Trust, social capital, and the bond market benefits of ESG performance

To construct our sample of corporate bonds on the secondary market, we start with the universe of bonds covered in the Enhanced Historic TRACE database from January 2006 to September 2019. Following Dick-Nielsen et al. (2012), we exclude variable- and zero-coupon, perpetual, foreign currency, preferred, puttable, and exchangeable issues as well as private placements and Yankee and Canadian bonds. We further restrict our selection to corporate debentures and corporate medium-term notes with a time-to-maturity of more than one month and 30 years or less. To be included in our sample, we require that data on fundamental bond contract attributes (i.e., issue size, offering and maturity dates, and coupon rates) are available on the Mergent Fixed Income Securities Database (FISD). To address liquidity biases and erroneous entries in TRACE, we follow the method in Dick-Nielsen (2014).⁶ We further apply the quantity, price, and yield filters used in Becker and Ivashina (2015) and Han and Zhou (2016) to remove outliers and observations with likely data errors.⁷

We merge this sample with E&S ratings data from the Thomson Reuters Refinitiv ESG database, which contains annual environmental, social, and governance ratings of large publicly listed companies, covering over 80% of global market capitalization, across more than 450 different ESG metrics. The data are compiled using information from corporate annual reports, sustainability reports, nongovernmental organizations, company websites, stock exchange filings, and news sources for publicly traded companies at an annual frequency. This database has been used in a number of recent studies examining the causes and consequences of firms’ environmental and social performance (e.g., Dyck et al. 2019; Albuquerque et al. 2020). Its U.S. coverage comprises roughly the largest 3,000 companies.

A potential shortcoming of the Refinitiv ESG database is that some of the historical data may be revised over time to reflect new information, which could affect certain inferences (see Berg et al. 2021). As a sensitivity test, we re-estimate our main specifications using E&S metrics constructed from the MSCI ESG KLD STATS database for our sample of firms. Our main findings hold using this metric as well.^8,9

We obtain annual fundamentals from Compustat, daily stock returns from the Center for Research in Security Prices (CRSP), bond duration from the Wharton Research Data Services (WRDS) bond returns database, bond ratings from Mergent FISD, and probability of default data from the Credit Research Initiative of the Risk Management Institute at the National University of Singapore.

Merging these databases yields a final sample of 5,631 corporate bonds, issued by 808 firms, with secondary market trade data from January 2006 to September 2019, as noted in Panel A of Table 1. Panel B outlines the distribution of the sample across the Fama–French 12 industries. The finance industry constitutes the largest proportion of bond issues (23.1%), while the other sectors have a fairly balanced representation in the overall sample (except consumer durables, which represents only 1.5% of all bonds).

Our main independent variable is the E&S index, which we construct using environmental (E) and social (S) performance data from the Thomson Reuters Refinitiv ESG database. Refinitiv ESG evaluates firms’ environmental performance in three categories: resource use, emissions, and innovation. Social performance is measured in four areas: workforce, human rights, community, and product responsibility. For both E and S performance, Refinitiv provides, for each firm in each year, a score that ranges from 0 to 100, which we use to form our E&S index.¹⁰ Following Albuquerque et al. (2020), we define the E&S index as the average of the Refinitiv E score and the S score. We do not include the governance metric in our E&S index because governance speaks most directly to the relation between the firm and its shareholders, while investments in social capital reflect the relation between the firm and its stakeholders, as captured by E&S performance. However, since prior studies have shown strong governance to be beneficial to bondholders (e.g., Bhojraj and Sengupta 2003; Klock et al. 2005; Bradley and Chen 2011, 2015), we also control for governance in our estimations.

Our main dependent variable is a bond’s credit spread, computed as the difference between the bond’s yield to maturity from the Enhanced Historic TRACE database and the Treasury yield matched by maturity (e.g., Campbell and Taksler 2003; Chen et al. 2007; Huang and Huang 2012).¹¹ As in Becker and Ivashina (2015), we employ the median yield of all transactions taking place on the last active trading day of a given month to compute the bond spreads.

Table 2 provides summary statistics on the characteristics of the bonds in our sample, the E&S index, environmental and social scores, credit spreads, and other control variables. The Appendix contains detailed definitions of all the variables employed in our analyses. Panel A reports the fundamental bond contract features that remain constant over the life of bond. As such, we count each bond issue only once in the summary statistics. The mean issue size in our sample is $711 million with an average annual interest rate just below 4.7%.

Panel B of Table 2 contains the bond characteristics that could vary on a monthly basis. As such, we count each bond-month pair as a separate observation. The bonds in our sample have a mean time-to-maturity of just over 9.8 years (118 months). Credit ratings are converted to numerical values, starting with 1 for AAA ratings through 21 for C ratings. The mean credit rating of 8.2 indicates that the bonds in our sample are rated between BBB and BBB + , on average.¹² There is considerable variation in credit spreads, with an average of just under 179 basis points and a standard deviation of 545 basis points. We do not winsorize the spreads because they are larger during the financial crisis and winsorization would disproportionally affect these observations. However, all of our findings persist if we winsorize spreads at the 99^th percentile, albeit with reduced, but still considerable, economic significance.

Panel C of Table 2 reports summary statistics on firm characteristics. All of them vary annually, except for stock return volatility and probability of default, which are computed monthly. The firms in our sample are large (average market capitalization of $25.4 billion) and profitable (operating income to sales of 20.8%).

In this section, we seek to understand whether the bond market payoffs to firms’ E&S activities are more pronounced when overall trust is low and a firm’s social capital may become more valuable. To do so, we focus on the financial crisis, which constituted an exogenous shock to public trust in corporations, capital markets, and institutions and led to a decline in stock prices and an increase in bond spreads for the vast majority of firms.

The exogenous nature of the financial crisis helps alleviate the endogeneity concerns associated with Eq. (1). The underlying assumption is that the crisis is exogenous with respect to firms’ decisions to engage in E&S activities. In particular, firms decide on the optimal level of E&S investments during normal times, when the probability of a crisis and a decline in overall trust is relatively low. During normal times, some firms do not engage much in E&S activities because they do not view them as worth the cost, while others invest in E&S activities because they expect them to be beneficial. When the crisis hits, the value of social capital that firms have built through E&S investments becomes apparent. For firms that invested little in E&S during the pre-crisis period, however, it is too late to make such investments, as corporate social capital takes time to build.¹⁶

Our sample period for this analysis begins in January 2007, prior to the onset of the crisis, and ends in September 2019. We adopt a quasi-difference-in-differences approach and examine whether firms that entered the crisis period with higher E&S ratings enjoyed relatively lower spreads during the crisis.¹⁷ Specifically, we estimate the following model:

where, as before, Credit spread_ijt denotes the spread of firm i’s bond j in month t, X_ijt-1 is a (K×1) vector of bond-level controls measured at time t-1, and Z_it-1 is a (L×1) vector of firm-level controls measured at time t-1. We include firm fixed effects, FFE_i, to control for unobservable time-invariant credit risk factors, and time fixed effects, TFE_t, specified at the monthly level.¹⁸ We measure the E&S index as of year-end 2006, well before the onset of the financial crisis, to eliminate the concern that firms might have adjusted their E&S activities in anticipation of the crisis.¹⁹Crisis_t is an indicator variable that takes the value of 1 for the crisis-of-trust period, which starts in August 2008 and ends in March 2009 (as in Sapienza and Zingales 2012 and Lins et al. 2017), and Post-crisis_t is an indicator variable that takes a value of 1 from April 2009 to September 2019. As before, we double cluster the standard errors at the firm and time (monthly) levels to control for cross-sectional and time-series dependence, respectively. Inclusion of firm fixed effects and firm and bond characteristics ensures that the crisis-E&S effect is not due to healthier firms spending more on E&S activities and performing better during the crisis.

In Eq. (2), the coefficient on the interaction term E&S index_i2006×Crisis_t, β₁, captures the difference between the effect of E&S performance on credit spreads in the crisis versus the pre-crisis periods (the pre-crisis effect itself is captured by the time and firm fixed effects). The coefficient on the interaction term E&S index_i2006×Post-crisis_t, β₂, captures the difference between the effect of E&S performance on credit spreads in the post-crisis versus the pre-crisis periods. This coefficient could also be negative given that overall trust in corporations, markets, and institutions continued to be low after the crisis for some time. However, in absolute terms, we expect β₁ to be larger than β₂, because the most pronounced erosion of trust occurred during the crisis.

The results from estimating Eq. (2) are reported in Panel A of Table 4. In model (1), we include only firm and time fixed effects and the E&S index interactions. We then control for bond attributes in model (2) and add firm characteristics in model (3). All three models indicate that E&S performance has a statistically and economically significant impact on bond spreads during the crisis. Based on the regressions reported in model (3), a one standard deviation increase in the pre-crisis E&S index is associated with 106 basis points lower spreads during the crisis period.²⁰ The benefit that accrued to high-E&S firms during the crisis disappears in the post-crisis period (the difference between β₁ and β₂ is statistically significant at the 1% level in all three specifications).

In model (4), we also control for corporate governance using the governance pillar score from the Refinitiv ESG database. Prior evidence suggests that better-governed firms have lower bond spreads. These firms also performed better during the crisis (Lins et al. 2013; Nguyen et al. 2015); thus, if governance is correlated with our measure of E&S performance, we could be suffering from an omitted variable bias. The coefficient on the E&S index remains virtually unchanged in this specification; hence, the impact of E&S performance on bond spreads during the crisis cannot be attributed to better governance, as captured by the G pillar score. The governance score itself is not significantly related to bond spreads.²¹

Figure 1 presents our findings graphically. To construct this figure, we partition our sample firms into two groups based on their E&S index at the end of 2006. We then estimate a panel regression of credit spreads as a function of all control variables (including the governance pillar score) as well as separate monthly time dummies for high- and low-E&S firms. The coefficients on these time dummies are akin to monthly intercepts for high- and low-E&S firms. The figure presents the plot of these monthly dummies over time. The variation in the differential between characteristic-adjusted spreads of high- and low-E&S firms over time is striking. For most of the sample period, there are only small differences between the two groups. However, after August 2008, the difference between the two groups shoots up, reaching its maximum level in November 2008. The differential remains high until March 2009, when the stock market hit its lowest point of the crisis; afterwards, the difference declines and the two lines often cross and overlap. The period of August 2008–March 2009 (the shaded area in the figure), when the difference becomes considerable, coincides with the crisis of trust as defined in our regression models.²²

Next, we investigate whether these findings persist for various subsamples and when we divide the E&S index into its environmental and social components. Panels B–D of Table 4 present the regression models separately for nonfinancial firms, financial firms, and financial firms that did not receive TARP funding. For the sake of brevity, we report only the main E&S effects in the tables. We find significant effects for all sets of firms. However, the effect is significantly larger for financial firms, which is not surprising, given that their credit spreads increased much more during the crisis than the credit spreads of nonfinancial firms. After removing financial firms that received TARP funding, the effect becomes even larger.²³ Note that for the latter subsample, we also find a significant effect in the post-crisis period. Given that trust in banks was not fully restored after the financial crisis, a continued negative relation between spreads and E&S performance is consistent with our conjecture.²⁴ In terms of economic significance, increasing the pre-crisis E&S index for financial firms by one standard deviation (18.0) led to lower spreads of 167 basis points during the crisis and 65 basis points in the post-crisis period (based on model (3) of Panel C of Table 4).²⁵

In Panel E of Table 4, we partition the E&S index into its environmental and social components. Each component is significantly related to credit spreads in the specifications that include both firm and bond controls (models (3) and (4)), suggesting that social capital stems from both environmental and social elements. Increasing the environmental component by one standard deviation (25.9) reduces spreads by 49 basis points during the crisis, while increasing the social component reduces spreads by 69 basis points (based on model (4)).

From these analyses, we conclude that the bond spreads of high-E&S firms increased less during the financial crisis than the spreads of low-E&S firms. For financial firms, this effect persisted in the post-crisis period. These findings are consistent with bondholders valuing a firm’s social capital and its “earned trust” more in periods when being trustworthy is particularly important, such as in a crisis of trust.

Next, we conduct further analyses to corroborate that our results are indeed driven by a shock to market-wide trust rather than a shock to the supply of credit. In July 2007, LIBOR rates started to increase dramatically as the solvency of the banking sector weakened, which had a negative impact on the ability of firms to borrow (e.g., Duchin et al. 2010; Ivashina and Scharfstein 2010). This shock to the supply of credit persisted until at least March 2009, and thus partly overlaps with the period during which there was a shock to trust. If high-E&S firms were less affected by the credit crunch, the differential in the spreads that we document could be due to this phenomenon rather than a shock to trust. High-E&S firms may have been more able to borrow over the credit crunch, given that the agency costs of debt argument that we describe could hold in any crisis. Our contention, however, is that if a firm’s E&S investments engender trust, the effect of E&S performance on credit spreads should be particularly salient when trust is more valued.

As discussed, Fig. 1 suggests that the difference in spreads between high- and low-E&S firms becomes noticeable starting in August 2008 and not earlier. However, to investigate bond spreads during the credit crunch more formally, we augment Eq. (2) with an interaction term between the E&S index and the “pure” credit-crunch period, which we define as the period of July 2007–July 2008. During this period, the shock to credit supply had already happened, but the shock to trust had not yet occurred (Sapienza and Zingales 2012; Lins et al. 2017). As in Panels A–E of Table 4, we estimate various specifications of this augmented regression, starting with a more parsimonious model and sequentially adding controls in subsequent specifications. The findings are reported in Panel F of Table 4. Across all models, we find that the impact of the E&S index on credit spreads is much more pronounced during the crisis than in the surrounding periods. There is some evidence of a negative relation during the credit crunch and the post-crisis periods, but with a much lower economic significance. Moreover, the effect of the E&S index on credit spreads is significantly different between the crisis and the credit crunch and between the crisis and the post-crisis periods across all specifications. Overall, the results reported in Panel F of Table 4 indicate that the effect of E&S performance on bond spreads is most pronounced during the loss-of-trust period relative to the surrounding periods.

To better understand the mechanisms behind our findings, we conduct six additional tests. First, we split the sample into two groups based on the probability of default at the end of 2006 and estimate separate models for each subsample. Firms with a higher probability of default have more of an incentive to engage in asset substitution to expropriate their bondholders, since they are closer to bankruptcy. If E&S activities reduce the agency costs of debt, we expect the influence of E&S performance on spreads to be particularly germane for this group of firms. The estimated models include all control variables (including the governance pillar score), equivalent to model (4) of Table 4. The results are reported in models (1) and (2) of Panel A of Table 5. While the effect of the E&S index on bond spreads is significant for both groups of firms, it is much larger for firms with a higher probability of default, and the difference between the two is statistically and economically significant; increasing the E&S index by one standard deviation reduces bond spreads of firms with a high default probability by 99 basis points, compared to 21 basis points for firms with a low probability of default. This result supports the notion that our findings are due to the perception of reduced agency costs of debt in high-E&S firms. We also note that there is some evidence of narrower spreads in the post-crisis period for high-E&S firms with a higher probability of default. This finding is consistent with the fact that, in the post-crisis period, trust had not been fully restored (for example, the trust component of the Global Competitiveness Index of the World Economic Forum was still lower in September 2016 than in September 2007). Thus, social capital remains relevant after the crisis for firms that are more likely to engage in asset substitution.²⁶

Second, we investigate whether the effect of E&S performance on spreads during the crisis is more pronounced in firms with low asset tangibility. Williamson (1988) and Johnson (2003) argue that these firms have more of an opportunity to engage in asset substitution when distress risk increases. If the spreads of high-E&S firms are lower during the crisis than those of low-E&S firms because bond investors expect less asset substitution from high-E&S firms, then this effect should be more pronounced for firms that have more opportunities to shift risk. We investigate this possibility by splitting the sample into two groups according to asset tangibility, defined as property, plant, and equipment (net) divided by total assets. Firms are assigned to a group based on tangibility as of year-end 2006, and this grouping remains unchanged throughout the full sample period. In model (3) of Panel A of Table 5, we report the results of the spreads regression for firms with tangibility below the median (< 18.24%). For this group, the E&S index has a strong negative impact on spreads during the crisis period, but not afterwards. In model (4), we report the results for the high tangibility group. The coefficient on the E&S index×Crisis interaction for this subsample is less than half the coefficient of the low tangibility sample, and the difference between the two coefficients is statistically significant. The fact that our results are much stronger for the subgroup of firms that have more opportunities to engage in asset substitution supports our contention that bond investors believe that high-E&S firms are less likely to take advantage of that opportunity.

Third, we examine whether our results are stronger for firms incorporated in states that provide weaker bondholder protection in case of insolvency. In particular, we use the classification of Wald and Long (2007) and Mansi et al. (2009) to divide states into two groups, depending on whether or not they allow firms with negative book equity to make payouts. Mansi et al. (2009) find that bond yields are higher in states without payout restrictions, which indicates that bondholders penalize firms for the possibility that cash flows will be diverted to shareholders in case of financial distress. The results for this analysis are reported in models (5) and (6) of Panel A of Table 5. The effect of the E&S index on credit spreads during the crisis is significant in both groups, but larger in states where firms face no restrictions on payouts during insolvency. The difference between the two is not statistically significant at conventional levels, however. We also find some evidence that the effect of the E&S index on spreads persists in the post-crisis period for firms incorporated in states without payout restrictions, consistent with the view that social capital has remained important for the firms most prone to asset diversion.

In models (7) and (8) of Table 5, we combine both the tangibility and payout criteria. In model (7), we focus on firms with either low tangibility or no state-level payout restrictions, or both. These firms have higher agency costs of debt, and social capital, for them, is likely more important during the financial crisis. This is exactly what we find. Model (8) includes firms with high tangibility that also face state-level payout restrictions. For these firms, agency costs of debt are lower and social capital is likely to have a smaller influence on bond spreads. The results support this notion, as the coefficient on the E&S index×Crisis interaction is half that of model (7).

Fourth, we study whether our findings are affected by the salience of E&S investments. We focus on two elements that reflect the importance that companies attribute to E&S activities. First, does the company publicly disclose its E&S activities either in the form of a standalone ESG (CSR or sustainability) report or a separate section in its annual report?²⁷ If bond investors find these disclosures to be material, we would expect our findings to be more pronounced for firms that report on their E&S activities.²⁸ Second, does the company have an ESG committee or team? Having such a team would suggest that E&S activities are not peripheral to the company’s strategy and business model. To compile data on these two elements, we employ the Refinitiv ESG database, augmented by two databases: (a) The Corporate Register, the world’s most comprehensive database of ESG reports, and (b) The Corporate Sustainability Assessment database provided by RobecoSAM, an international asset management company focused on evaluating corporate sustainability practices. We obtain this information for year-end 2006 and keep it constant throughout the sample period. In Panel B of Table 5, we report results for sample splits based on these two data items. In the first two columns, we report results for splits based on ESG reporting, and in models (3) and (4) we report splits based on having an ESG committee. The partition based on the availability of an ESG report supports the notion that E&S performance has a significantly stronger effect on bond spreads during the crisis for firms whose E&S activities are more salient. The presence of an ESG committee, on the other hand, has no effect on the E&S index-credit spread relation. In models (5) and (6), we compare firms that have at least one of these two elements to firms that lack E&S salience altogether. We continue to find that the E&S index has a stronger effect on spreads during the crisis for firms with more salient E&S activities. These findings suggest that bond spreads are not only influenced by the underlying E&S performance but also by the salience of those activities.

Fifth, since Lins et al. (2017) find that high-E&S firms earned excess stock returns during the crisis compared to low-E&S firms, we seek to determine whether the bond spread effect that we document is incremental to the stock return effect or whether the bond market performance is merely a reflection of superior stock market performance. To discriminate between these two possibilities, we conduct two tests. In the first test, we control for the firm’s contemporaneous stock returns in the spreads regression of model (2). Moreover, we allow the effect of returns to vary during the crisis- and post-crisis periods. Specifically, we estimate the following augmented regression model:

where R_it is firm i’s raw stock return during month t and all other explanatory variables follow earlier definitions. The findings from estimating this regression are reported in Table 6. In model (1), the effect of contemporaneous stock returns is held fixed throughout the period, while in model (2) we allow the stock return effect to vary across subperiods. Both models illustrate that the effect of E&S activities on bond spreads during the crisis is incremental to the stock price effect and therefore cannot be inferred from the prior evidence on stock returns. Moreover, the coefficient on the E&S index is similar to that in the models that do not control for stock returns.

In the second test, we investigate whether the cross-sectional patterns documented for bond spreads in Panel A of Table 5 can be inferred from the stock market findings of Lins et al. (2017). To do so, we replicate their stock market results for our sample, splitting the data according to probability of default, tangibility, and the ability to make payouts when distressed (our agency cost of debt proxies). If the reduction in credit spreads for high-E&S firms during the crisis is driven by the same factors that explain stock returns, then we should observe similar cross-sectional patterns in stock returns (i.e., high-E&S firms should have higher stock returns during the crisis when they have a higher probability of default, lower tangibility, and are more able to make payouts when in distress). In untabulated tests, we do not find this to be the case: there is no consistent pattern in stock returns across these three proxies.

Sixth, instead of controlling for stock returns in the bond regressions, we control for operating performance using the four measures employed by Lins et al. (2017): operating profitability, gross margin, sales growth, and sales per employee. While we find that credit spreads are lower for firms with better operating profitability, the effect of the E&S index on bond spreads during the crisis persists with similar economic and statistical significance (not reported in a table).

Overall, the findings from these additional tests indicate that the effect of E&S performance on bond spreads during the crisis reflects bondholders’ expectations of the likelihood of asset substitution or diversion taking place. As such, the key bond market benefit of E&S investments during the financial crisis is a reduction in the perceived agency costs of debt. This effect appears to be particularly pronounced in firms whose E&S activities are more salient.

To investigate bond originations on the primary market during the financial crisis, we use sample selection criteria similar to those described in Sect. 2 for secondary market bond trades. From Mergent FISD we obtain origination data for bonds that were issued between January 2007 and September 2019 by U.S. domiciled and incorporated publicly listed firms, excluding bonds with non-standard features. This procedure yields 5,797 new issues by 868 firms. After merging these data with annual fundamentals and market data from Compustat and CRSP, respectively, our resulting bond-origination sample for issuers with an E&S rating in 2006 contains 3,606 corporate bonds issued by 406 firms.

We present summary statistics for the variables used in our bond origination estimations in Table 8. The firm characteristics are for the fiscal year-end prior to the year of the bond issue. The average issue size is 3.4% of assets with a median of 2.1%. The bond proceeds average $834 million for a maturity of 8.3 years. The mean credit spread for new bond issues is 1.61%.

We first examine whether high-E&S firms were able to raise more debt in the primary market during the crisis by estimating the following regression for all issuing firms:

where Issue_ijt is defined as the offering amount scaled by total assets (firms that do not make an initial or seasoned public bond offering are excluded from this analysis), and Z_it-1 is a (L×1) vector of lagged firm-level controls that are typically used in studies on new bond issuance (e.g., Leary and Roberts 2005; Badoer and James 2016). Specifically, we control for Market equity, Book-to-market, Profitability, Debt-to-capitalization, Cash holding, Tangibility, Capital expenditure, Dividend indicator, and Investment-grade indicator. We also include firm and time (quarter) fixed effects. As with earlier estimations, we update the firm-level variables three months after a firm’s fiscal year-end, and we double cluster standard errors at the firm and time level. As in Eq. (2), the main effect for the E&S index, measured at the end of 2006, is not included because it is subsumed by the firm fixed effect.

Table 9 contains the estimation results. We report models for the full sample as well as for nonfinancial and financial firms separately. Models (1), (3), and (5) include only the pre-crisis E&S index interacted with the crisis and post-crisis dummies and the fixed effects as explanatory variables, while models (2), (4), and (6) include all control variables. For the full sample, estimation results from models (1) and (2) indicate that high-E&S firms that accessed the corporate bond market during the financial crisis were able to raise more funds than low-E&S firms. In terms of economic significance, based on model (2), increasing the E&S index by one standard deviation (22.1) increases the amount issued as a percentage of assets by 2.9 percentage points during the crisis. The crisis effect is substantial when compared to the average issuance of 3.4% of assets over the entire sample period and 3.1% (untabulated) of assets during the crisis months. Thus, high-E&S firms were essentially able to almost double the size of their bond issues during the crisis relative to the average firm. In the post-crisis period, on the other hand, E&S efforts have no significant impact on the amount of debt raised in the primary bond market.

When studying nonfinancial and financial firms separately in models (3)–(6), we find that the issuance results are solely driven by nonfinancial firms. This is not surprising, given that bond markets were essentially shut for most financial firms during the crisis; only six banks issued bonds during this period.

Next, we examine the effect of E&S performance on the pricing and maturity of new bond issues, both during and after the financial crisis. To the best of our knowledge, this is the first investigation of the effect of social capital on bond contracting terms. We adopt a similar approach as in the prior tests on the amount raised, and estimate the following specification:

where Issue term_ijt is the dependent variable of interest. We study both at-issue credit spreads and the maturity of the bond. The vectors of bond and firm controls, X_ijt-1 and Z_it-1, are the same as in Eq. (2), with the exception of duration, which is not available until several months after the issue date.³¹ As in all models, we also control for firm fixed effects, FFE_i, to capture unobservable time-invariant firm-specific determinants of credit risk, and time fixed effects, TFE_t. Standard errors are again double clustered at the firm and time level, and the firm’s 2006 E&S index itself is captured by the firm fixed effect.

In Panel A of Table 10, we report the results from estimating Eq. (5) for at-issue credit spreads for our sample of bonds issued from January 2007 to September 2019. We report results for the full sample and for nonfinancial and financial firms separately. Models (1), (3), and (5) include all controls except for the governance pillar score, while models (2), (4), and (6) include governance as a control variable. The effect of the E&S index on at-issue spreads is negative and significant only during the crisis, for both nonfinancial and financial firms, when all controls are included in the model. During this period, the effect is also economically important. For instance, based on the coefficient estimates presented in model (2), a one standard deviation increase in the pre-crisis E&S index (22.1) is associated with 27 basis points lower spread on bonds issued during the crisis period.³² The effect of the E&S index on spreads during the post-crisis period is not statistically significant, consistent with our findings for the secondary market. Finally, the difference between the coefficients for the crisis and post-crisis periods is statistically significant  for the full sample.

Next, we assess the relation between E&S performance and bond maturity. Imposing a shorter maturity can be viewed as an extreme type of covenant, given bondholders’ limited flexibility in recontracting due to unanimous consent requirements (e.g., Rey and Stiglitz 1993; Berger and Udell 1998). If E&S activities engender trust, high-E&S firms may be able to issue bonds with relatively longer maturities when prevailing trust levels have been eroded. To assess the impact of E&S performance on bond maturity, we regress time-to-maturity on bond- and firm-level controls as in Eq. (5). The results from this estimation are reported in Panel B of Table 10 and show a significant positive relation between the E&S index and bond maturity during the crisis for the full sample and for the subsample of nonfinancial firms. According to model (2), a one standard deviation increase in the pre-crisis level of the E&S index translates into a 17-month-longer time-to-maturity during the crisis (equivalent to over 17 percent of the mean level of maturity in the sample), compared to the pre-crisis period.³³

In sum, our primary bond market tests provide further evidence that bondholders value the trust earned from building social capital via E&S investments: during the financial crisis, high-E&S firms were able to raise more debt, at more favorable interest rates, and for a longer period of time.³⁴