Do LGBT-Supportive Corporate Policies Improve Credit Ratings? An Instrumental-Variable Analysis

The Don’t Ask, Don’t Tell Repeal Act of 2010 established a process for ending the Don’t ask, don’t tell policy, thus allowing gays, lesbians and bisexuals to serve openly in the United States Armed Forces. It became effective in September 2011.

In Obergefell v. Hodges (2015), The Supreme Court of the United States ruled that the fundamental right to marry is guaranteed to same-sex couples by both the Due Process Clause and the Equal Protection Clause of the 14 Amendment. The 5–4 ruling requires all 50 states to perform and recognize the marriages of same-sex couples on the same terms and conditions as the marriages of opposite-sex couples with all the accompanying rights and responsibilities.

2 Weeks after the announcement was made by President Trump, two lawsuits were filed by The National Center for Lesbian Rights (NCLR) and GLBTQ Legal Advocates and Defenders (GLAD). These lawsuits resulted in a nationwide preliminary injunction halting the ban while it was being heard by the court. To date, four lawsuits have been filed against the ban, each securing a preliminary injunction. In February 2018, The Pentagon released guidelines where the ban against transgender individuals was much more limited than earlier announced. Even the limited restrictions are being litigated.

Cespa and Cestone (2007) and Surroca and Tribo (2008) argue that managers use CSR as an entrenchment strategy, making it more difficult more shareholders to remove them. They find empirical evidence in favor of this argument. Oh et al. (2011), using a sample of Korean firms, report that managerial ownership is inversely related to CSR, implying that insiders may use their equity ownership to obtain personal benefits at the expense of other stakeholders. Borghesi, Houston and Naranjo (2014) find that CSR investments are related to CEO-specific attributes. They argue that CSR is not driven solely by altruistic reasons. Rather, it is motivated, at least partially, by personal reasons specific to each individual CEO. Finally, Chintrakarn, Jiraporn, Kim, and Kim (2016), using the governance metrics provided by the Institutional Shareholder Services (ISS), find that firms with more effective governance invest significantly less in CSR, suggesting that managers over-invest in CSR and are forced to cut back when corporate governance is more effective.

Additional studies that support the agency view of CSR include Cespa and Cestone (2007), Surroca and Tribo (2008), Borghesi et al. (2014), Oh et al. (2011), and Chintrakarn et al. (2016).

The Gallup conducted another survey earlier in 2012–2013 on sexual orientation with a smaller sample of 473,243. According to Gallup, there is a modest increase in the percentage of LGBT populations between two surveys. As acknowledged by Gallup, however, most of the state-level changes between the two surveys are not statistically significant (the correlation of the percentages of LGBT population by state between the two surveys is 99.93%). The results based on the earlier estimate from the 2012–2013 Gallup survey remain similar.

Our sample period ends in 2011, given that KLD discontinued the coverage of LGBT-friendly policy (div-str-g) in 2012.

Both the IV analysis and the two-stage selection models are well known in prior studies (e.g., Antonakis et al. 2010) to be useful in establishing causality when there is possible endogeneity. Both of them also rely on instrumental variables with simultaneous equations. The two-stage selection models, however, are often criticized for being overly sensitive to changes in model specifications, resulting in non-robust, inconsistent, and unreliable inferences (Lennox et al. 2012). On the other hand, the IV analysis is widely-recognized, and used extensively in economics and finance because it is useful in coping with various sources of endogeneity including, but not limited to, the omitted-variable bias, reverse causality, and measurement errors (Antonakis et al. 2010). For the aforementioned reasons, we prefer the IV analysis over the two-stage selection models in our study.

We use Institutional Shareholder Services (ISS, formerly known as RiskMetric) database to calculate board independence and board size, respectively.

The KLD database reports data on several aspects of CSR, such as diversity, employees, products etc. The overall CSR score is the sum of the strengths minus the concerns for all of these CSR categories.

An alternative way to control for unobservable characteristics is to run a fixed-effects model. However, our state-level data on the LGBT population is only cross-sectional. So, it is not possible to run a fixed-effects model. It is important to note, however, that the LGBT population by state does not change much at all over time. Several other variables in our analysis also change slowly over time, such as the propensity for firms to be LGBT-supportive. Even with a panel data set, a fixed-effects analysis is unlikely to produce significant results.

Leif (1984) and Porter (2000) discuss a number of factors that determine the locations of headquarters. For instance, the following factors are often considered: firm size, stage of development, proximity to major customers or suppliers, access to various services, access to information, marketing complementarities etc. The sexual orientation of the population in a specific location is not one of the factors.

COMPUSTAT reports only the current location of the firm. So, to the extent that some firms may have moved their headquarters, there may be some measurement errors. This problem is not particularly serious as headquarters relocations are quite rare (Pirinsky and Wang 2010; Alli et al. 1991). In any event, to deal with this issue, we execute a number of additional tests. Pirinsky and Wang (2010) find that the vast majority of the headquarter relocations in their sample come from small firms. Large firms rarely relocate their headquarters. We divide our sample into three groups based on total assets. Then, we run a regression on the group with the largest firms and obtain a similar result. Because relocations are rare for this group of large firms, it does not appear that our results are significantly driven by firms that relocate their headquarters. Second, we run a number of error-in-variables regression, where possible measurement errors are recognized. We assume that our data on the number of firms in the same state is only 80% accurate (20% measurement errors). Even with the assumption, we still obtain a consistent result. So, measurement errors do not appear to be a serious problem. Finally, we include only observations in the most recent year of each firm. This is to ensure that the headquarters locations are current and accurate. Again, the results are consistent.

By including the state-average credit ratings, we control for state-specific factors. Nevertheless, as robustness checks, we also execute a regression analysis that includes state-specific variables such as population, population change, and population density. The results remain consistent.

We put the Employee-friendly Score (Residuals) in place of CSR Score (Excluding LGBT-friendly) in the analysis to cope with possible multicollinearity among them.