled-out census forms
# of NGOs	This variable measures the number of non-government organizations normalized by the population in the county
# of associations	This variable measures the number of associations normalized by the population
Norm	This is the first component of a principal component analysis that uses Electoral participation and Census response rate
Network	This is the first component of a principal component analysis that uses # of NGOs and # of Associations
Industry dummies	This variable is a set of binary variables constructed based on the Fama–French 48-industry grouping. Source: COMPUSTAT

Appendix 2: Additional Tests

	(1)	(2)	(3)
	DV = FinclFraud
Panel A: The norms appear to have a stronger impact than the network
Norms	−0.221***		−0.211***
	(0.002)		(0.003)
Network		−0.157	−0.090
		(0.216)	(0.453)
Firm-level controls	Yes	Yes	Yes
County-level controls	Yes	Yes	Yes
Industry dummies	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes
Observations	85,743	85,743	85,743
(Pseudo) R2	0.076	0.074	0.076
Prob > χ2	(0.000)	(0.000)	(0.000)

	(1)	(2)	(3)	(4)
	DV = FinclFraud
Panel B: The common component of the four measures of social capital has a stronger effect on financial fraud
Electoral participation	−2.050***
	(0.004)
Census response rate		−1.448
		(0.151)
# of NGOs			−0.007
			(0.136)
# of associations				−0.424
				(0.127)
Firm-level controls	Yes	Yes	Yes	Yes
County-level controls	Yes	Yes	Yes	Yes
Industry dummies	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes
Observations	85,743	85,743	85,743	85,743
(Pseudo) R2	0.0750	0.0741	0.742	0.741
Prob > χ2	(0.000)	(0.000)	(0.000)	(0.000)

	(1)	(2)	(3)	(4)	(5)
	DV = FinclFraud
				year ≤ 2002	year > 2002
Panel C: State-fixed effects
SocialCapital	−0.175	−0.173*	−0.177*	−0.093	−0.623**
	(0.209)	(0.079)	(0.069)	(0.567)	(0.028)
State-fixed effect	Yes	Yes	Yes	Yes	Yes
Firm-level controls	Yes	Yes	No	Yes	Yes
County-level controls	Yes	No	No	Yes	Yes
Industry dummies	Yes	Yes	Yes	Yes	Yes
Year dummies	Yes	Yes	Yes	Yes	Yes
Observations	84,896	84,896	84,896	51,982	25,996
(Pseudo) R2	0.0826	0.0812	0.0766	0.0841	0.0979
Prob > χ2	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)

1. The DV refers to the dependent variables. The county-level controls are the same as in Table 1. The p values are in the brackets. They are based on the robust standard errors clustered at the county level. The ***, **, and * represent significance at the 1, 5, and 10% levels, respectively. The variable descriptions are in “Appendix 1” section. All continuous variables are winsorized at the 1st and the 99th percentile