Strange bedfellows? Voluntary corporate social responsibility disclosure and politics
Abstract
We show a reliable association between voluntary corporate social responsibility (CSR) disclosure and company political interests, which we proxy by company employees’ contributions to political action committees and statewide voting in presidential elections. This relation is most pronounced for the contributions of Democratic employees at companies in states that vote for the Democratic presidential candidate. We also show a positive association between corporate political contributions and excess stock returns. A portfolio strategy of investing based on company size, CSR disclosure intensity and corporate political contributions produces a significant positive mean excess stock return of 4.5 per cent over 3 months following CSR disclosure.
1. Introduction
Over the last four election cycles, corporate individuals have contributed an aggregate of more than $5 billion to federal committees registered with the U.S. Federal Election Commission (FEC).1 Given such corporate political contributions, many would contend that we should observe politics as a powerful factor in company decision-making. Yet, compared with the considerable evidence on how economic incentives and events affect company and market behaviour, and despite the obvious importance of politics to investors, creditors and other stakeholders, we know relatively little about how politics influences company behaviour and investment decision-making (2. reviews the relevant literature). This knowledge deficit applies especially to the question of whether politics might relate to disclosures about corporate social responsibility (CSR), because most would view CSR and politics as natural rather than strange bedfellows.
This paper tests hypotheses about whether political interests at the federal level influence companies’ decisions and investors’ responses regarding corporate voluntary CSR disclosure. First, we predict and test that company individuals’ and stakeholders’ political interests associate positively with the intensity of voluntary CSR disclosure (H1). Incremental to company individuals’ and stakeholders’ political interests, we further predict and test that additional variables such as the ongoing implicit claims of customers and suppliers and financial slack also have a significant influence on voluntary CSR disclosure intensity but do not rule out the explanatory power of the political variables. We then ask the question of whether investors use the published CSR information consistent with these political interests. This leads to our second hypothesis of an investor response to CSR news that varies positively with voluntary CSR disclosure intensity after controlling for other variables that explain investor response (H2). Because our first-stage model predicts a positive association between political variables and CSR disclosure intensity, this effectively links stock market performance to political interests to the extent that the latter help explain CSR disclosure intensity.2
We would expect investor behaviour of this kind to occur whether the political interests and companies’ activities related thereto are likely to result in beneficial regulatory, legislative or other outcomes for the company, and shareholders’ knowledge of the expected beneficial outcomes from these activities conditions their response to voluntary CSR disclosure. Some political interests, for instance, might seek to strengthen disclosure regulation (for example, the Sarbanes-Oxley Act 2002, Pub.L. 107–204, 116 Stat. 745 and the Dodd-Frank Reform Act 2010, Pub.L. 111–203, H.R. 4173) and favour the disclosures of companies with Democratic employees or shareholders, which could be beneficial for their stock price. Other political interests might promote weaker disclosure regulation (e.g. JOBS Act 2012)3 and favour the disclosures of companies with Republican interests, which could be beneficial for those companies’ stock price.4
We develop our hypotheses from voluntary disclosure theory, which predicts that companies make optimal disclosures in the best interests of shareholders (Verrecchia, 1983; Diamond and Verrecchia, 1991). As discussed below, the prior studies (and this investigation) provide reliable support for this theory in the context of environmental and social responsibility disclosure, in particular, the finding that investors react favourably to voluntary CSR information released by companies, whose decisions sensibly balance the costs and benefits of disclosure. We condition our hypotheses about whether political interests and variables representing stakeholders’ interests influence companies’ decisions and investors’ responses to voluntary CSR disclosure on this theoretical foundation as a maintained hypothesis.
Consistent with the preceding discussion, we produce two key findings. First, our study documents a reliable association between CSR disclosure intensity and company individuals’ and stakeholders’ political interests. This relates mostly to Democratic individuals’ political contributions to registered federal election committees, rather than Republican individuals’ contributions, and to disclosures made by companies headquartered in Democratic states, whose stakeholders presumably have similar political interests. This result supports our first hypothesis. Moreover, it holds after we control for customers’ and suppliers’ interests and other stakeholders’ claims regarding CSR information, as well as other company-related factors that could affect CSR disclosure intensity, such as environmental performance and financial slack.
Our second result is that investors’ response to voluntary CSR news announcements varies positively with CSR disclosure intensity, and we show further that companies with Democratic individuals versus Republican individuals generate more positive returns around CSR disclosure events. Our research design also considers the possibility that CSR disclosure intensity – which reflects political interests – and investor response are co-determined variables. However, when we control for co-determination, our results align more with the view that CSR disclosure intensity causes or conditions investor response and not vice versa. We also find that investors respond positively to a wider array of CSR disclosure types than shown in earlier research. This result girds a key underlying assumption of this paper, that corporate individuals consider them important and act sensibly in making voluntary CSR disclosures, and without apparent detriment to company shareholders.
These findings contribute uniquely to the literature in two major ways. To the best of our knowledge, we present new findings showing that political party affiliation makes a difference to company decision-making about CSR disclosure, which is the key question we raised at the outset. In other words, we show empirically that politics and CSR disclosure – as significant contributors to company economic value – make natural rather than strange bedfellows.
Our results contribute in a second way, in that we derive our results by analysing a heretofore un-researched data set on companies’ voluntary CSR disclosures made through the CSRwire news service. CSRwire provides immediate dissemination of CSR news by public companies and other member organizations to investors and other interested parties. CSRwire also claims to be the leading global distributor of CSR news and information (csrwire.com). This makes our analysis novel because other studies of the relation between CSR information and politics base their CSR measures on ratings or survey data and not on a comprehensive sample of company-specific, date-stamped, voluntary CSR disclosures, as we do here.
These contributions should have practical implications as well to the extent that they may help companies develop rational CSR disclosure policies and aid investors in using information about companies’ and stakeholders’ political interests to develop appropriate responses to CSR news and events. A strategy of investing in small companies with Democratic individuals and frequent CSR disclosure produces a significant mean portfolio excess return of 4.5 per cent over the three months following CSR disclosure.
Our study continues as follows. 2. discusses the literature and develops the hypotheses more fully. 3. describes the data and political variables and summarizes the CSRwire disclosure sample. 4. states the models and documents the tests of hypotheses. 5. summarizes the sensitivity and robustness checks. 6 concludes and offers suggestions for future work.
2. Literature and hypotheses
Much of the literature on voluntary corporate disclosure supports the prediction of the theory – that companies appropriately trade off the costs and benefits of disclosure in making optimal disclosure decisions (Verrecchia, 1983, 2001; Diamond and Verrecchia, 1991). These costs and benefits include the effects of competitive disadvantage, adverse selection, agency costs and the mix of information between inside and outside investors (information asymmetry). Empirical studies confirm the predictions of voluntary disclosure theory in a variety of settings, including management earnings forecasts (Frankel et al., 1995), franchise offerings (Price, 1999) and seasoned equity offerings (Lang and Lundholm, 2000). More recently, Ba et al. (2012) and Griffin and Sun (2013) show that investors respond to voluntary CSR disclosures in ways consistent with this theory. Other studies of CSR information examine the response of investors by exploiting events such as CSR disclosure initiation (Dhaliwal et al., 2011) and release of a sustainability report (Guidry and Patten, 2010) and by testing for differences in information quality (Plumlee et al., 2012) and industry segments (Ingram, 1978). These studies do not test for factors relating to the announcement effects of a comprehensive sample of companies’ voluntary CSR disclosures, however, which is one purpose of our study.5
A second literature looks at the determinants of voluntary corporate disclosure through the lens of how a company might satisfy the implicit claims of nonshareholder stakeholders who have a vested interest in company performance. This literature contends that stakeholders exert pressure for certain disclosure outcomes from the company's accounting and reporting system and have implicit claims on both to protect their interests (Klein and Leffler, 1981; Ullmann, 1985; Bull, 1987; Kale and Shahrur, 2007; Banerjee et al., 2008; Raman and Shahrur, 2008). Building on this second literature, Roberts (1992), Cormier and Magnan (1999), Cormier et al. (2005) and Clarkson et al. (2008) identify and test for similar stakeholder determinants of CSR voluntary disclosure, for example, those relating to shareholder, creditor and government interests; company economic, environmental and social performance; and financial policy and strategy.
For example, Clarkson et al. (2008) document a positive association between environmental performance and a disclosure-index-based companies’ sustainability reports. The authors further suggest that their findings contradict the notion that companies increase voluntary CSR disclosure to dissuade stakeholders from forming unfavourable views or to establish legitimacy, which is another way of explaining increased voluntary CSR disclosure. Bowen et al. (1995) identify similar determinants of disclosure with regard to accounting choice and, consistent with stakeholder theory, document implicit claims as relating to the interests of customers, suppliers, employees and short-term creditors. In the case of CSR disclosure, these claims might relate, for example, to information about product ethics (customers), the use of conflict-free materials (suppliers), safe labour conditions (employees), fair credit practices (lenders) and the interests of the public (Cormier et al., 2005).
In addition to stakeholders’ interests, we contend that voluntary CSR disclosure should relate to the political interests of company individuals and stakeholders, because we posit that these interests should be significant factors in legislation, regulation or other relationships expected to produce positive outcomes for the company, for instance, through stronger financial and environmental performance, risk reduction or higher enterprise value from favourable disclosures about the legislative benefits. We investigate two proxies for companies’ political interests. First, we base company individuals’ or inside political interests on the contributions of corporate individuals to political committees registered with the FEC. Second, we base stakeholders’ or outside political interests on state-wide voting in presidential elections, conditional on the assumption that state-wide voting correlates positively with the political orientation of the stakeholders in that state.6 We observe, anecdotally, that political interests can be highly significant in company disclosure regulations. For example, the Securities and Exchange Commission (SEC) voted along party lines (three Democrats versus two Republicans) to adopt the requirements of section 1502 of the Dodd-Frank Reform Act of 2010, requiring disclosure of the use of conflict minerals in companies’ manufacturing processes and supply chains (SEC 2012), presumably for the benefit of stakeholders and the public.7 Also, section 1502 was originally sponsored by a Democratically controlled congress in 2009 as part of the Conflict Minerals Trade Act of 2009. The Bill's 22 co-sponsors included 19 Democrats, two independents and one Republican.8
A third literature identifies several finance and accounting papers on the role of political interests in company decision-making.9 In an early study, Roberts (1992) finds a positive association between political contributions and social disclosure ratings (issued by the Council on Economic Priorities in 1986) and explains the increased disclosure as a rational response to political exposure. Roberts (1992) also finds that social disclosure ratings vary positively in return on equity and financial leverage. Hong et al. (2012), on the other hand, find that companies invest in social responsibility (SR) projects during periods of financial slack, where they view lower leverage and increased stock repurchases as signs of financial slack and proxy for SR activity using the ratings of Kinder, Lydenberg, Domini Research & Analytics (KLD). The role of financial slack suggests an excess funds explanation for SR investment, which could also influence CSR disclosure in addition to the stakeholder variables. We use the same variables as Hong et al. (2012), who contend that lower leverage and increased stock repurchases are good proxies for increased financial slack. However, in a related paper, Hong and Kostovetsky (2012) find that Democratic fund managers invest more in SR companies as measured by KLD ratings independent of financial slack, although such tilt to SR companies does not affect investment fund performance. Rubin (2008) also finds that companies with high (low) KLD ratings locate in Democratic (Republican) states. Similarly, Di Giuli and Kostovetsky (2012) find that companies with Democratic CEOs or headquartered in Democratic states have higher KLD ratings than companies with Republican CEOs or headquartered in Republican states.10
Importantly, none of these studies test whether company individuals’ (inside) political interests affect investment returns or investors’ response to CSR announcements. Thus, although company and investment managers apparently exhibit variation in their political interests, and this helps explain variation in companies’ social disclosure ratings, the prior research does not explain or document how politics relates to actual company decisions or outcomes; in this case, whether companies with predominantly Democratic or Republican employees engage in more voluntary CSR disclosure and whether stakeholders’ interests or implicit claims might further condition company individuals’ CSR disclosure decisions or outcomes. Some research, also, has questioned whether KLD ratings actually measure corporate social performance (Chatterji et al., 2009), which is a further reason to analyse CSR disclosures around announcement dates. Dimson et al. (2012) comment similarly.
To resolve this conundrum, we base the political interests of company individuals on their contributions to committees registered with the FEC and predict an overall positive relation between CSR disclosure intensity and company individuals’ contributions to those political committees. However, given the inherent nature of CSR disclosure (which reflects stakeholders’ interests in greater environmental, social and governance disclosure) and the preceding results from the literature (earlier this section), we also expect a stronger positive CSR disclosure relation for Democratic company individuals’ political contributions than Republican company individuals’ political contributions. We contend that, among other factors, including stakeholders’ implicit claims, company individuals’ political views are manifested, in part, in voluntary CSR disclosure policy, because disclosure is one means by which these individuals can affect a favourable political or regulatory outcome for investors. Later in this section, we explain why CSR disclosure should favour Democratic versus Republican interests.
We define stakeholders’ political interests as the percentage of votes cast in the state in which the company is headquartered for the Democratic presidential candidate less the votes cast for the Republican presidential candidate in the 2000, 2004 and 2008 general elections. For example, we denote the stakeholders of Intel Corp. as more Democratic than Republican in 2004-2007, as Intel has its headquarters in California and the state voted 55 per cent for Kerry (versus 44 per cent for Bush) in the 2004 general election. We score the state in years 2004-2007 as +11 (55-44) per cent. We do not literally mean, however, that all Intel stakeholders are Democrats and vote in California, but, simply, that Intel stakeholders in general reflect the same voting choices as Californians in the general election.11
Given this definition, we then test for an empirical relation based on the natural and historical inclination of Democrats to place more emphasis on SR investment and CSR disclosure than Republicans. But unlike Rubin (2008) and Di Giuli and Kostovetsky (2012), who find a positive relation between CSR activity based on KLD ratings and politics, we test for a positive relation between CSR disclosure and company individuals’ and stakeholders’ political interests in the context of company individuals’ actual decisions to disclose CSR news to investors and the public.
In contrast to the literature on SR investment and politics, in which party affiliation appears to play a role, several studies find that Democratic and Republican managers’ investment returns are broadly equivalent. For example, Goldman et al. (2009) find a positive stock price response to the announcement of the nomination of a politically connected board member, but this holds for board members from both parties. Hong and Kostovetsky (2012) find no difference in the performance of investment funds managed by Democratic versus Republican company individuals, despite the former's tilt to socially responsible stocks. Lee et al. (2012) find a negative relation between politically similar company individuals and boards and shareholder value, which they reason occurs because the common political orientation increases agency costs; but these results too do not differ conditional on party affiliation. Consistent with these results, Ansolabehere et al. (2003, 2004) find no association between political contributions and votes cast by politicians or regulators on issues critical to companies. Thus, based on the prior work, we would not expect investors to respond to voluntary CSR disclosures conditional on company individuals’ political contributions or stakeholders’ party affiliation.
On the other hand, it is reasonable to assume that the political interests in and the resources contributed by company individuals to registered political committees would not be entirely altruistic and, thus, would be expected to have some benefits for the company and company stakeholders, for example, in the form of favourable legislation or regulation. Why and when those benefits might be reflected in shareholder value is unclear, however. For example, Faccio (2006) documents a positive association between political contributions and investor returns, reasoning that this stems from the connectedness of companies and politicians, although a later paper (Chaney et al., 2011) concludes that politically connected companies disclose low-quality earnings information, which could reduce such positive relation. Chen et al. (2012) find a positive relation between corporate political expenditures and future investor returns, but only for the companies in the top 20 per cent of political expenditures (scaled by total assets), which leads them to conclude that most investors receive no benefits.12
Cooper et al. (2010) offer a different explanation based on the notion that companies view political contributions as positive net present value expenditures (also, Stigler, 1971). Under this view, stock returns reflect pay-offs to those expenditures. Also, under this view, one channel that could reflect those pay-offs would be that which provides timely CSR disclosure to shareholders and investors, because we contend and show in this study that political interests in part drive the intensity of CSR disclosure. Moreover, the CSRwire service makes an ideal choice in this regard, as it was the first to be established (in 1999) and, today, claims to be the leading global platform used by companies and others for timely and credible distribution of CSR and sustainability news (cswire.com).13
As such, while the null hypothesis would be that the stock market does not condition its response to CSR announcements on political interests, the alternative would be to predict a positive market response, namely, that such response varies positively with political interests and the presumed benefits of CSR regulation and legislation expected to accrue to the company.
These benefits could result from either party. But given the long history of Democratic politics in environmental and social regulation and legislation and evidence of Democrat's interests from research (e.g. Kamieniecki, 1995; Gruber, 2001; and Davis and Wurth, 2003) and legislative tracking by groups such as the League of Conservation Voters of House and Senate Democrats’ and Republicans’ votes on environmental or social legislation (scorecard.lcv.org), our expectation is that we should observe a more positive market response for companies with Democratic individuals and/or companies located in Democratic states, and we should observe this positive relation either indirectly through the link between CSR disclosure intensity and company individuals’ or stakeholders’ political orientation or directly as a positive relation between market response and company individuals’ political orientation.14
| Variable | N | Mean | SD | Q1 | Median | Q3 |
|---|---|---|---|---|---|---|
| Panel A: CSR sample | ||||||
| LogAT | 4,478 | 23.360 | 2.153 | 22.402 | 23.612 | 24.602 |
| BTM | 4,458 | 0.517 | 0.411 | 0.252 | 0.445 | 0.623 |
| ROA | 4,477 | 0.041 | 0.102 | 0.013 | 0.046 | 0.088 |
| LogFREQ | 4,781 | 2.166 | 1.445 | 1.099 | 2.197 | 3.332 |
| LogCOGS | 4,467 | 22.293 | 2.060 | 21.401 | 22.775 | 23.720 |
| LogRECT | 4,365 | 21.311 | 2.403 | 20.166 | 21.392 | 22.611 |
| XAD | 4,478 | 0.016 | 0.034 | 0 | 0 | 0.018 |
| USELIFE | 3,866 | 2.518 | 0.547 | 2.221 | 2.539 | 2.938 |
| LEV | 4,477 | 0.176 | 0.147 | 0.052 | 0.157 | 0.251 |
| DURABLE | 4,781 | 0.277 | 0.447 | 0 | 0 | 1 |
| REPURCHASE | 4,519 | 0.721 | 0.449 | 0 | 1 | 1 |
| Panel B: Compustat population | ||||||
| LogAT | 113,644 | 19.074 | 2.800 | 17.270 | 19.258 | 21.011 |
| BTM | 101,784 | 0.542 | 1.237 | 0.225 | 0.500 | 0.882 |
| ROA | 112,944 | −0.234 | 0.785 | −0.118 | 0.009 | 0.052 |
| LogCOGS | 106,167 | 18.094 | 2.735 | 16.292 | 18.104 | 20.058 |
| LogRECT | 106,431 | 16.982 | 3.092 | 15.035 | 17.197 | 19.181 |
| XAD | 132,898 | 0.008 | 0.035 | 0 | 0 | 0 |
| USELIFE | 93,573 | 2.363 | 1.014 | 1.793 | 2.367 | 2.875 |
| LEV | 113,425 | 0.177 | 0.235 | 0 | 0.079 | 0.273 |
| DURABLE | 132,898 | 0.242 | 0.428 | 0 | 0 | 0 |
| REPURCHASE | 132,898 | 0.246 | 0.431 | 0 | 0 | 0 |
| Panel C: Political variables | ||||||
| LogCON | 4,781 | 4.348 | 5.014 | 0 | 0 | 9.568 |
| LogCON_D | 4,781 | 3.510 | 4.636 | 0 | 0 | 8.666 |
| LogCON_R | 4,781 | 3.858 | 4.741 | 0 | 0 | 8.772 |
| BLUERED | 4,781 | 6.655 | 15.965 | −2 | 10 | 17 |
| Number of companies | No. CSR releases | Percentage |
|---|---|---|
| Panel D: Frequency of CSR disclosures | ||
| 4 | 528 | 11.04 |
| 9 | 961 | 20.10 |
| 16 | 1,432 | 29.95 |
| 29 | 1,934 | 40.45 |
| 45 | 2,406 | 50.32 |
| 70 | 2,869 | 60.01 |
| 110 | 3,346 | 69.99 |
| 185 | 3,847 | 80.05 |
| 351 | 4,304 | 90.02 |
| 785 | 4,781 | 100.00 |
| 391 | 1 | 8.18 |
| 115 | 2 | 2.41 |
| 67 | 3 | 1.40 |
| 26 | 4 | 0.54 |
| 16 | 5 | 0.33 |
- This table presents descriptive statistics for all the key variables. Panel A provides statistics for all the financial variables for the main sample. Panel B provides statistics for the similar financial variables for Compustat population between 2000 and 2011. Panel C provides statistics for all the political determinants for the main sample. Panel D provides the frequency of CSR announcements for the CSRwire sample. See the 1 for variable definitions.
As noted above, prior research suggests that such positive response could result from the connectedness of Democratic legislators and Democratic company individuals (Faccio, 2006) or the expectation of net pay-off to the company from the political contributions (Cooper et al., 2010). Our research design is unable to distinguish which channel might best explain the relation, however. In addition, we might expect higher CSR disclosure intensity for Democratic companies for other reasons as well. One might stem from shared beliefs among Democratic stakeholders, company individuals and legislators that CSR regulation and legislation carry benefits for companies and their constituents. Another might relate to the perceived higher quality of CSR reporting by companies in Democratic states, which could make CSR disclosure a lower risk signal for investors in Democratic states. Yet, a third might relate to differences in how stakeholders in Blue states versus Red states impose penalties on socially irresponsible behaviour.
In sum, we state and test two models, which we specify as regression equations in the results section. Model 1 focuses on the company's decision to disclose CSR news and specifies a relation between CSR disclosure intensity and political interests after controlling for stakeholders’ implicit claims and other company-specific factors that might drive the disclosure decision. Model 1 tests our first hypothesis (H1). Model 2 extends the analysis of investor response to CSR news in the prior literature by specifying and testing political and economic variables to explain cross-sectional variation in that response. These factors include Democratic and Republican company individuals’ political interests, stakeholders’ political interests and other factors known to explain returns in non-CSR settings such as company performance, size and growth opportunities. Model 2 tests our second hypothesis (H2).
3. Data and samples
3.1. CSR data and sample
We obtain our data from CSRwire, which made available to us their data archive of 14,561 date-stamped CSR releases distributed during January 2000 to December 2011 and classified into 23 CSR categories. We then matched the releases made by companies in the archive to companies listed in the CRSP/Compustat merged database. This produced a final sample of a maximum of 4,781 CSR observations representing 784 CRSP/Compustat companies and 1,683 company-years. We also extracted financial and stock return data from the CRSP/Compustat merged file. Because we constrained the CSR disclosure sample companies to have a CSRP/Compustat PERM number, the selection of variables from this database caused a small reduction in sample size in some analyses, due mostly to missing values.
Panels A and B of Table 1 summarize the merged CSRwire/CSRP/Compustat sample and the CRSP/Compustat population based on common characteristics. Panel A first shows that a log transformation produces reasonably symmetrical variables, with similar means and medians. We use log transformations in our later regression tests. Second, panel A indicates little reduction in sample size from the constraints of financial and stock price data from CRSP/Compustat, except that not all CSRwire companies report long-lived assets and depreciation, which we require to compute the estimated useful life of property, plant and equipment. Third, panel A shows that the CSR sample is on average larger in size (LogAT), more successful (ROA), has longer-lived assets (USELIFE) and engages in more share repurchases (REPURCHASE) than the broader CRSP/Compustat population.15 As such, our results do not generalize to the larger Compustat population. Panel C summarizes the political variables and shows mean LogCON of 4.348, which if unlogged translates to an overall mean sum of individual contributions per company per election cycle of $39,451. Equivalently, LogCON_D and LogCON_R translate to mean corporate contributions per election cycle of $42,065 and $37,056 to Democratic and Republican committees, respectively.16 Panel D of Table 1 shows the frequency of CSR disclosures per company. Some companies disclose more frequently than others. For example, four companies account for 11.04 per cent of the CSR disclosure sample (and two companies – Alcoa and Dow Chemical – comprise 6.5 per cent of the sample). Also, 8.18 per cent of the companies made only one disclosure, and 12.86 per cent disclosed at most five times.
Finally, we use Factiva to check a random sample of 10 per cent of the CSR disclosure sample for contemporaneous company press releases. While company press releases do not occur for all CSRwire releases, in those cases of a match, we observe no more than a 1-day lag between the company press release date and the CSRwire release date. From a design standpoint, this means that we are reasonably sure that CSRwire releases provide fresh information to investors and others about company CSR activities and events.
3.2. Federal election commission data and political variables
To proxy for company individuals’ political interests at the time of a voluntary CSR disclosure, we use data from the FEC on the contributions of corporate individuals to registered political committees. The FEC data files cover the period 1979–1980 to the present, which more than sufficiently covers the CSR study period of 2000–2011. We sourced the data at http://www.fec.gov/. We first extract for each two-year election cycle (1999–2000 to 2011–2012) the individual contributions data from the FEC's ‘Contributions by Individuals’ file and the related political committee information from the FEC's ‘Committee Master File’. The contributions file contains the name, employer, job title, and date and amount of contribution by each individual to a registered committee if the contribution was at least $200. The ‘Committee Master File’ contains one record for each committee registered with the FEC, and the committee affiliation can be Democrat (hereafter, Dem.), Republican (hereafter, Rep.) or one of several other parties.
Next, we match the individual contributions file with the committee master file to obtain the names of the individuals, their contributions, employers, titles, the registered committee to which they contribute and the political affiliation of the registered committee (we recorded either Dem. or Rep. only). Because the FEC compiles the contribution files biennially, each individual contribution is specific to a biennial period (for example, 2009–2010) and not to a single calendar year. As such, we assign contributions for a given biennial period to both calendar years, which means that we assume that an individual's political interests in one year carry over to the next. For example, for the 2009–2010 contributions file, we assign years 2009 and 2010 to all observations in that file when we match the individual Dem. and Rep. political contributions at each company to our CSR sample file by company and year.
Panel A of Table 2 provides additional detail. We eventually obtain political contributions data made by Dem. and Rep. individuals aggregated at the company level for 2,134 CSR disclosures, comprising 44.6 per cent of the 4,781 CSR disclosure sample. While we determine that the vast majority of individuals state managerial titles such as executive, manager, president, treasurer and director, we use the term ‘company individuals’ in the paper, as we seek to proxy for the overall political orientation of the company, and not that of a few key individuals at head office (e.g. CEO or CFO), who may not represent the company's overall political orientation regarding CSR.17 We are also able to identify the home state for 3,863 of the 4,781 CSR company disclosures.
| No. | Step | No. of records | Record type |
|---|---|---|---|
| Panel A: Data collection procedure | |||
| 1. | Prepare the contribution file | ||
| a. | Match contributions to the associated registered committee file for calendar years 2000, 2002, 2004, 2006, 2008, 2010 and 2012, respectively (raw form) | 8,211,847 | Individuals |
| b. | Double the biennial records by assigning the individual contributions to both calendar years | 16,423,694 | Individuals |
| 2. | Select the CSR disclosure sample | ||
| a. | CSR releases from CSRwire.com between 2000 and 2011 | 14,561 | Announcements |
| b. | CSR releases in 2(a) with GVKEY and PERMNO | 4,781 | Announcements |
| 3. | Match the FEC contributions file with the CSR disclosure sample in 2(b) | ||
| a. | Match using the first word in the company name and the calendar year, and remove mismatched records based on a manual check | 126,293 | Announcements/individuals |
| b. | In a given year, sum individuals’ contributions at a particular CSR company made to a Democratic or Republican registered committee | 5,435 | Company-years summed over individuals’ contributions |
| 4. | Match the file from 3(b) with the CSR disclosure file 2(b). CSR observations with FEC contributions data | 2,134 | Announcements |
| 5. | Match with CSR disclosure file 2(b) with Democratic (blue) and Republican (red) states based on CSR company home state and voting in 2000,2004 and 2008 presidential elections | 3,863 | Announcements |
| Sum of individuals’ contributions per CSR company | No. of Democratic Committees | No. of Republican Committees |
|---|---|---|
| Panel B: Distribution of contributions to FEC committees for CSRwire sample | ||
| $1,000,000 and above | 3 | 5 |
| $500,000 | 4 | 4 |
| $250,000 | 28 | 33 |
| $100,000 | 53 | 66 |
| $50,000 | 140 | 146 |
| $10,000 | 61 | 61 |
| $5,000 | 124 | 132 |
| $1,000 | 95 | 99 |
| $200 | 9 | 10 |
| Mean CSR company contribution | $42,065 | $37,056 |
| t-test of difference in mean | t = 6.31 | p < 0.0001 |
| N (companies) | 1,810 | 1,974 |
| Median | $12,456 | $15,620 |
| Minimum | $200 | $200 |
| Maximum | $2,571,407 | $1,080,035 |
- This table presents FEC sample selection and distribution of contributions. Panel A lists the steps of data collection from the Federal Election Commission (FEC). Panel B summarizes the distribution of the sum of Dem. and Rep. contributions per company per election cycle across the CSR disclosure sample. The same person could have made a contribution to a Democratic and Republican FEC committee on the same date. CSR, corporate social responsibility.
Panel B of Table 2 summarizes the distribution of the sum of Dem. and Rep. contributions per company per election cycle across the CSR disclosure sample. These amounts range from a statutory minimum of $200 for a company with one individual contribution (both parties) to a high of $2.57 million ($1.08 million), which is the maximum sum of individuals’ contributions per company per election cycle to a Democratic (Republican) committee. In some cases, the corporate contribution per election cycle comprises thousands of individuals at the company. For example, in the 2007–2008 election cycle, our data show that 4,547 individual contributions at AT&T produce an overall corporate contribution of $707,385 to registered FEC committees, comprising 582 contributions totalling $214,112 to Republican committees and 3,965 contributions totalling $493,273 to Democratic committees.
The table also shows that the mean company Dem. contribution of $42,065 (averaged over Dem. individuals per company per election cycle) significantly exceeds the mean company Rep. contribution of $37,056 (averaged over Rep. individuals per company per election cycle) based on a t-test (t = 6.31, p < 0.0001). On the other hand, the median values are lower than the means for both groups, especially for Dem. individuals, which are skewed by more extreme amounts. We correct for skewness in our empirical tests using a logarithmic transformation.
4. Results
4.1. Disclosure intensity and political variables
Our first set of results on politics and CSR pertains to model 1 (stated formally below), which regresses CSR disclosure intensity on independent variables that would explain disclosure intensity. We split these independent variables into three kinds, namely political, ongoing implicit claims and other company characteristics. Because some companies disclose CSR news more frequently than others, we view this model as an expression of companies’ decision to disclose voluntarily. We realize that CSR disclosure frequency is but one aspect of disclosure intensity, and a complete analysis would combine additional factors such as disclosure length, tone, content and placement into a single measure.18 However, while more complex measures are possible, we start with a straightforward, ordinal measure of companies’ decision to disclose based on frequency of disclosure.
(1)Following prior research (e.g. Rubin, 2008; Di Giuli and Kostovetsky, 2012), we expect positive coefficients for β1 and β2, and a more significant β1 for Dem. individuals when we split politicali,1 into the sum of Dem. individuals’ contributions (LogCON_D) and the sum of Rep. individuals’ contributions (LogCON_R). While not shown in model 2, we also interact LogCON (and LogCON_D and LogCON_R) with BLUERED. While we would expect a positive interaction coefficient for LogCON*BLUERED, as our descriptive data show that Blue state political contributions per company per election cycle exceed Red state contributions per company per election cycle (Table 2, panel B), we are unsure about the coefficient signs for Dem. individuals’ contributions in a Blue state versus a Red state (LogCON_D*BLUERED) and Rep. individuals’ contributions in a Blue state versus a Red state (LogCON_R*BLUERED). This could depend on whether corporate individuals’ and stakeholders’ interests are reinforcing when they represent the same party or whether same-party interests have offsetting effects, for example, Rep. individuals might wish to disclose more frequently in Blue states than Red states.
Our next set of variables relates to the ongoing implicit claims of nonshareholder stakeholders, where claimsi,1 = suppliers’ interests, which we proxy by the log of cost of goods sold (LogCOGS), and claimsi,2 = customers’ interests, which we proxy by the log of accounts receivable (LogRECT) and claimsi,3 = dummy variable for those industries that make durable products (DURABLE). Following Bowen et al. (1995) and others, we expect positive χ coefficients for these variables based on the notion that the greater the amount or presence of these variables the more interested the company is in supplying CSR (and non-CSR) information to satisfy those stakeholders’ claims.21
We define a third set of variables, otherm = 1 to 5, where, similar to the above, each relates to a CSR disclosure year. First, we include estimated useful life of property, plant and equipment (other1 = USELIFE), as research shows that companies with longer-lived assets have higher disclosure rankings (Clarkson et al., 2008) and greenhouse gas emissions (Griffin, 2013), which should increase CSR disclosure. Second, we include advertising expenditure (other2 = XAD), as such expenditures have been shown to relate positively to customers’ interests and company image (Bowen et al., 1995). Third, we include variables to control for financial slack, namely leverage (other3 = LEV) and the presence of stock repurchases (other4 = REPURCHASE). Fourth, we control for industry as a fixed effect (other5 = INDUSTRY) based on a company's membership in a one-digit SIC code.22 We also use CSRwire's assignment of a company to one of 18 industry categories as an alternative classification. Because Hong et al. (2012) find that SR activity increases in periods of financial slack, we expect a negative ϕm coefficient for LEV, as higher leverage indicates less financial slack, and we expect a positive ϕm coefficient for REPURCHASE, as stock repurchases often occur when a company has excess cash or financial slack. As a fixed effect, we are uncertain about the signs of the industry coefficients.
We assume the regression residuals, ε, represent uncorrelated residuals subject to clustering by company and year. We adjust our significance tests of the regression coefficients in model 2 for potential cross-correlations using the methods in Cameron et al. (2011). Failure to control for observation clustering can be critical in panel regression analysis because the regression standard errors based on conventional hypothesis tests can be under-estimated and lead to over-rejection of the null hypothesis of no effect. In sum, model 1 allows us to test our first hypothesis (H1) of a positive relation between voluntary CSR disclosure intensity and political variables, with controls for other factors that might also explain disclosure intensity.
Table 3 presents the results. We show seven regressions, where each combines the implicit claims and financial slack and other variables with different combinations of the political variables. First, regarding the implicit claims variables, our proxies for customers’ claims (LogRECT) and suppliers’ (LogCOGS) claims have positive and mostly significant χcoefficients, especially customers’ interests.23 Thus, consistent with the prior research that links financial reporting choices to stakeholders’ economic interests (Bowen et al., 1995), those same proxies also reflect stakeholders’ ongoing interests in nonfinancial information, in this case, the demand for disclosure about social responsibility and environmental impact. The dummy variable for industries with durable products (DURABLE), on the other hand, shows insignificant coefficients across the regressions, suggesting that customers’ concerns about industries with durables, such as products that require long-term availability of parts and servicing, do not seem to matter, although this effect could also be subsumed by INDUSTRY fixed effects.24
| Regression no. | (1) | (2) | (3) | (4) | (5) | (6) | (7) | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Variable | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. |
| Intercept | −5.4971 | *** | −5.5560 | *** | −5.5464 | *** | −5.1207 | *** | −5.0270 | *** | −5.1547 | *** | −5.0299 | *** |
| (6.60) | (6.49) | (6.36) | (6.17) | (6.00) | (6.31) | (6.10) | ||||||||
| BLUERED | 0.0150 | *** | 0.0197 | *** | 0.0144 | *** | 0.0138 | *** | 0.0009 | ns | 0.0009 | ns | ||
| (3.11) | (3.92) | (3.26) | (3.21) | (0.15) | (0.15) | |||||||||
| LogCON | 0.0422 | *** | 0.0303 | ** | ||||||||||
| (3.65) | (2.32) | |||||||||||||
| LogCON_D | 0.0537 | *** | 0.0493 | ** | ||||||||||
| (2.72) | (2.39) | |||||||||||||
| LogCON_R | 0.0027 | ns | −0.0049 | ns | ||||||||||
| (0.14) | (0.25) | |||||||||||||
| LogCON*BLUERED | 0.0024 | *** | ||||||||||||
| (2.54) | ||||||||||||||
| LogCON_D*BLUERED | −0.0012 | ns | ||||||||||||
| (0.98) | ||||||||||||||
| LogCON_R*BLUERED | 0.0037 | *** | ||||||||||||
| (3.10) | ||||||||||||||
| LogRECT | 0.1784 | *** | 0.1552 | *** | 0.1633 | *** | 0.1560 | *** | 0.1500 | *** | 0.1343 | *** | 0.1302 | *** |
| (3.66) | (3.01) | (3.24) | (3.04) | (2.92) | (2.68) | (2.61) | ||||||||
| LogCOGS | 0.0854 | ns | 0.1097 | * | 0.1013 | ns | 0.0808 | ns | 0.0799 | ns | 0.1055 | * | 0.1015 | * |
| (1.38) | (1.64) | (1.53) | (1.27) | (1.25) | (1.71) | (1.64) | ||||||||
| DURABLE | −0.0084 | ns | −0.1194 | ns | −0.2010 | ns | −0.2028 | ns | −0.2206 | ns | −0.1835 | ns | −0.2016 | ns |
| (0.04) | (0.56) | (0.98) | (0.97) | (1.06) | (0.87) | (0.96) | ||||||||
| USELIFE | 0.2335 | * | 0.1862 | ns | 0.1695 | ns | 0.1725 | ns | 0.1820 | ns | 0.1466 | ns | 0.1476 | ns |
| (1.69) | (1.45) | (1.37) | (1.43) | (1.52) | (1.28) | (1.29) | ||||||||
| XAD | 0.3589 | ns | 0.2414 | ns | −0.0175 | ns | 0.6197 | ns | 0.6612 | ns | 0.8438 | ns | 0.8822 | ns |
| (0.19) | (0.13) | (0.01) | (0.33) | (0.35) | (0.45) | (0.47) | ||||||||
| LEV | −0.6401 | ns | −0.5688 | ns | −0.5104 | ns | −0.4621 | ns | −0.4612 | ns | −0.6592 | ns | −0.6395 | ns |
| (1.40) | (1.17) | (1.02) | (0.93) | (0.92) | (1.53) | (1.47) | ||||||||
| REPURCHASE | 0.3272 | *** | 0.3022 | ** | 0.2735 | ** | 0.2898 | ** | 0.2851 | ** | 0.3024 | ** | 0.3069 | ** |
| (2.61) | (2.42) | (2.14) | (2.34) | (2.30) | (2.47) | (2.49) | ||||||||
| INDUSTRY | YES | YES | YES | YES | YES | YES | YES | |||||||
| Adj R2 (%) | 23.11 | 24.71 | 26.12 | 26.64 | 27.60 | 27.84 | 28.89 | |||||||
| N | 3,756 | 3,756 | 3,756 | 3,756 | 3,756 | 3,756 | 3,756 | |||||||
- This table presents an analysis of the relation between CSR disclosure frequency and political, implicit claims and other variables. BLUERED in regressions 2, and 4–7 refers voting percentage difference if year<2004, Gore-Bush; else if year <2008, Kerry-Bush; else Obama-McCain. BLUERED in regression 3 refers to voting percentage difference if year<2002, Gore-Bush; else if year <2006, Kerry-Bush; else if year<2012 Obama-McCain. *, **, *** indicate significance at the 0.10, 0.05 and 0.01 levels, respectively, using two-tailed tests, and ns indicates not significant. t-statistics and p-values are calculated using clustered standard errors by company and year. See the : 1 for variable definitions.
Second, the regressions show uniformly positive χ coefficients for the average useful life of property, plant and equipment (USELIFE), although they are not all significant. This could occur to the extent that longer useful life associates with higher carbon emissions, which means that we might also expect CSR disclosure intensity to increase in asset useful life. Third, Table 3 shows greater CSR intensity for companies that have financial slack, as proxied by leverage (LEV) and share repurchases (REPURCHASE). The χcoefficients for REPURCHASE are significantly positive, and the for LEV are uniformly negative (although not significant). These financial slack results are also consistent with the literature (Orlitsky et al., 2003; Hong et al., 2012). Overall, our results for the implicit claims control variables are broadly consistent with the literature, and companies that accumulate more financial slack associate with higher CSR disclosure intensity or, as Hong et al. (2012, p. 4) state, ‘less constrained firms spend more on goodness’.
We now turn to the political variables and a test of our first hypothesis. Table 3 documents five results of interest. First, we show that a variable for each CSR disclosure/year defined as the percentage in a state in which the company is headquartered voted Democrat less the percentage that voted Republican in a general election makes a difference. The βcoefficient for this variable (BLUERED) is positive and significant across all the regressions that include it as a noninteractive explanatory variable (regressions 2–5) or an interactive explanatory variable (regressions 6 and 7). In other words, consistent with expectations, we document that CSR intensity is higher in Blue states than in Red states. Based on the literature on home bias, which suggests that investors prefer local stocks (Coval and Moskowitz, 1999; Campbell, 2006) and stocks of their employer (Mitchell and Utkus, 2004), we further contend that BLUERED should associate positively with shareholders’ political interests in local companies, although not exclusively as this proxy also likely reflects the political leanings of all stakeholders and the voting public in general.
Second, we show that the sum of contributions by company individuals (LogCON) has a positive and significant βcoefficient (regressions 4 and 6). Hence, inside political contributions in general influence CSR disclosure intensity. Third, we split the variable LogCON into two variables, namely LogCON_D and LogCON_R. Now, only the coefficient for LogCON_D is positive and significant, that is, Dem. company individuals’ contributions have more explanatory power. For example, in regression 5, the t-value for LogCON_D equals 2.72, whereas the t-value for LogCON_R equals 0.14. Fourth, regressions 6 and 7 include a variable that interacts the dummy variable BLUERED with corporate individuals’ political contributions. While the interaction coefficient is significant for LogCON*BLUERED (regression 6), the coefficients for LogCON_D*BLUERED and LogCON_R*BLUERED are insignificant and significantly positive, respectively (regression 7). In other words, when stakeholders’ and corporate individuals’ political parties disagree (LogCON_R*BLUERED), CSR disclosure intensity increases (β = 0.0037 t = 3.10), and when they agree (LogCON_D*BLUERED), they seem to have offsetting effects as CSR disclosure intensity is unaffected (β = −0.0012 t = 0.98). Fifth, we observe from the regressions that the addition of variables representing corporate individuals’ political interests makes little difference to the explanatory power of BLUERED and vice versa. Both factors apparently contribute uniquely to explaining CSR disclosure intensity, either directly or through interaction effects.
In sum, the results in Table 3 support our first hypothesis, namely, that the political interests of corporate individuals and stakeholders and the combination of the two relate significantly to voluntary CSR disclosure intensity (H1). To the best of our knowledge, this finding is new to the literature, in that the earlier results that explain CSR activities as a function of political variables use CSR ratings from surveys and not date-stamped companies’ voluntary CSR disclosures, as we do here.25
4.2. Excess stock return and political variables
Having established an empirical link between CSR disclosure intensity and corporate individuals’ and stakeholders’ political interests, we now examine whether those same political interests might associate with excess stock returns around the CSR announcement date. For the reasons enumerated below, we use a two-stage regression approach to conduct this analysis, where we first predict CSR disclosure intensity using the estimated coefficients from model 2 (based on implicit claims and the other variables, but not the political variables). We then use predicted CSR disclosure intensity from model 2 as an instrument in the second-stage regression. We adopt this two-stage approach because of the possibility that disclosure intensity might not be exogenous to announcement period stock return and, more broadly, because it is unclear whether CSR disclosure intensity helps drive investors’ response, is driven by investors’ response to CSR disclosure, or both. If stock return and disclosure intensity are co-determined variables, this can bias the coefficient for LogFREQ in a regression of stock return on that variable, causing misleading inferences. While much of the prior empirical literature (Botosan, 1997) adopts the view that the causation runs from disclosure to stock market response, Venky et al. (2003) document that stock market-based managerial incentives increase the frequency of voluntary management earnings guidance (MEG) disclosures, and Dhaliwal et al. (2011) find that higher past cost of equity capital (based in part on stock market variables) associates with greater voluntary CSR disclosure in the current year.26 In other words, the causation could go both ways, and hence the need for an approach such as what we use here. Also, Shin (2006) contends that good news increases investors’ likelihood of more company information in the future, so that in these cases the favourable stock price response (from the good news) could help increase the frequency of future disclosures.
We contend that these effects could also occur in a CSR setting, because under the preceding theory, it is reasonable that a manager who expects a positive stock price response to CSR news would be more inclined to increase CSR disclosure in the future. In short, in addition to a favourable current stock price response to CSR news, such favourable response could also contribute to higher disclosure frequency in the future.
(2)| Regression no. | (1) | (2) | (3) | (4) | ||||
|---|---|---|---|---|---|---|---|---|
| Variable | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. | Coeff. | Sig. |
| Intercept | 0.0440 | *** | 0.0432 | *** | 0.0457 | *** | 0.0456 | *** |
| (2.89) | (2.82) | (2.98) | (2.97) | |||||
| PredLogFREQ | 0.0041 | ** | 0.0039 | ** | 0.0035 | ** | 0.0034 | ** |
| (2.53) | (2.34) | (1.98) | (1.97) | |||||
| LogCON | 0.0002344 | * | ||||||
| (1.70) | ||||||||
| LogCON_D | 0.0002967 | * | ||||||
| (1.78) | ||||||||
| LogCON_R | −0.0000635 | ns | ||||||
| (0.18) | ||||||||
| BLUERED | 4.80E-06 | ns | 7.50E-06 | ns | 1.93E-05 | ns | 1.70E-05 | ns |
| (0.08) | (0.13) | (0.33) | (0.29) | |||||
| LogAT | −0.0016 | *** | −0.0015 | *** | −0.0017 | *** | −0.0017 | *** |
| (3.00) | (2.90) | (3.14) | (3.11) | |||||
| BTM | −0.0026 | ns | −0.0026 | ns | −0.0029 | ns | −0.0028 | ns |
| (0.83) | (0.81) | (0.89) | (0.88) | |||||
| ROA | 0.0271 | * | 0.0270 | * | 0.0269 | * | 0.0268 | * |
| (1.70) | (1.69) | (1.68) | (1.67) | |||||
| INDUSTRY | YES | YES | YES | YES | ||||
| Adj R2 (%) | 1.08 | 1.07 | 1.09 | 1.05 | ||||
| N | 3,596 | 3,596 | 3,596 | 3,596 | ||||
- This table presents an analysis of the relation between excess stock return over CSR event days −1 to 1 and predicted CSR disclosure intensity and political variables. PredLogFREQ in regression 1 is a dummy variable partitioned by the median predicted value of log frequency by regressing LogFREQ on LogCON, BLUERED, and controls (LogRECT, LogCOGS, DURABLE, USELIFE, XAD, LEV, REPURCHASE and INDUSTRY). PredLogFREQ in regressions 1 and 2 is a dummy variable partitioned by the median predicted value of log frequency by regressing LogFREQ on LogCON_D, LogCON_R, BLUERED and other controls. PredLogFREQ in regressions 3 and 4 is a dummy variable partitioned by the median predicted value of log frequency by regressing LogFREQ on controls. *, **, *** indicate significance at the 0.10, 0.05 and 0.01 levels, respectively, using two-tailed tests, and ns indicates not significant. t-statistics and p-values are calculated using clustered standard errors by company and year. See the : 1 for the variable definitions.
where XRETi = the sum of the return over days −1 to 1 for stock i less the return of the value-weighted stock market portfolio (cumulative market-adjusted return), where day 0 is the CSRwire release date, and the political and other variables are defined as before. Model 2 is therefore an extension of a conventional event study, which would predict that mean XRET ≥ 0 on CSR disclosures on event days −1 to 1, since given a choice, companies would choose not to disclose if expecting an ex ante negative return. Model 2 extends the analysis to test whether both α > 0 and PredLogFREQ and/or the political variables reflect a positive association with the stock price response across our company observations, based on the notion that investors understand and respond to the expected benefits of CSR disclosure intensity and political contributions, which are partially revealed to investors through the channel of a CSRwire disclosure.27
Table 4 shows two key results in this regard. First, we observe a positive intercept coefficient (α > 0), which is significant in all four regressions. Hence, on average, stock prices respond positively to CSR news around days −1 to 1 after controlling for other factors. This is essentially the same result established in prior event study research, but with the inclusion of several additional factors to control for differential investor response across the sample. We also reproduce this result for our sample of CSR disclosures. For example, untabulated analysis shows a positive and significant market response to CSR disclosures over days −1 to 1 relative to zero (p = 0.0455) and relative to nonannouncement excess returns over days –20 to 20 excluding days –1 to 1 (p = 0.0433). This response strengthens for small companies relative to zero (p = 0.0427) and is significantly greater than zero for environmental disclosures (p = 0.0978) but not for social or governance disclosures (based on CSRwire's classification into these categories).
Second, Table 4 shows uniformly positive and significant γ coefficients for PredLogFREQ. Hence, as variables that help explain disclosure intensity, the political and implicit claims variables that make up expected disclosure intensity also indirectly explain excess stock returns. However, the political variables may also have a direct influence on stock returns as distinct from PredLogFREQ, because the implicit claims and political variables, in part, explain <25 per cent of CSR disclosure intensity (Table 3).
We therefore include additional variables in model 2 to reflect corporate individuals’ political contributions and stakeholders’ political interests directly (as well as indirectly through disclosure intensity). This analysis shows positive and significant χ coefficients for the relation between corporate individuals’ political contributions and stock returns around CSR announcement date; these significant χ coefficients are incremental to the positive and significant γ coefficients. That is, the coefficient in regression 3 for LogCON is significant (χ = 0.0002344, t = 1.70), and the regression 4 coefficient for LogCON_D is also significant (χ = 0.0002967, t = 1.78). Hence, consistent with the earlier results, which imply a stronger demand for CSR disclosure by Dem. individuals, Table 4 also shows that stock prices respond favourably conditional on Dem. individuals’ political contributions. Contrariwise, stock returns show no such positive association conditional on Rep. individuals’ political contributions, because the χ coefficient for LogCON_R in regression 4 is insignificant (χ = −0.0000635, t = 0.18).
The other variables in model 2 act as controls, with results as expected. For example, LogAT has a significantly negative ϕ coefficient (investors respond less to the announcements of larger companies), book-to-market ratio (BTM) has a negative ϕ coefficient (although it is not significant), and ROA has a significantly positive ϕ coefficient, meaning that investors respond to CSR announcements in the same direction as recent financial performance (ROA). These results are robust to INDUSTRY defined as a company's one-digit SIC code or CSRwire's designation as one of 18 categories or when we exclude INDUSTRY from the regression.
We interpret these results as favouring the view of both a direct link between stock returns and corporate individuals’ political interests and an indirect link through CSR disclosure intensity, for if this were not the case, PredLogFREQ would not continue to be significant in regressions 3 and 4 when we add the LogCON political variables. The results in Table 4 also adjust for the possibility of causation running in both directions, although PredLogFREQ shows positive and significant ϕ coefficients in all the regressions in the presence of the controls and political variables, and the event of CSR disclosure apparently produces positive excess returns over disclosure days −1 to 1, primarily for smaller companies and companies with environmental disclosures (discussed earlier).28
Finally, we note that whereas the variable BLUERED helps explain CSR disclosure intensity (Table 3), it has little effect on investors’ response at CSR disclosure date. For example, Table 4 shows insignificant coefficients for BLUERED. Thus, we find that company individuals’ political contributions associate significantly and positively with higher stock prices, but this is not so for stakeholders’ interests based on the presidential voting percentage in the state of the company's headquarters. Simply stated, higher corporate political contributions in the aggregate apparently pay off in higher stock returns to company shareholders, whereas presidential voting patterns in the state of the company's headquarters are inconsequential for investors. In the latter case, we conjecture that this could occur if it is unclear to investors how statewide presidential voting patterns might bestow benefits on shareholders (and other stakeholders), either through the actions of company individuals or the perceptions of market investors.29 Also there could be offsetting return effects due to market participation, as Republicans have higher stock market participation rates than Democrats (Bonaparte and Kumar, 2013; Table 2).
| Holding period | N | 0–20 days | 0–30 days | 0–90 days | 0–20 days | 0–30 days | 0–90 days |
|---|---|---|---|---|---|---|---|
| Statistic | Mean | Mean | Mean | Median | Median | Median | |
| Panel A: Portfolio return on company size | |||||||
| Small | 2,235 | 0.003 | 0.006 | 0.019 | 0.001 | 0.004 | 0.005 |
| Large | 2,243 | 0.001 | 0.002 | 0.004 | −0.002 | −0.002 | −0.001 |
| Difference (%) | 0.20 | 0.40 | 1.50 | 0.30 | 0.50 | 0.60 | |
| p-value for difference | 0.450 | 0.298 | 0.018 | 0.280 | 0.086 | 0.065 | |
| Sig. | ns | ns | ** | ns | * | * | |
| Panel B: Portfolio return on company size and disclosure intensity | |||||||
| Small/high frequency | 843 | 0.005 | 0.009 | 0.029 | 0.005 | 0.009 | 0.029 |
| Large/low frequency | 846 | −0.005 | −0.004 | 0.000 | −0.005 | −0.004 | 0.000 |
| Difference (%) | 1.00 | 1.20 | 2.90 | −1.00 | 1.20 | 2.90 | |
| p-value for difference | 0.017 | 0.019 | 0.001 | 0.005 | 0.006 | <0.0001 | |
| Sig. | ** | ** | *** | *** | *** | *** | |
| Panel C: Portfolio return on company size, disclosure intensity, and Blue/Red state | |||||||
| Small/high frequency/Blue | 422 | 0.006 | 0.011 | 0.017 | 0.003 | 0.008 | 0.026 |
| Large/low frequency/Red | 423 | −0.001 | 0.001 | 0.023 | −0.002 | −0.012 | 0.003 |
| Difference (%) | 0.70 | 1.00 | −0.70 | 0.50 | 2.00 | 2.30 | |
| p-value for difference | 0.28 | 0.242 | 0.633 | 0.1 | 0.041 | 0.023 | |
| Sig. | ns | ns | ns | ns | ** | ** | |
| Panel D: Portfolio return on company size, disclosure intensity, and political contributions | |||||||
| Small/high frequency/high contribution | 145 | 0.014 | 0.011 | 0.053 | 0.012 | 0.002 | 0.037 |
| Large/low frequency/low contribution | 186 | −0.003 | 0.003 | 0.008 | −0.003 | −0.01 | −0.013 |
| Difference (%) | 1.70 | 0.80 | 4.50 | 1.40 | 1.20 | 5.00 | |
| p-value for difference | 0.117 | 0.580 | 0.056 | 0.048 | 0.051 | 0.023 | |
| Sig. | ns | ns | * | ** | * | ** | |
- Panel A presents the buy and hold return of long small companies and short large companies over an interval of 20, 30 and 90 days, respectively. Panel B presents the buy and hold return of long small and disclosure intensive companies and short large and nondisclosure intensive companies over a interval of 20, 30 and 90 days, respectively. Panel C presents the buy and hold return of long small, disclosure intensive, and Blue state companies and short large, nondisclosure intensive, and Red state companies over a interval of 20, 30 and 90 days, respectively. Panel D presents the buy and hold return of long small, disclosure intensive, and high contributions companies and short large, nondisclosure intensive, and low contribution companies over an interval of 20, 30 and 90 days, respectively. *, **, *** indicate significance at the 0.10, 0.05 and 0.01 levels, respectively, using two-tailed tests, and ns indicates not significant.
4.3. Portfolio excess returns following CSR new release
Table 4 also helps us develop some practical implications for investors in that our regression results suggest factors that investors might use to achieve superior short-run returns. In this case, investors might benefit purchasing the stocks of companies at the time they make a CSR announcement (Table 4 shows significantly positive intercept coefficients) where the stocks purchased relate to companies that are (i) smaller in size (Table 4 shows significantly negative LogAT coefficients), (ii) have a longer history of CSR disclosure (Table 4 shows significantly positive LogFREQ coefficients) and (iii) have corporate individuals that contribute to registered political committees (Table 4 shows significantly positive coefficients for LogCON and LogCON_D).
We test this practical investment strategy by forming hypothetical portfolios at the time of CSR disclosure based on subsets of companies in the above- and below-median partitions of factors (i) through (iii) and holding the resulting extreme portfolios formed on CSR release day 0 for the next 20, 30 or 90 days.30 We take a long (short) position if the variable relates positively (negatively) to announcement period excess stock returns as per our Table 5 analysis. For example, if we focus on two factors, we might choose to buy small companies with high disclosure intensity and short sell large companies with low disclosure intensity. We measure the pay-off as the mean/median cumulative excess return from disclosure day 0 to the end of the holding period for stocks in a long position less the mean/median cumulative excess return from day 0 for stocks in a short position.31
Table 5 presents the results in four panels. Focusing on the 90-day holding period, the hypothetical portfolio earns a net 1.5 per cent excess return based on low and high size (panel A), 2.9 per cent net excess return based on low and high size and disclosure frequency (panel B), and 4.5 per cent net excess return based on low and high size, disclosure frequency and political contribution (panel D). On the other hand, a portfolio formed on Red state or Blue state earns an insignificant net excess return (panel C), but we would have expected this given that BLUERED is insignificant in Table 4. In sum, Table 5 documents that a hypothetical investment strategy that is long and short in equal amounts and neutral to overall market trends produces a positive mean and median excess return to investors; although we caution that these results are retrospective and, thus, may not be indicative of future results. Additionally, these positive excess returns could reflect compensation for additional stock market risk, because they reflect purchases of small companies (which tend to be riskier than large companies), and Table 4 shows negative coefficients for LogAT. On the other hand, Dhaliwal et al. (2011) show that increased CSR disclosure associates with a reduction in the future cost of equity, so that the positive excess returns we observe in Table 5 could also result from a cost of equity reduction and/or the underlying mechanisms that might support this reduction such as the diffusion of knowledge about companies that have higher CSR intensity. For instance, CSR disclosers might experience future increased institutional ownership and analyst following or signal such through CSR disclosure (Lys et al., 2012), and this could lead to higher market returns.
5. Additional tests
5.1 Non-CSR voluntary disclosure
As an additional test, we apply model 1 to a sample of non-CSR voluntary disclosures made by the same companies. We select MEG disclosures as our candidate. If companies publish MEG subject to similar incentives and constraints relating to voluntary CSR disclosure, then the coefficient signs of the factors that explain CSR and MEG disclosure should be the same, although not necessarily the same in magnitude. Contrariwise, a finding of different coefficient signs would threaten to our results, because voluntary disclosure theory does not explain why the CSR coefficients would switch in sign for MEG disclosures.
To test this idea, we extract MEG disclosures from First Call's Company Issued Guidance database for the same study period and same sample CRSP identifiers. This step identifies a sample of 17,326 company-announcement observations with disclosure days distributed over 2000 to 2011. We then add the constraint that an MEG disclosure shall not include an earnings announcement, as these result from mandated disclosure. This second step identifies a sample of 12,127 company-announcement MEG disclosures. These disclosures also have the feature that the calendar year distribution of the disclosure days differs from the CSR sample, for example, MEG disclosures cluster more around earnings announcement dates. Third, we calculate MEG disclosure intensity in the same way for CSR releases, that is, we define the variable LogFREQ_MEG for each company as the natural log of the number of MEG disclosures up to and including the most recent MEG disclosure between 2000 and 2011.
We then run the same regressions as in Table 3, but now with LogFREQ_MEG as the dependent variable. Untabulated results indicate the following. First, similar to Table 3, the political variables BLUERED, LogCON, LogCON_D and LogCON_R*BLUERED reflect uniformly positive (and mostly significant) coefficients. Hence, the coefficients for the political variables that explain MEG disclosure have the same sign as the coefficients for the political variables that explain CSR disclosure. Similarly, the coefficients for the implicit claims variables (except for LogRECT) and the financial slack variables have the same sign for MEG and CSR disclosure. But as noted earlier, we expected similar coefficient signs to the extent that both disclosure types reflect common factors, for instance, voluntary disclosure incentives and constraints, although these factors would not necessarily influence disclosure intensity with the same response magnitude.32 Another reason for similar signs is that our data on company individuals’ political contributions and state-wide voting do not differentiate between political contributions intended to affect CSR activities (and potentially reflected in CSR disclosures) and political contributions intended to affect non-CSR activities (and potentially reflected in MEG disclosures).
5.2 Alternative test procedures and variable definitions
First, we repeat the analysis in Table 3 by conditioning on firms’ past environmental performance. This could attenuate the effect of political interests on CSR disclosure intensity, although the studies are mixed as to whether we should observe a positive or negative relation between environmental performance and environmental disclosure (Patten, 2002; Al-Tuwaijri et al., 2004; Clarkson et al., 2008). We proxy for environmental performance using the previous year's abnormal greenhouse gas emission (actual or estimated GHGE minus the industry median), and estimate abnormal GHGE following the steps in Griffin (2013). Untabulated results show that the coefficients for the political variables in Table 3 remain qualitatively the same, that is, we still find positive and significant coefficients for BLUERED and LogCON when we include an indicator variable for high or low environmental performance in model 1 or interact that indicator variable with BLUERED or LogCON. The coefficients for the implicit claims variables remain qualitatively similar also. Interestingly, we also observe a negative and significant fixed effect for the environmental performance indicator variable. Consistent with the prior work (cited above), this suggests that companies with higher/lower than expected GHGE tend to have lower/higher CSR disclosure intensity, apart from the effects of the other variables.33
Some have also advanced the view that litigation risk might condition CSR disclosure, based on the notion that additional CSR disclosure may reduce investors’ needs to elicit risk information in more costly ways (Cormier and Magnan, 1999). Similar to our tests conditioning on environmental performance, we establish an indicator variable to proxy for litigation risk based on the presence or absence of an Accounting and Auditing Enforcement Release in the 5 years prior to a CSR disclosure and include that variable in model 1 as a fixed effect and interact it with BLUERED and LogCON.34 Untabulated analysis shows that the coefficients for BLUERED and LogCON continue to be positive and significant as per Table 3, and the litigation risk interaction coefficients are not significant.
We also replicate Tables 3 and 4 using alternative definitions of the variables. In general, these additional tests do not change our findings and conclusions. We re-estimated models 1 and 2 with CON (and CON_D and CON_R) scaled by total assets (AT). This analysis generated coefficients for the scaled political variables of the same sign, although none was significant. But this could have resulted from using inconsistent scaling on the right and left hand side of model 1. We considered different forms of BLUERED by shifting the election year forward 2 years, so that, for example, the Bush-Kerry 2004 election results apply to 2003–2006 rather than 2005–2008. We examined different models for PredLogFREQ by excluding the political variables in the estimation model. We re-estimated model 2 with LogFREQ instead of PredLogFREQ to check that our two-stage approach did not produce dissimilar results and found this to be the case. We also considered the following alternative definitions of the implicit claims variables, with no material change in the results. Instead of LogRECT as a proxy for customers’ claims, we tested RECT/AT. Instead of LogCOGS as a proxy for claims for suppliers, we tested the mean of COGS/AT from years t−2 to t or COGS/AT at t. Instead of XAD as a proxy for stakeholders’ claims or customer interests, we tested the mean of XAD/AT from t−2 to t or XAD/AT at t. Instead of LEV or REPURCHASE as a proxy for financial slack, we tested the Kaplan-Zingales (KZ) score as used in Hong et al. (2012). Finally, we re-estimated model 1 using total CSR frequency over 2000–2011 as the dependent variable, rather than cumulative disclosure frequency from 2000 up to and including the most recent CSR disclosure and found that this alternative definition did not materially change the results in Tables 3 and 4.
6 Conclusions
This paper derives new results about companies’ voluntary CSR disclosure based on a unique and comprehensive data set of voluntary disclosures from CSRwire, a leading distributor of CSR news. First, we investigate companies’ decisions to disclose CSR news and test whether politics might be influential in explaining why some companies have higher CSR disclosure intensity than others. We provide new evidence that corporate individuals’ contributions to federal election committees and presidential voting in the home state of the disclosing company significantly explain CSR disclosure intensity. In addition, we find that CSR disclosure intensity is driven mostly by corporate individuals’ contributions to Democratic committees whose stakeholders reside in a state favouring the Democratic presidential candidate (Blue state). While this result resembles Rubin (2008) and Di Giuli and Kostovetsky (2012), who find that companies with Democratic CEOs have higher survey ratings than companies with Republican CEOs, it goes further by linking date-stamped disclosures to political interests rather than CSR ratings from surveys.
Second, we document new evidence of a positive relation between political contributions by company individuals and investors’ short-term response to CSR news, especially for Democratic company individuals. This result comports with the theory espoused by Cooper et al. (2010) whereby managers (and investors) view companies’ political contributions as positive net present value expenditures, whose pay-offs to the company are, in part, revealed through the mechanism of investor’ response to CSR news. We also find that this response strengthens for companies with Democratic political interests, which we posit occurs because of the connectedness of Democratic regulators and legislators and Democratic company individuals and/or the likelihood that they share common beliefs that CSR regulation, legislation and disclosure carries benefits for companies and their constituents. Other factors or mechanisms could explain this result also, such as a link between politics and the cost of equity capital, which could arise directly or indirectly through the role of Democratic company individuals or voting in Blue states as determinants of CSR disclosure frequency. Third, we document a favourable or neutral stock market response to an array of corporate CSR news releases distributed through the CSRwire news service. This evidence updates previous work that has focused on a few categories only, which relate mostly to greenhouse gas emissions.
While each result is significant by itself, the combination of all three offers an interesting narrative on the choices and consequences of CSR reporting; for it suggests that, unlike earlier studies of how politics might influence CSR activities, corporate individuals’ political contributions appear to have a distinct impact on shareholder value when viewed through the lens of the intensity of companies’ voluntary CSR disclosures. These choice variables and the economic consequences thereof, moreover, associate reliably with contributions to federal committees by Democratic individuals who work at companies headquartered in Blue states.
We also show practical results by documenting significant and positive portfolio returns for up to 3 months following CSR disclosure when we exploit corporate individuals’ political orientation, CSR disclosure intensity and company size. These positive excess stock returns following CSR disclosure cannot be dismissed as a chance occurrence and, therefore, challenge the widely held belief that money does not seem to curry favour with politicians and regulators to improve the standing of shareholders (Porter, 2012). Perhaps in the broadest sense, this belief may be true. That said, it also seems inconceivable that individuals at the companies we study would have spent millions of political dollars for altruistic reasons alone. Political expenditures in conjunction with increased CSR disclosure could well serve the cause of building a sustainable brand or enhancing CSR reputation. In addition, the tests in the prior literature may have led to nonrejection of the null hypothesis (of no impact on investment returns) because they lacked control for the myriad of reasons why companies and company individuals make political contributions in the first place. Our focus on how politics affects CSR disclosure, on the other hand, narrows the heterogeneity and, therefore, provides a more powerful setting.
Finally, our work suggests new avenues for future research. For example, we would benefit from additional knowledge about the disclosure mechanisms by which federal election committee contributions might manifest themselves in investors’ beliefs and, hence, stock market returns. Such mechanisms and contributions, which seem to vary by political party, might also vary across companies and over time and further depend on factors such as product market competition, industry structure, unionization, compensation incentives and current and impending regulation. Lastly, missing variables of which we are unaware, which happen to mirror the political and other variables that we examine, may confound our results. We encourage additional research in this area, especially to understand at a less granular level the causes and consequences of what we document in this study.
Notes
Appendix :
Variable Definitions
| BLUERED | The percentage of votes cast by Democrats minus the percentage of vote cast by Republicans in the 2000, 2004 and 2008 general elections in the state in which the company has its headquarters |
| BTM | Company i's book value of equity scaled by market capitalization as of the fiscal year-end prior to CSR disclosure year |
| DURABLE | 1 if company i is in industries constructing and manufacturing durable products (SIC codes 150–179, 245, 250–259, 283, 301 and 324–399), 0 otherwise |
| INDUSTRY | 1 if company i is classified as in a one-digit SIC code industry code 0, 1, …, or 9, otherwise 0 |
| LEV | Company i's long-term debt scaled by its total assets as of the fiscal year-end prior to CSR disclosure year |
| LogAT | Company i’s natural log of total assets (AT) as of the fiscal year-end prior to CSR disclosure year |
| LogCOGS | Company i's natural log of cost of goods sold as of the fiscal year-end prior to CSR disclosure year |
| LogCON | Company i’s natural log of total individuals contributions to registered Democratic or Republican committees as of CSR disclosure year |
| LogCON_D | Company i's natural log of total individuals contributions to registered Democratic committees as of CSR disclosure year |
| LogCON_R | Company i's natural log of total individuals contributions to registered Republican committees as of CSR disclosure year |
| LogFREQ | Company i's natural log of the number of CSR disclosures up to and including the most recent CSR disclosure between 2000 and 2011 |
| LogFREQ_MEG | Company i's natural log of the number of management earnings guidance disclosures up to and including the most recent MEG disclosure between 2000 and 2011 |
| LogRECT | Company i's natural log of receivables as of the fiscal year-end prior to CSR disclosure year |
| REPURCHASE | 1 if company i has purchased common or preferred stocks as of the fiscal year-end prior to CSR disclosure year, 0 otherwise |
| ROA | Company i’s income before extraordinary items scaled by total assets as of the fiscal year-end prior to CSR disclosure year |
| USELIFE | Company i's natural log of property, plant and equipment as of the fiscal year-end prior to CSR disclosure year scaled by annual depreciation |
| XAD | Company i's advertising expenditure scaled by total assets as of the fiscal year-end prior to CSR disclosure year |
| XRET | Sum of the return over days −1 to 1 for stock i less the return of the value-weighted stock market portfolio (cumulative market-adjusted return), where day 0 is the CSRwire release date |
