Does social responsibility begin at home? The relation between firms’ pension policies and corporate social responsibility (CSR) activities

In Appendix 1, we list other examples of pension sponsors stepping up CSR activities following pension freezes.

While the CSR response is aimed at local communities, the intended audience for reputation-repair could be much broader than the stakeholder groups that specifically benefit from the CSR responses. All stakeholders who care about firms’ treatment of workers could notice and respond with sanctions—socially responsible investors could reduce their holdings, customers could boycott the firm’s products, and local communities could protest. Resulting media attention could bring scrutiny from politicians and regulators. This is in addition to employees, who are directly hurt by benefit cuts. Therefore sponsoring firms have strong incentives to take actions so as to repair their reputation as good corporate citizens with all these stakeholders. Theory does not specify a particular stakeholder with whom the firm will try to repair reputation, although certain stakeholders could be more important to certain firms.

Statement of Financial Accounting Standards (SFAS) No: 158 Employers’ Accounting for Defined-Benefit Pension and Other Postretirement Plans.

Godfrey et al. (2009), Wans (2017), and Barnett et al. (2018) document insurance-like consequences from CSR in archival empirical settings. Bolton and Mattila (2015), Joireman et al. (2015), and Eisingerich et al. (2011) demonstrate insurance-like effects of CSR in an experimental setting.

They document a sharp increase post restatement in what they label reputation-repair actions, targeting customers (e.g., improved warranty programs), employees (e.g., investing in winning “Best Employer” awards), and the community (e.g., charitable contributions and involvement with nonprofit organizations).

Notwithstanding this trend, defined-benefit pension plans remain economically important in the United States. These plans, across corporate and governmental sectors, controlled assets worth $8 trillion in 2013 (Towers Watson 2014). In 2017, the aggregate dollar amount of pension obligations (assets) for all U.S. defined-benefit sponsors on Compustat was $2.7 trillion ($2.3 trillion). From employees’ perspective, 37 million private sector employees and retirees rely on one of the 25,000 defined-benefit plans sponsored by the Pension Benefit Guaranty Corporation (PBGC) for retirement income. PBGC Annual Report 2018, available at: https://​www.​pbgc.​gov/​sites/​default/​files/​pbgc-annual-report-2018.​pdf.

In cash-balance plans, the employee’s account is credited each year with a pay credit (e.g., 5% of annual compensation) and an interest credit (which could be fixed or floating). Like traditional defined-benefit plans, they are protected by the Pension Benefit Guaranty Corporation, and the employer bears investment risks. Unlike traditional defined-benefit plans, which define the benefit as a series of monthly payments payable for life starting at retirement, cash-balance plans define the benefit as a stated account balance; e.g., on retirement, the participant will have an account balance worth $100,000, which she could choose to take as a lump-sum or convert to an annuity.

The literature examining why sponsors freeze defined-benefit plans has found some support for both supply-side and demand-side motivations for pension freezes. Among the supply-side motivations, sponsors are unwilling to continue defined-benefit plans because they impose too heavy or too volatile of a cash-flow burden for contributions (Munnell and Soto 2007).

In our sample there are 313 individual firms that freeze pension plans. Of these, 38 firms engage in multiple freezes. Dropping the second freeze of these firms helps more clearly isolate the impact of the first freeze, by ensuring that the post-freeze period of the first freeze is not tainted by pre-freeze period of the later freeze. Our results are also robust to dropping altogether the firms with multiple freezes.

KLD’s manual, available at https://​www.​msci.​com/​www/​research-paper/​esg-ratings-methodology/​0175943017, indicates that KLD analysts gather CSR information from a variety of public sources. MSCI claims that the ratings are used by over 1000 institutional investors. Recent examples of studies using KLD ratings include Lins et al. (2017) and Davidson et al. (2019).

We list examples of CSR-related keywords from the dictionary in Table 9 of the online appendix.

Available from http://​www.​amckenny.​com/​CATScanner/​

We exclude keywords from the employee relations category to be consistent with our focus on non-employee categories from KLD ratings.

Table 10 of the online appendix tabulates descriptive statistics about the persistence of CSR engagement over this longer window. Given the increasing trend in CSR engagement over time, these patterns are only suggestive.

The mapping of our sample to CSRWire is low for two reasons. Our sample starts in 1995, whereas CSRWire coverage starts from 2000. And our sample has many small firms, whereas CSRWire focuses on larger ones.

To keep data collection efforts manageable, we focus our manual research of CSRNewswire and Global Reporting Initiative only on the treatment firms. Since we do not have relevant data for the control group, we cannot rule out the possibility that the patterns in Table 3, Panels B–C, are affected by a trend of issuing more press releases/CSR reports over time. CSR reports have become more prevalent over time: by 2018, 86% of S&P 500 firms issued CSR reports (https://​www.​sustainability-reports.​com/​86-of-sp-500-index-companies-publish-sustainability-responsibility-reports-in-2018/​), whereas only 20% did so in 2011. Therefore we caution that the patterns in Panels B–C are only suggestive.

A better way to identify employers trying to “do the right thing” by their employees would be to identify those that offer employees a choice (to move to the new benefit structure or not). Unfortunately, due to the voluntary nature of such disclosure, only four employers explicitly mentioned a choice, too few for meaningful results.

See Bulkeley and Schultz (2003), Johnston (2004), and U.S. Senate Hearing 106–849 (2000).

Perceptions of unfairness were also heightened by firms’ attempts to promote cash-balance conversions to their employees. While defined-benefit freezes are often presented as a painful but necessary step to remain competitive, cash-balance conversions were presented to workers as a beneficial change. Janet Krueger, lead spokesperson for the IBM Employee Benefits Action Coalition, said in congressional testimony: “IBM told us they did the conversion to retain older workers.” She also said IBM claimed “employees would receive frequent statements listing their earned cash balance” and that “a cash balance plan is much easier to understand.” Krueger argued that none of these claims turned out to be true and that IBM removed a pension calculator from its internal website, inhibiting employees’ ability to grasp exactly how their benefits were changing. Pension consultants even touted as a “handy” feature of these plans that it was “difficult for employees to compare prior pension benefit formulas to the cash-balance approach”(Crenshaw 1999). So when employees eventually realized that they had lost substantial benefits, widespread anger resulted (Madland 2007).

The net effect to the firm from freezing defined-benefit pensions is not unambiguously clear. On one hand, pension freezes yield cost savings for firms (e.g., Rauh et al. 2013), which benefit stockholders. On the other hand, offsetting costs arise from loss of employee morale, difficulty in attracting and retaining employees, and reputational harm, which are harder to quantify. Unsurprisingly, empirical work with recent samples produces mixed results. Milevsky and Song (2010) find positive announcement-period returns, while McFarland et al. (2009) find negative or insignificant announcement-period returns. It is not always easy to identify the announcement date (Cocco 2014); for this reason, we do not examine announcement-period returns but rather returns over the fiscal year of announcement.

Theory does not specify a particular stakeholder with whom the firm will try to repair its reputation; however, certain stakeholders could be more important for certain firms. While developing these predictions is beyond the scope of our paper, we test whether the CSR response is stronger for consumer-facing firms, using proxies developed by Lev et al. (2010) and Servaes and Tamayo (2013). We find no meaningful results.

Prior to SFAS 158, the funding status (projected benefit obligation or PBO minus fair value of plan assets) was only disclosed in footnotes to the financial statements but not recognized on the balance sheet. On balance sheet, the funding status based on a smaller estimate of the obligation (the accumulated benefit obligation, ABO), was recognized but only if the plan was underfunded on an ABO basis. ABO-basis underfunding (i.e., ABO minus fair value of plan assets) was recognized directly through the statement of changes to stockholders’ equity, as a dirty-surplus item.

A body of literature supports the notion that recognition brings more visibility than footnote disclosure alone (e.g., Bernard and Schipper 1994; Aboody 1996). Yu (2013) provides evidence consistent with increased investor awareness of pension liabilities post SFAS 158.

Axenrod and Kisser (2020) demonstrate no significant reduction in affected sponsors’ leverage following SFAS 158; they also find no significant increase in interest costs for these firms, consistent with early evidence from Shaw (2008). Donovan et al. (2019) document no significant increase in default risk for underfunded sponsors post-SFAS 158. Collectively, these studies suggest that SFAS 158 did not discernibly reduce sponsors’ borrowing capacity.

Two tests in the paper provide evidence supporting the reputation-repair interpretation over the “freezes free up resources” interpretation. First, the strong increase in CSR following cash-balance conversions. These conversions do not clearly free up additional resources, relative to other kinds of freezes. However, they were especially controversial. Second, the CSR increase following SFAS 158, which potentially increased pension visibility but did not necessarily free up financial resources.

We tabulate all results discussed in this section in Tables 10, 11 and 12 of the online appendix.

Kotsantonis and Serafeim (2019) discuss many issues with CSR-related data, such as inconsistencies in the way companies report CSR metrics, distortions created by different peer group choices, varying methods for imputing missing metrics, and disagreements among ratings. In addition to the robustness tests discussed above, (i) we document that CSR disclosures, measured as Ln(TOTAL CSR WORDS), increase significantly following pension freezes—to the extent to which CSR performance and CSR disclosure are correlated, finding similar results with our word count-based measure (that is constructed very differently from KLD ratings) serves as a validity check that our results with KLD ratings are not artifactual to KLD’s data scoring procedures; (ii) our findings are robust to using an alternative data source to measure CSR performance, Asset4.

We document results discussed in this section in Tables 12 and 13 of the online appendix.