ESG did not immunize stocks during the COVID-19 crisis, but investments in intangible assets did

3.1 Data and sample

We obtain accounting information from the Compustat quarterly and annual databases, and stock market returns data from CRSP. Internal control weaknesses are taken from the Audit Analytics dataset, while institutional ownership data is derived from the Thomson Reuters 13f database. I/B/E/S data is the source of our analyst following measure, and the CEO tenure variable comes from the BoardEx database. Fama–French factors are obtained from Kenneth French's website.13 Measures of investor orientation are calculated using data from the Thomson Reuters 13f database merged with investor classification data obtained from Brian Bushee's website.14 In addition, we manually collected some executive start dates that were identified as missing in order to maximize the available sample.

Refinitiv's pre-April 2020 EIKON database is the primary source of our environmental, social and governance (ESG) scores for the COVID-19 analyses, reflecting the data that would have been available to the market at the onset of the pandemic. In light of current controversies regarding inconsistent rankings across commercial ESG data providers (Berg et al., 2020), we also replicate our findings using MSCI data for 2018 (i.e., the most recent data that would have been available in early 2020 at the onset of the pandemic). Finally, in light of the revelation that Refinitiv restated its data in April 2020 in a manner that made it historically more highly correlated with returns (Berg et al., 2021), we also verify that our key results hold using this revised data, which we refer to as “Refinitiv II.”15

In order to construct our primary dataset, we begin with all firm-year observations for which a Refinitiv ESG score is available for fiscal 2018, the last annual reporting period included in the database for most firms prior to the onset of the COVID crisis. We restrict our sample to US companies, and we drop all financial and real estate firms from our tests. We also drop all observations that have an undue influence on the determination of the coefficients in the fully specified returns regression (i.e., Model 1 introduced below).16 This results in a sample of 1652 firms for which all requisite data is available for our tests of the Q1 2020 COVID crisis period. Due to the removal of additional influential observations, the sample declines to 1642 firms for the full year returns analyses. Further details on the sample determination process are provided in Table 1.

3.2 Descriptive statistics

Descriptive statistics for our sample firms are provided in Table 2A for Q1 and Table 2B for the full year sample. In light of the highly overlapping nature of the samples, only the returns variables differ materially across the two panels. Given the relatively large firm bias of some of the databases being used for our analyses, sample firms tend to be somewhat larger, on average, than the CRSP-Compustat universe. The average (median) firm has been established for 28 (24) years and has nearly 10 (8) analysts providing earnings estimates for it. Twenty-one percent of sample firms reported negative earnings for fiscal 2019, while the average (median) firm's adjusted ROA for that year was approximately 3.3% (4.4%). The average abnormal buy-and-hold (raw) returns during the Q1 crisis period were −5.6% (−31.4%), which improved to −4.6% (20.6%) for the full year of 2020. The mean (median) overall Refinitiv ESG summary score for sample firms is approximately 47 (43) out of a theoretical maximum of 100, the interquartile range is about 25, and the full range of this variable for our sample firms is 7 to 95 (untabulated), all suggesting that there is a good deal of cross-sectional variation in the test variable of interest.

Panels A and B of Table 3 present the pairwise Pearson correlations between select regression variables for each of Q1 and the full year, respectively. As shown in Panel A, Cash and RD&SGA stock are the two fundamental variables that are most significantly positively correlated with firms’ buy-and-hold abnormal returns (BHAR) for Q1 2020, while short- and long-term debt are those most significantly negatively correlated with Q1 returns. These simple pairwise correlations confirm our expectations that firms with a lot of liquidity and innovation-driven assets are more resilient during the market implosion, whereas those with more significant debt burdens are less so. For the full COVID year, Panel B reveals that RD&SGA stock is again the variable that is most highly correlated with BHAR, while Cash and LTDebt continue to exhibit significant, albeit diminished, positive and negative correlations with BHAR, respectively. ROA and Size are surprisingly significantly negatively correlated with returns for the full year period. In unreported tabulations (for parsimony) we find that traditional market-based measures of risk are also highly significantly associated with abnormal returns.

Consistent with widespread reports in early 2020 that ESG was a resilience factor during the Q1 market implosion, the correlations reported in Panel A suggest that ESG scores and abnormal returns are positively correlated, albeit not nearly as strongly as the previously mentioned cash, internal investments in intangible assets, and some of the market-related risk factors, nor as statistically significantly as the negative correlations found for the debt variables. Contrary to similar claims of ESG outperformance for the full year of 2020, however, the correlation between ESG and each of BHAR and raw returns is negative over this longer period. In other words, even the pairwise correlations do not support the claims that ESG was a resilience factor for the full COVID year of 2020. It remains to be seen whether the association between ESG and returns will be significant for either period once the competing fundamental accounting and market risk factors are simultaneously incorporated into a multiple regression analysis. We turn to this next.

4.1 What explains returns during the COVID-19 Q1 2020 crisis period?

For our primary tests, ESG is measured using Refinitiv's pre-April 2020 overall summary score of ESG performance. Although some prior studies focus only on the ES portion of Refinitiv's ESG score under the assumption that these are the components that best capture the social capital that is purported to lead to resilience, we consider Refinitiv's comprehensive ESG score to be the most appropriate measure for our tests. Our reasoning relies upon a review of the subcomponents included in Refinitiv's “G”, or the governance pillar, of the total ESG score.

These include variables designed to capture board and executive gender and cultural diversity (i.e., arguably as much “S” as “G”), an indicator for whether the firm is reporting under GRI standards and the percentage of the firm's activities covered in its E&S reporting (i.e., impacting both “E” and “S” measurement), as well as numerous other variables that border upon, or are intimately intertwined with, the E and S dimensions of corporate sustainability.18

We control for the firm's liquidity and leverage with Cash, LTDebt and STDebt and for accounting-based performance with ROA and the Loss indicator variable. Given widespread concerns regarding supply chain issues during the early months of the pandemic, we include the industry-adjusted inventory turnover ratio (InvTurn), measured following the approach suggested by Platt and Platt (1991), to capture this dimension of the firm's performance.

We also include variables to capture the firm's past investments in acquired and internally developed intangible assets, AcqIntang and RD&SGAstock, respectively.

RD&SGAstock is a measure designed to capture the notion that, although required to be expensed under US GAAP, firm expenditures on R&D, as well as some portion of expenditures on SG&A, are actually investments in internally developed intangible assets. More specifically, the calculation of RD&SGAstock assumes that all R&D expenses and an assumed rate of ⅓ of SG&A expenses represent investments in intangible assets that will have a five-year life. We use the assumed rate of ⅓ of SG&A as a conservative estimate of the “investment” portion of SG&A expenditures although this rate has been growing over time (Enache & Srivastava, 2018). We notionally capitalize these expenditures and amortize them linearly over a five-year period, which is also a conservative estimate of the duration of these benefits (Lev & Sougiannis, 1996). So, for example, RD&SGAstock for fiscal 2019 = FY2019 (R&D + ⅓SGA)*100% + FY2018 (R&D + ⅓SGA)*80% + FY2017 (R&D + ⅓SGA)*60% + FY2016 (R&D + ⅓SGA)*40% + FY2015 (R&D + ⅓SGA)*20%.

Additional control variables include the firm's dividend payout ratio (DivPayout), Size, market share (MktShare) and the number of years since the firm first appeared in Compustat (Age), allowing the latter to enter non-linearly with the inclusion of Age². We control for the firm's reporting timeliness and quality by including the average earnings announcement speed over the prior four quarters (MeanAnnSpeed) (Gallemore & Labro, 2015) as well as a count of internal control weaknesses noted in the most recent fiscal year available prior to the crisis (ICweakness). We capture investor horizon (InvestorOrient) using the measure proposed by Serafeim (2015), while InstOwners is the average percentage of institutional investors in the firm's stock, truncated at 100% consistent with Gompers and Metrick (2001). Analyst is the firm's analyst following, a measure designed to capture the firm's information environment, and lnCEOtenure is the natural log of the length of the CEO's tenure with the firm measured in days. Market-based measures of risk and/or growth opportunities include the book-to-market ratio (BTM), an indicator set to one if BTM is negative (BTMneg), prior stock price momentum (Momentum), idiosyncratic risk (IdioRisk), as well as the four Fama–French factor loadings (MKTRF, SMB, HML, MOM). All specifications also include industry fixed effects, where industries are defined using two-digit SIC codes. All variables other than InstOwners and ESG are winsorized at the top and bottom 1%, and each is defined in greater detail in the Appendix.

The results from estimating variants of Model 1 using the abnormal buy-and-hold returns from the first quarter of 2020 as the dependent variable are presented in Table 4. The specification in the first column includes only ESG and the industry indicators, while the regression in the second column also includes market-based measures of risk. Consistent with the early pandemic hype, ESG is significantly positively related to returns in each of these specifications, prior to other controls being included in the regression. As shown in column (3), when Size is added to the regression, this control variable is positive and significant, whereas the coefficient on ESG declines by more than half, and ESG is no longer significant at any conventionally acceptable level.19 These results suggest that the exclusion of a simple control and known determinant of returns (i.e., firm size) leads to incorrect inferences regarding the role of ESG as a determinant of returns.20

Column (4) presents the results from a more completely specified regression that includes fundamental accounting-based measures of performance, supply chain management, liquidity, leverage, investments in intangibles and financial reporting quality, while column (5) adds further control variables such as analyst following, firm age, CEO tenure, and share ownership characteristics. As shown, ESG remains insignificant in each of these specifications, whereas the coefficient signs, magnitudes, and significance levels of most of the other control variables remain largely consistent across models.

As shown in column (5), firms with higher levels of institutional ownership performed less well during the market downturn (i.e., InstOwners is negative and significant), a finding that is broadly consistent with the results reported in Glossner, Matos, Ramelli, and Wagner (2020). These authors show that institutional investor ownership falls more in firms with weak financials (i.e., lower cash holdings and higher leverage), which they interpret as institutional investors prioritizing financial strength over ESG performance in the face of the COVID-19 Q1 2020 tail risk event.

Although it is not surprising that idiosyncratic risk and the four Fama–French factors all remain statistically significant in explaining returns in this more fully specified regression, it is nevertheless notable that they do not dominate the more fundamental accounting-based measures of the firm's expected resilience or reporting quality. Specifically, the firm's inventory turnover rate (InvTurn), its level of cash stores (Cash) and its reporting timeliness (MeanAnnSpeed) are positive and significant determinants of returns during the crisis period, while the firm's short- term and long-term debt levels are negatively associated with returns. These results suggest that traditional accounting-based measures of the firm's reporting timeliness, supply chain management, liquidity and leverage are all important indicators of a company's share price resilience during the early days of the unexpected global COVID crisis although ESG is not.

RD&SGAstock, the measure of the firm's stock of innovative assets (i.e., the unamortized portion of the capitalized internally-developed R&D- and SG&A-related intangible assets) is also statistically and economically significant. The results from column (5) suggest that a one standard deviation increase in the stock of internally developed intangible assets is associated with an approximately 1.7% increase in abnormal returns during the Q1 2020 crisis period, or about 7% on an annualized basis.21 Overall, the results from the fully specified regressions in Table 4 clearly indicate that investments in intangible assets led to share price resilience during the Q1 2020 market crisis period of the COVID pandemic, whereas ESG performance did not.22

In untabulated specification checks, we confirm that the insignificance of ESG is not due to the inclusion of RD&SGAstock in the regression nor due to the associated remeasurement of net income and total assets resulting from the notional capitalization of R&D and (a portion of) SG&A expenses. In other words, our results are robust to using unadjusted net income in the calculation of ROA, using unadjusted total assets as the scalar, and to dropping RD&SGAstock from the regression. Thus, our combined findings of significance for RD&SGAstock and insignificance for ESG should not be interpreted as an indication that RD&SGAstock is serving as an alternative proxy for firms’ investments in CSR that somehow manages to dominate the ESG performance measures derived from commercial databases. Indeed, the negative pairwise correlations (of modest magnitudes) between ESG and RD&SGAstock presented in Table 3 further suggest that, statistically speaking, these two variables are not substitute proxies for the same underlying construct. Furthermore, practically speaking, it would be unlikely that RD&SGAstock would dominate ESG in capturing CSR performance given that the ESG scores are actually designed to capture performance (i.e., the outcomes of efforts and expenditures on CSR-related initiatives), whereas the RD&SGAstock measure derives from expenditures (i.e., inputs, only some of which are likely to be related to CSR initiatives, and not all of the latter are likely to have a positive pay off in terms of manifesting in higher CSR performance). In other words, RD&SGAstock does not emerge as the winner of a “horse race” between intangible asset investments and ESG because ESG becomes insignificant even in the absence of RD&SGAstock being in the regression. Instead, these two variables are capturing quite different elements of the firm's activities, of which only the investments in intangibles are a resilience factor during COVID-19.

In order to gain a better understanding of the relative importance of accounting, market, industry membership and other variables in explaining crisis period returns, we undertake an Owen-Shapley decomposition as explained by Huettner and Sunder (2012).23 Using this approach, we are able to estimate the proportion of the explanatory power for returns that each set of variables contributes. Table 4 reports that our most complete regression from column (5) explains approximately 42% of the overall cross-sectional variation in the Q1 COVID crisis period returns for the firms in our sample. Figure 1 presents a pie chart depicting the proportion of this 42% that is explained by each group of variables. As shown, the set of market-based measures contributes the most to the overall R², with 40% of the explained variation being due to these variables. Industry membership is a relatively close second, accounting for 34% of the explained variation. Measures derived from the firms’ financial statements combined with measures of the firms’ financial reporting quality together account for 20% of the explained variation in stock returns, of which 6% is due to RD&SGAstock, while other variables (e.g., ownership characteristics, market share and firm age) contribute just 4% of the overall explanatory power. Most notable, however, is the finding that the ESG summary score is the least important category, contributing less than 1% of the total 42% explained variation in returns during the first quarter of the COVID crisis.

Taken together, our results from these regression analyses and the Owen-Shapley decomposition suggest that classic market-based determinants of returns, industry fixed effects, as well as financial statement variables that capture the firm's financial and supply chain performance, liquidity, leverage and investments in intangible assets are all important in explaining Q1 COVID crisis period stock returns, whereas ESG is not. We conclude that ESG was not a share price vaccine during the market's implosion in the early months of the global pandemic, but that investments in internally generated intangible assets were positively and economically significantly associated with share price resilience.

4.2 Determinants of returns for the full COVID year of 2020

Table 5 presents the results of the regressions from variants of Model 1 using the abnormal buy-and-hold returns from the full year of 2020 as the dependent variable. The overall explanatory power of the most complete model for this full year period is just 16.5%, which is considerably below the 42% of explained variation previously documented for the Q1 crisis period. In contrast to the findings for Q1, the full year return results presented in column (1) of Table 5 indicates that ESG is a weakly negative and significant determinant of returns in the absence of any additional market risk and accounting-based control variables being included in the regression. Once additional standard controls are added to the regressions in columns (2)–(4), ESG becomes insignificant as an explanatory variable for the full COVID year 2020 returns.

Consistent with the much lower level of overall explanatory power, several variables that were important in explaining Q1 returns are no longer significant when considering returns for the entire year. For example, idiosyncratic risk, three out of the four Fama–French factors and inventory turnover are all insignificant in the full year returns regressions. Offsetting the loss in explanatory power of these variables, dividend payout and firm age both become negatively associated with returns in the full year model, whereas the loss indicator becomes positively significant.

Interestingly, long-term debt is largely insignificant in explaining full year returns (versus negatively significant in Q1), while cash and short-term debt are insignificant (versus positively and negatively significant for Q1, respectively, which was in line with expectations). These full- year results for the liquidity and debt variables, which are both counter-intuitive and inconsistent with the findings in prior GFC period studies and with those reported in the previous section for the Q1 COVID crisis period, may perhaps be explained by government interventions that began in the second quarter of 2020 and that continued in various forms through to the end of the year (i.e., traditional measures of financial structure vulnerability are no longer relevant when the government effectively steps in as a sponsor or guarantor). As leverage and liquidity are primarily control variables in our study, we leave it to future research to explain why their signs and significance differ from expectations over the longer period of the COVID pandemic.

The results in columns (3) and (4) indicate that investments in internally generated intangibles, captured by RD&SGAstock, remain significantly positive determinants of COVID full-year returns. Using the coefficient estimate from column (4), we find that a one-standard deviation increase in RD&SGAstock results in an economically significant 9.3% increase in full year abnormal returns (0.769731 * 0.12033202 ≈ 9.3%, again calculated using the industry- orthogonalized standard deviation of RD&SGAstock for the full year sample) even after controlling for industry fixed effects.

In order to understand the importance of each set of variables in explaining market returns for the full year period, we once again undertake an Owen-Shapley decomposition of the 16.5% explained variation in share price changes. Figure 2 presents the results of this analysis, with the findings indicating that there are some notable changes in the roles of different sets of variables in explaining the Q1 versus full year returns. Specifically, market-based risk proxies capture just 19% of the explained variation in annual returns, versus nearly 40% in Q1. Industry indicators are the most important set of variables in explaining full year returns, capturing 38% of the explained variation (versus an almost equivalent 34% in Q1). The explanatory power of the accounting variables, led by the importance of RD&SGAstock, increases dramatically in the full year returns to capture 33% of the explained variation (versus 20% in Q1). Of this 33%, more than half (i.e., 18%) is due to the internally generated intangible asset variable. The collection of other non-accounting/non-market variables are also more important in explaining annual versus Q1 returns, capturing 10% versus 4% of explained variation in each respective period. Finally, ESG remains a relatively useless variable, capturing less than 0.5% of the explained variation in returns for the full year returns period, similar to its irrelevance in the Q1 returns regressions.

Overall, the combined results from our analyses present a clear and consistent picture related to ESG as a potential resilience factor during the global pandemic. Specifically, in contrast to previous studies’ findings of a positive association between CSR and returns during the GFC period, and to the conventional wisdom during the recent crisis, ESG is not significantly associated with returns during the COVID pandemic. This finding is consistent across each of the Q1 market implosion and full COVID year of 2020 periods. By contrast, an accounting-based measure of firms’ investments in internally developed intangible assets is positively associated with returns for both periods, and its association is both statistically and highly economically significant. Our analyses do not address whether ESG is insignificant due to Refinitiv's score being a noisy measure of CSR performance versus whether the insignificance is due to a lack of economic relation between returns and the true underlying construct of interest. The specification checks in the next section explain this issue.

4.3 Specification checks

In this section, we report the results from a series of s²⁹pecification checks using alternative measures for each of the returns dependent variable, as well as our ESG test variable of interest. For each specification check, we use the fully specified regression depicted in Model 1 with all variables included, but for the sake of parsimony we present only the coefficients on the ESG and RD&SGAstock variables of the interest.24

In the first two columns of Table 6, we report the results for each of the fully specified Q1 and full year regressions using raw returns as the dependent variable. In columns (3) and (4), we use the natural log of (1+BHAR) as the dependent variable.25 As shown, across all columns ESG remains insignificant, while RD&SGAstock remains economically and statistically significantly positively associated with returns. The results in Table 6 thus confirm that our key inferences are robust to alternative measures of returns.

In Table 7, we report the results of specification checks that focus more on the ES or corporate social responsibility (CSR), dimensions of ESG, leaving out the “G”, or governance element. In the first two columns, we report the results of fully-specified Q1 and full year BHAR regressions using MSCI data to capture CSR. Specifically, we follow Lins et al. (2017) to construct an ESG_MSCI variable by first dividing the number of strengths (concerns) for each of five categories of performance (community, human rights, diversity, employee relations and environment) by the maximum possible number of strengths (concerns) in that category. We then construct a net ESG measure for each category by subtracting the scaled number of strengths from the scaled number of concerns, and finally we sum the net measures across categories to obtain the ESG_MSCI summary score. As shown in the table, ESG_MSCI is insignificant in explaining COVID Q1 crisis and full year 2020 returns, while RD&SGAstock remains economically and statistically significant.26

In the last two columns of Table 7, we use just the ES component of Refinitiv's ESG as the test variable of interest (ES).27 As shown, ES is insignificant in explaining returns for both periods, while the prior inferences related to RD&SGAstock are unaffected. We conclude from the results in Table 7 that our inferences regarding the insignificance of ESG are robust to measures of just the environmental and social components (ES) of CSR.

In Panel A of Table 8, we report the results of fully-specified Q1 and full year BHAR regressions using what we refer to as the Refinitiv II dataset to capture ESG. In April 2020, Refinitiv restated previously reported ESG measures, and Berg et al. (2021) have noted that these changes are systematic and related to past performance, suggesting that depending upon whether the original Refinitiv versus Refinitiv II rewritten data are used, the observed relation between ESG and returns may be different. As shown in Panel A of Table 8, however, even using the restated Refinitiv II data, ESG_ReftvII is not significantly associated with returns in either Q1 or for the full year of 2020.

In our last specification check, we consider the potential impact that “greenwashing” may have on our reported results. It is conceivable that firms that are greenwashing may benefit from enhanced ESG scores (i.e., that the ratings purveyor does not adequately “see through” this window dressing). In other words, it is plausible that ESG scores are “noisy”, especially through the middle of the distribution, in discriminating between firms with moderately low, moderate, or moderately high actual (i.e., as opposed to measured) environmentally and socially responsible behavior, whereas the theory suggests that it is the actual behavior that should lead to resilience in times of crisis. Under the assumption that firms that are mainly greenwashing would not attain a very top ESG score, we create a relatively greenwashing-free measure of ESG performance by setting an indicator variable, ESG_top, equal to one for firms in the top decile of our sample on the basis of their original Refinitiv ESG scores and setting it to zero otherwise.28 Panel B of Table 8 reports Q1 and full year regressions using this indicator variable to capture ESG performance. As shown, for both periods ESG_top is insignificant, while our inferences related to RD&SGAstock remain unchanged. In other words, these results indicate that the stock returns of even the highest performing ESG firms, which are those that are most likely to be truly performing well on ESG rather than merely gaming the ratings, were not higher during the Q1 2020 market implosion, nor for the full COVID year, after controlling for other standard determinants of returns.

In summary, the specification checks in this section confirm that our results are robust to alternative measures of stock returns and to numerous alternative measures of ES and ESG performance. Specifically, our results consistently indicate that ESG is not a significant explanatory variable for returns for the Q1 crisis period, nor for the returns of the full COVID year of 2020. Furthermore, RD&SGAstock remains a positively and statistically significant variable in explaining returns across all specifications. The combined results strongly support the conclusion that ESG is not a share price vaccine during the COVID pandemic, but that investments in internally generated intangible assets are an economically significant resilience factor during the current global crisis. While we cannot definitively rule out that the numerous ESG measures derived from three different databases all provide such poor proxies for CSR performance that this noisy measurement is causing the insignificance across all specifications, we consider it to be more likely that the global insignificance is due to a lack of robust economic relation between generic, summary level ESG (or ES) performance and stock returns during the COVID crisis.