Corporate social responsibility, performance and risk taking in Europe. An inter-industry comparison between banks and non-banks.

Corporate social responsibility, performance and risk

taking in Europe. An inter-industry comparison between

banks and non-banks.

P.S. Veldmana

Combined thesis MSc. Finance & MSc. Economics University of Groningen

Faculty of Economics & Business Supervisor: dr. A. Dalò

Abstract:

This paper studies the relationship between corporate social responsibility (CSR), financial performance and risk-taking with a focus on banks. The constituents of the STOXX 600 are investigated during the period 2003-2018. By implementing an inter-industry comparison between banks and non-banks this paper shows that the sensitivity of performance and risk-taking to CSR is significantly different for banks compared to non-banks. For non-banks a positive relationship between CSR and financial performance is found. Moreover, CSR reduces risk-taking significantly for this group of firms. On the other hand, for banks a neutral relationship is found between CSR, financial performance and risk-taking. Additionally, a Granger causality test provides evidence that causality mainly runs from CSR to financial performance.

JEL classification: M14, G21, L2 and G32

Keywords: corporate social responsibility, banks, firm performance and risk

2

1. Introduction

In the aftermath of the financial crisis the role of financial institutions as the cause of the crisis has attracted a significant amount of researchers’ attention. A substantial part of this focusses on subjects such as the banking competition-stability debate, governance structures or excessive executive compensation. These are all important subjects itself, but many are affected by one important business philosophy; a corporate social responsibility (CSR) philosophy. While CSR is a well-known concept, there is no consensus in the literature on the proper way to define CSR. Dahlsrud (2006) provides an overview of the different definitions used. The most used definition provided by the commission of the European communities (2001) states: “CSR is a concept whereby companies integrate social and environmental concerns in their business operations and in their interaction with their stakeholders on a voluntary basis”. Broadly speaking, CSR integration means going beyond a profit only focus and incorporation of a broader set of stakeholder interests into decision making. Hence, a stronger CSR engagement might have prevented the individualistically oriented choices, which resulted in excessive risk taking, empire building or unrealistic executive compensations.However, the complete role of CSR in conjunction with the crisis has been overlooked or disregarded in the literature for a long period of time. A reason for inadequate CSR adoption could be reluctance of firms to allocate sufficient resources to this because it might negatively affect financial performance. Therefore, it is relevant to empirically investigate if this is a sound reason. There is already a strong grounding of literature discussing the relationship between CSR and firm performance in general. Surprisingly, given the role of banks in the financial crisis, research has mainly been conducted for firms in general and research focusing on the financial industry is still rather scarce. Additionally, due to the fact that many studies have been executed within the United States (US), Nollet et al. (2016) suggest that future research should focus on effects of CSR in other regions (with a different legal framework) or in different sectors. Accordingly, this study addresses this shortfall by executing an inter-industry comparison between European banks and European firms not operating in the banking industry (non-banks), resulting in the main research question:

3

By studying the time period 2003-2018 this paper shows that non-banks with a stronger CSR commitment have a better financial performance when looking at general performance measures such as return on assets (ROA) and Tobin’s Q. While causality remains an issue in the social-financial performance relationship, this paper provides evidence that the relationship is mainly running from CSR to performance. This indicates that the positive relationship found is likely to be a causal relationship as well. However, for banks the positive relationship is non-existent and a neutral relationship is found. Both banks and non-banks should thus not shy away from CSR because it can be detrimental for performance. Moreover, the shape of the relationship has been assessed. Allowing for a non-linear relationship does not provide evidence for the U-shaped relationship previously found in the literature (Nollet et al, 2016 and Barnett and Salomon, 2012). Additionally, this study shows that a strong CSR focus can mitigate firm risk-taking, but when analyzing banks there is again a neutral relationship found. The conclusion can be drawn that doing business in a responsible way can have advantages in terms of performance and firm’s risk-taking behavior. Nevertheless, it is crucial to differentiate between banks and non-banks as this relationship between CSR, performance and risk-taking is not found in the banking industry.

In the next section the most relevant literature connecting CSR, performance and risk-taking is discussed. The first subsection of this literature review focusses on the general CSR-performance relationship, while the second subsection reviews CSR and banking CSR-performance explicitly. The third subsection discusses relevant literature concerning bank risk-taking. Subsequently, the methodology and further analysis is outlined. Afterwards, sample selection criteria and descriptive statistics are provided and discussed. Then, results are presented and reflected. Lastly, I sum up the main conclusions and point towards relevant future research.

2. Literature review

2.1 CSR and financial performance in general

4

CFP in his 1984’s work. Firms engaging in socially responsible activities improve their relationship with stakeholders and this results in lower transaction costs. Furthermore, socially responsible activities could increase market opportunities and enhance performance in this way. Empirically, numerous studies investigate the relationship between CSR and financial performance, but the overall picture is not clear-cut as well. There are positive, neutral and negative relationships found. Friede et al. (2015) give an overview of 2000 empirical studies about the CSR-CFP relationship. In the remainder of this subsection I focus on the main papers and the developments in this research area.

The majority of studies find a positive correlation between CSR and performance. For example, an early study of Waddock and Graves (1997) finds a positive relationship between CSR and firm performance and observes large differences between industries. However, this was only the start of the CSR-CFP debate. The measures chosen for performance and CSR namely might predetermine the ultimate outcome according to Griffin and Mahon (1997). An influential study of Orlitzky et al. (2003) provides more convincing evidence by executing a more rigorous study using a large dataset of 33878 observations. Their finding is in line with Waddock and Graves (1997) and, moreover, they find that the magnitude of the results depends on the measures of performance used. CSR has a stronger correlation with accounting-based measures of performance than with market-based measures of performance. A couple of more recent studies such as Saedi et al. (2015), Mishra and Suar (2010) and Hull and Rothenberg (2008) again find a positive relationship.

Next to this, there are also several early studies finding no general relationship between CSR and financial performance. For instance, Aupperle et al. (1985) and McWilliams and Siegel (2000, 2001). This indicates CSR at least does not harm financial performance by drawing away scarce resources from profitable investments. However, it also does not create new business opportunities. Additionally, McWilliams and Siegel (2000) show that misspecifying the model by excluding R&D investments as a control variable causes an upward bias of the effect of CSR. Another study failing to find a significant relationship is Nelling and Webb (2009). They argue that unobservable firm characteristics are mainly driving the relationship between CSR and financial performance.

5

financial performance. Next to this, Nollet et al. (2016) find that CSR and return on capital are negatively related using their linear model.

Not only the sign of the relationship is of question, but causality too. In the literature there is no consensus yet in which direction the relationship goes. Preston and O’Bannon (1997) already stress that not only the sign of the relationship should be tested, but causality should be addresses as well. This leads to six hypotheses in their paper. Firstly, a positive relationship going from CSR to financial performance. This relationship is the view advocated by Freeman. Secondly, a negative relationship going from CSR to financial performance. This is the opinion of Friedman discussed previously. Thirdly, the relationship can be positive going from financial performance to CSR. This is called the available funds theory. CSR draws away scarce resources of the firm. Therefore, mainly firms with more available resources will invest in CSR. Fourthly, the relationship is negative going from financial performance to CSR. This can be explained by managerial greed and self-enrichment after good performance, while trying to cover poor performance by increased spending on CSR. Furthermore, Preston and O’Bannon (1997) test a fifth and sixth hypothesis where the relationship is bidirectional and the sign of the relationship can be positive or negative. However, they do not clarify why they are testing for a bidirectional relationship.

In order to better understand the motives of firms to engage in CSR Baron (2001) provides three reasons: strategic choice, altruism and image building without really changing the underlying business. The latter is in the literature known as greenwashing. Dam et al. (2009) test these three hypotheses and find evidence for the strategic choice motive. So, firms mainly engage in CSR for profit maximizing reasons. The idea is that CSR engagement changes the firm’s profit and cost function and firms outweigh the costs and benefits of CSR.

6

2.2 CSR and financial performance for banks

According to Scholtens (2009) there is a growing role of CSR in the banking sector, but research studying the effects of CSR in the banking sector started relatively recent. A lot of early papers studying the general CSR-CFP relationship namely exclude the financial industry from their analysis. The reason for this is that the financial industry is perceived to be very different than other industries. This view resulted in two stances in the literature. Firstly, as discussed in the previous subsection, studies investigating the general CSR-CFP relationship excluding financial firms. Secondly, studies which focus on the banking industry by itself. This paper is set apart from this by including all industries and differentiating between banks and non-banks. This makes inter-industry comparisons possible and in combination with the view that banks are completely different this leads to my first hypothesis.

Hypothesis 1: the relationship between CSR and financial performance differs between banks and non-banks

The first study that draws the CSR-CFP debate to the specifics of the banking industry is

7

and Shen (2013) find a positive relationship using a worldwide sample. As a result, they argue strategic choice is the leading motive for banks to engage in CSR. Additionally, a number of studies find a positive relationship in specific countries such as Hafez (2015) in Egypt and Fayad et al. (2017) in Lebanon.

On the other hand, there are also studies finding no relationship at all. For example, Fijalkowska et al. (2018) show that investments in CSR do not result in higher profits at the bottom line for banks. They extend the preceding studies by focussing on banks in central and eastern European countries (CEEC) exclusively. Advantageous of this study is that CSR is again measured differently. Instead of using a scoring system, they classify a bank as responsible if they disclose a CSR report. Furthermore, they measure the intensity of CSR based on a content analysis of the CSR report. Another study by Islam et al. (2012), executed in Bangladesh, is unable to find any relationship. Although they did find a positive correlation between CSR and CFP, this was not statistically significant. This insignificant result can probably be traced back to the small sample size of the study.

Despite most studies finding a positive or non-negative relationship going from CSR to performance, Gonenc and Scholtens (2019) obtain a conflicting result. Using a large dataset of worldwide banks, they mainly find evidence for the available funds theory. They report that CSR is not driving performance, but performance is driving CSR. Furthermore, they find evidence that the financial crisis might have changed the CSR-CFP performance relationship. An advantage of this study is the large sample size. Earlier studies, mainly in specific countries, suffer from small sample sizes. In my paper the CSR-CFP relationship is revisited and this leads, given that the majority of papers finds a positive relationship, to the second hypothesis.

Hypothesis 2: there is a positive relationship between CSR and (bank) financial performance.

8

2.3 CSR and risk-taking

As a response to the financial crisis there are numerous studies investigating causes of bank risk taking. This is of importance because financial instability and excessive risk taking can fuel a worldwide recession. It is crucial to know how this could have happened and especially how this can be prevented in the future. In this research area, much research is focussed on the role of competition and its effect on risk-taking. For instance, Jiménez et al. (2013) find evidence that more market power reduces bank risk-taking. Another topical issue is the effect of the long period of very low interest rates on bank risk-taking. Altunbas et al. (2010) and Delis and Kouretas (2011) both find that low-interest rates trigger bank-risk taking. The role that CSR can play has been left unexamined for a long time. Boddy (2011) argues in a theoretical paper that corporate directors in the financial industry have caused the crisis by greed and self-enrichment. A stronger CSR focus and internal enforcement mechanism could lead to more ethical behaviour. However, the literature in this regard mainly focusses on corporate governance instead of the complete role of CSR. For example, Pathan (2009) argue that board structure is an important factor in determining bank risk-taking based on their finding that small bank boards, less restrictive boards and a low number of independent directors are positively associated with risk-taking. Besides, a study not specifically focussing on banks (Harjoto et al., 2018) finds that firms with more diverse boards take less risk. As in the CSR and financial performance relationship, current literature is once again separated by papers focussing on the banking industry exclusively and papers including a broad range of firms, but excluding the financial industry. This paper closes this gap in the literature by connecting these two stances and investigate if differentiating between banks and non-banks has an empirical grounding. This results in the third hypothesis.

Hypothesis 3: the relationship between CSR and risk-taking differs between banks and non-banks

There are only a small number of papers investigating the complete role of CSR on risk taking. For instance, by studying a broad US sample, Harjoto and Laksmana (2018) find that CSR acts as a control system to avoid excessive risk taking and limit excessive risk

9

CSR on performance, but combines this with the risk-taking behaviour of banks. He finds no relationship between CSR and risk-taking in general. Even though a neutral relation between CSR and risk-taking is found, subdividing the total KLD-score into a KLD-business and KLD-discretionary score shows these two effects offset each other. KLD-business (discretionary) was negatively (positively) associated with bank risk-taking. This

demonstrates that only CSR activities related to core operating activities reduce bank risk taking. Secondly, Scholtens and van ‘t Klooster (2019) investigate the relationship between CSR, bank’s default risk and their contribution to the financial sector’s overall risk. They find CSR reduces default risk and the contribution to systematic risk. Compared to Bolton (2013) this is found for total CSR and not just CSR investments related to core operating activities. Given these findings the following hypothesis is investigated.

Hypothesis 4: there is a negative relationship between CSR and (bank) risk-taking

There is no generally accepted measure of risk-taking in the literature. Therefore, Bolton (2013) looks at risk-taking from two perspectives. On the one hand, general risk-taking is measured as the potential for distress. This is done by the Z-score. On the other hand, banks needing financial assistance of the U.S. government during the crisis are perceived as riskier. Scholtens and van ‘t Klooster (2019) only use accounting-based measures to proxy risk taking. They measure risk-taking by the Z-score and as a robustness check they proxy risk-risk-taking by the volatility of return on equity (ROE).

3. Methodology

10

of CSR for banks specifically from a general firm regression. This ensures a sufficiently large sample size and prevents draining up the degrees of freedom as many studies in the banking industry suffer from a small sample size. Furthermore, a general regression gives insight into the effect of CSR on performance and risk-taking for banks compared to non-banks.

Firstly, a linear panel regression is estimated. Secondly, I estimate a panel regression including a linear and a quadratic term, which tries to capture the U-shaped relationship advocated by Barnett and Salomon (2012) and Nollit et al. (2016)1. An accounting-based measure (ROA2) as well as a market-based measure (Tobin’s Q) is used for financial performance. ROA is calculated as net income / total assets and Tobin’s Q is approximated using the method of Chung and Pruitt (1994), which is (market value of equity + book value of debt) / total assets. CSR is proxied by the environmental, social and governance (ESG) combined score. Apart from CSR as the variable of interest, a dummy variable (Bi) interacting with CSR is included. This dummy variable takes the value of 1 if the firm is a bank and 0 otherwise. In this way, I infer if and how the effect of CSR in the banking industry differs from the effect of CSR for non-banks. Control variables are size (logarithm of total assets), risk (standard deviation of monthly stock returns within the year), leverage (debt-to-total assets) and liquidity (cash-to-total assets). Theoretically, R&D should be the final control variable. Including R&D is crucial because omitting this variable might bias the effect of CSR upward (McWilliams and Siegel, 2000). However, R&D data is not widely available and results in a significant number of missing values. Therefore, R&D is proxied by intangibles (intangible assets-to-total assets). Just as R&D, intangible assets signal a firm’s innovativeness and willingness to invest now in order to achieve an uncertain return in the future. Size is included because larger firms might outperform due to economies of scale or more public attention (Cornett et al., 2016). However, size can also make a firm more bureaucratic and cause inefficiencies (Waddock and Graves, 1997). Leverage controls for different capital structures, while risk controls for differences in perceived risk by investors. Liquidity controls for a potential liquidity-profitability trade-off. Having more cash decreases liquidity risk, but also results in having fewer resources available to fuel profitability. This relationship is especially relevant during a crisis (Adjei, 2013).

1 _{Modelling the relationship as in Barnett and Salomon (2012) and Nollit et al. (2016) may cause a}

multicollinearity problem because CSR and CSR2 _{are likely to be highly correlated. Even though this is not a}

correlation by chance because CSRand CSR2 _{cannot vary independently of each other, this issue is addressed in}

the robustness subsection.

2_{After dropping outliers, ROE has a correlation of 0.7 with ROA. Therefore, only ROA is used as accounting}

11

Furthermore, literature (Nelling and Webb, 2009) suggests that the effect of CSR on performance is mainly driven by unobservable firm characteristics. Therefore, the redundant fixed effects test is executed to infer if there is unobservable heterogeneity across firms. In case of significance, fixed or random effects are added. A Hausman test determines if fixed or random effects are suitable. Fixed effects are added in case of significance of the Hausman test, while in case of insignificance random effects are added. This is denoted by 𝜇_𝑖. Time fixed effects are included to capture time trends in the relationship. Besides, this captures the effect of the crisis. Time fixed effects are denoted by 𝜆_𝑡. Furthermore, standard errors are clustered at the firm level to correct for heteroskedasticity and autocorrelation due to significance of the Wooldridge test for autocorrelation in panel models. Additionally, for all dependent and independent variables stationarity is tested to prevent a spurious regression. A Fisher unit root test for panel data rejected the null hypothesis that all panels contain unit roots for every variable at 1% significance, which implies the variables are stationary. An overview of all variables is given in Appendix A. In summary, this results in the following regressions:

𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝑖,𝑡 = 𝛼𝑖+ 𝛽1∗ 𝐶𝑆𝑅𝑖,𝑡+ 𝛽2∗ 𝐶𝑆𝑅𝑖,𝑡∗ 𝐵𝑖+ 𝛽3∗ 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡+ 𝜇𝑖+ 𝜆𝑡+ 𝑣𝑖,𝑡 (1)

𝑃𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝑖,𝑡 = 𝛼𝑖+ 𝛽1∗ 𝐶𝑆𝑅𝑖,𝑡+ 𝛽2∗ 𝐶𝑆𝑅𝑖,𝑡2 + 𝛽3∗ 𝐶𝑆𝑅𝑖,𝑡∗ 𝐵𝑖+ 𝛽4∗ 𝐶𝑆𝑅𝑖,𝑡2 ∗ 𝐵𝑖+ 𝛽5∗

𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡+ 𝜇𝑖+ 𝜆𝑡+ 𝑣𝑖,𝑡 (2)

12

To address the main endogeneity concern - reverse causality between CSR and financial performance - the general accepted method is to implement an instrumental variable (IV) approach which is estimated by two-stage least squares (2SLS). This requires finding a valid - relevant and exogenous - instrument. A valid instrument strongly affects CSR, while it should only affect performance through CSR. As a rule of thumb, a F-statistic of at least 10 is required in the first stage for an instrument to be relevant (Staiger and Stock, 1997). Given the broad set of control variables and the estimation by company and time fixed effects it is problematic to find such a strong instrument. Additionally, exogeneity is a strong requirement as performance is affected in much ways. Due to these problems the cure can be worse than the disease in this case (Bound et al., 1993). So, as the main issue of endogeneity is reverse causality, a granger causality test is executed. A crucial assumption in order to retrieve reliable inferences by this test is stationarity of the variables. As has been stressed earlier this assumption is satisfied. The proposed test of Granger (1969) to establish causality is:

𝑦𝑖,𝑡= α𝑖+ ∑ 𝛾𝑖 (𝑘) ∗ 𝑦𝑖,𝑡−𝑘 𝐾 𝑘=1 + ∑ 𝛽𝑖 (𝑘) ∗ 𝑥𝑖,𝑡−𝑘 𝐾 𝑘=1 + 𝜀𝑖,𝑡 (3)

In order to establish causality, the null hypothesis of 𝛽_𝑖(𝑘) = 0 (no causality from x to y) is tested against the alternative hypothesis 𝛽_𝑖(𝑘)≠ 0 (causality from x to y). In this study the Granger causality is executed between CSR and performance following a variant implemented by Dyck et al. (2019). I execute two symmetric sets of regressions to establish causality. Firstly, performance is regressed on lagged performance, lagged CSR and lagged control variables. Secondly, CSR is regressed on lagged performance, lagged CSR and lagged control variables. In this way, causality is allowed to go in both directions. For panel data the Granger causality test can be executed by pooling the data and including firm fixed effects via least squares dummy variables estimations. This is valid as long as the number of time periods is relatively small compared to the number of cross-sectional units (Holtz-Eakin et al., 1988). The Granger causality test is conducted to infer the reliability of the results found by regression (1) and (2) and is, therefore, addressed in the robustness section.

3.2 CSR and risk-taking

13

probability of insolvency. Z-score is calculated as (return-on-assets + capital-to-assets) / standard deviation of return-on-assets. Additionally, both papers use another measure of risk-taking as a robustness check. Bolton (2013) perceives banks which received financial assistance during the crisis as riskier. Scholtens and van ‘t Klooster (2019) only rely on accounting-based measures and use volatility of ROE as second proxy of risk-taking. Consistent with the literature, I use the Z-score to proxy risk-taking as well. However, results can be very dependent on the variable choice (Griffin and Mahon, 1997). Limiting risk-taking to the Z-score as dependent variable might predetermine the ultimate outcome. This necessitates estimating the regression using a broader set of risk-taking variables. Therefore, the annualized volatility of monthly stock returns is used as a market-based proxy of risk-taking. This is consistent with Habib and Hasan (2017) and Harjoto and Laksmana (2018). Another generally accepted proxy used by Habib and Hasan (2017), Yung and Chen (2018) and Harjoto and Laksmana (2018) is R&D expenses to total assets. R&D expenses have a highly uncertain return, low likelihood of success and the potential pay-off is very distant. So, R&D expenses to total sales proxy the willingness of a firm to take risk. However, as explained in section 3.1, availability of R&D expenses is very industry dependent and missing for a significant number of firms. To prevent biasing the results by dropping complete industries, I do not use R&D / total sales as a proxy for risk-taking, but once more intangible assets / total assets. The ESG combined scores are used to measure CSR again. Control variables are the same as in (1) and (2) without risk and intangibles because these are both proxies of risk-taking. Leverage is included because a higher leverage increases default risk. Liquidity is a control variable because cash generates a buffer to outside shocks and, consequently, reduces risk. Size is included because size is positively correlated with risk-taking for banks (Bhagat, Bolton and Lu, 2012). Additionally, ROAt-1 is

included as a control variable. Risk-taking can be very dependent on last year’s performance. Firms having a bad performance in the previous year might be tempted to take on more risk in order to improve performance (Bromiley, 1991). A dummy (taking the value of 1 for banks and 0 otherwise) interacting with CSR is included to estimate if risk-taking has a different sensitivity to CSR for banks than for non-banks. For the same reason and using the same procedure as in section 3.1, company fixed effects and time fixed effects are added and standard errors are clustered at the firm level. Furthermore, for the newly added variables the Fisher unit root test for panel data again rejects non-stationarity at 1% significance. This results in the following regression to assess the relationship between CSR and risk-taking:

14

4. Data

4.1 Sample selection

This paper focusses on Europe. Liang and Renneboog (2017) find evidence that the institutional framework in which firms operate impacts the outcomes. Using a sample of European firms, which are active in a comparable institutional framework, minimizes the effect of institutional differences. Therefore, the sample consists of the constituents of the STOXX 600 on 30 September 2019. The STOXX 600 consists of 600 small, medium and large capitalization firms across 17 developed European countries. A large part of these countries is member of the European Union (EU). European countries should converge economically and institutionally in order to join the EU. After joining, the EU members are subject to the same European regulations. Furthermore, part of this sample is member of the Economic and Monetary Union (EMU) and because of this operates under the same monetary policy regime. STOXX 600 constituents as well as all other data are extracted from Thomson Reuters Eikon. The data downloaded from this database are ROA, total assets, industry, market capitalization, book value of debt, book value of equity, cash, intangible assets, monthly stock returns and the ESG combined scores and their respective pillars. Subsequently, the relevant variables (Tobin’s Q, leverage, intangibles, liquidity, Z-Score and risk) are constructed using this data. The main variable of interest in this study is CSR. Consequently, it is crucial to measure CSR as precise as possible. Appendix B displays how the ESG combined score is constituted. Thomson Reuters Eikon covers several stock market indices in their ESG database dependent on geographical region. For Europe it covers the STOXX 600. By using the constituents of the STOXX 600 good ESG coverage is ensured. The data period covers the last 16 years of ESG data (2003-2018). The most important reason for this is to get a sufficiently large European sample. All variables are measured on a yearly basis.

4.2 Descriptive statistics

15

methodology to calculate Tobin’s Q, show similar descriptive statistics for Tobin’s Q. They find an average of 1.51 and a maximum of 12.16, which is consistent with my descriptive statistics. Secondly, a high Z-Score can occur if the firm has a stable ROA over the years and only a limited number of observations. This inflates the Z-Score. A high maximum Z-Score is not a mismeasurement, but just a consequence of using a specific variant of the Z-Score. For instance, Bolton (2013) has an average Z-score of 29.57, which is even slightly higher. However, their maximum Z-score (97.831) is significantly lower than mine. On the other hand, Scholtens and van ‘t Klooster find an even higher maximum of 1189.02. Therefore, the Z-scores of my sample are not out of line with other literature. Thirdly, there are firms with a very low leverage. This is not inconsistent as Barnett and Salomon (2012), using a similar metric for leverage, observe the same. In appendix C the number of observations per industry and year are presented for the CSR variable. This gives additional insight into the composition of the sample. Table 1. Descriptive statistics.

The columns describe the number of observations (N), average (Mean), standard deviations (Sd), minimum (Min.) and maximum (Max.). For the definition of all row variables I refer to Appendix A.

Variables N Mean Sd. Min. Max. CSR 6,948 52.97 15.91 7.29 93.64 ROA 6,096 6.32 5.96 -14.20 35.48 RISK 6,875 7.54 3.84 2.22 41.79 Z 6,067 27.02 27.50 -3.24 291.81 Tobin’s Q 6,655 1.36 1.44 0.03 13.46 LEV 6,751 0.24 0.15 0.00 0.71 LIQ 6,844 0.01 0.01 0.00 0.08 SIZ 6,947 23.51 1.92 18.79 28.57 INT 6,510 0.07 0.10 0.00 0.56

16

literature. Nevertheless, many papers do not provide correlation coefficients or use different variables, so focus is on the most prominent correlations. For instance, leverage is negatively correlated with performance, which is in accordance with Nollet et al. (2016) and Nelling and Webb (2009). Size and performance are strongly negatively correlated in Nelling and Webb (2009) too. The strong positive correlation of liquidity with performance is also found by Nguyen et al. (2015). Although they use a different measure of liquidity (current ratio), the variables capture the same notion. A last thing to note is the positive correlation between performance and intangibles. This is consistent with literature finding a positive relationship between R&D (as intangibles proxies R&D) and performance. This positive correlation is found by Nollet et al. (2016) as well.

Table 2. Matrix of correlations

Correlation matrix of all relevant variables. For the definition of all variables I refer to Appendix A.

Variables Tobin’s Q ROA Z CSR LEV LIQ SIZ INT RISK Tobin’s Q 1.00 ROA 0.64 1.00 Z -0.01 -0.05 1.00 CSR -0.01 -0.02 0.05 1.00 LEV -0.03 -0.11 0.02 0.00 1.00 LIQ 0.46 0.23 -0.06 -0.03 -0.14 1.00 SIZ -0.44 -0.47 -0.05 0.03 -0.08 -0.25 1.00 INT 0.18 0.14 0.02 -0.01 0.18 0.04 -0.17 1.00 RISK -0.07 -0.15 -0.18 -0.10 0.01 0.06 0.01 -0.05 1.00

5. Results

The results from the analysis of CSR and financial performance are presented in table 3. In both linear regressions the interaction term between CSR and being a bank is significantly different from zero at one percent significance. This means that the relationship between CSR and performance differs between banks and non-banks. In this case, the sensitivity of

performance to CSR is lower for banks than for non-banks. This observation provides

17

exists for both ROA and Tobin’s Q as dependent variables. The coefficients of CSR are significantly different from zero at the one percent significance level. As the CSR coefficients only measure the effect of CSR on performance for non-banks, evidence is found for

hypothesis 2 that the sign of the relationship between CSR and performance is positive for non-banks. Besides a linear relationship, previous studies are suggested a non-linear (U-shaped) relationship. In order for this relationship to be true for non-banks, a significant negative sign for CSR and a significant positive sign for CSR2 would be expected. In that case, the marginal effect would be negative first, but change sign for higher CSR scores. However, the quadratic models provide no evidence for this. In table 3 column 3 and 4 the sign for CSR is positive and for CSR2 negative. This is in contradiction with the expected signs and would indicate a positive, but decreasing marginal effect of CSR on performance. Nevertheless, none of the quadratic terms is significantly different from zero. Consequently, no evidence is found for any quadratic relationship. Past findings of a U-shaped relationship might thus be very sample dependent and not directly generalizable. Secondly, from table 3 the effect of CSR on performance for banks can be inferred by summing up the coefficients of CSR and B*CSR 3. Subsequently, in all regressions it has been tested if the CSR coefficients for banks are significantly different from zero 4. This shows that these coefficients are not significantly different from zero in any regression. Similarly, the coefficient CSR2 _{for banks is}

retrieved by adding up the coefficient of CSR2 and B*CSR2. In consensus with non-banks, these coefficients are not found to be significantly different from zero. Therefore, both in the linear and non-linear model, there does not seem to be any significant association between CSR and performance in the banking industry. Hence, for banks specifically I fail to find evidence for hypothesis 2. This finding is in line with several other papers in the banking industry such as Fijalkowska et al. (2018) and Islam et al. (2012).

3 _{To illustrate, in column 1 the CSR coefficient is 0.010 and the interaction coefficient of B*CSR is -0.013. This}

means the effect of CSR on financial performance for banks is 0.010 - 0.013 = -0.003. In the remaining footnotes, the sum of the CSR coefficient and the interaction coefficient of B*CSR is denoted by γ. So, in column 1, γ = -0.003.

4 _{To test the significance of the CSR coefficient for banks the null hypothesis γ = 0 is tested against the}

18

Table 3. Relationship between CSR and performance.

Estimation of the regressions (1) and (2). The first two models are a linear model, while the last two models are non-linear. The dependent variables are ROA and Q and are stated in the different column headings. All explanatory variables are in the rows. For the definition of all variables I refer to Appendix A. B is a dummy taking the value 1 if the firm is a bank and 0 otherwise. The models are estimated by fixed effects instead of random effects due to significance of the Hausman test. There is unobservable heterogeneity across firms due to significance of the redundant fixed effects test. This necessitates to estimate by fixed effects instead of pooled OLS. Standard errors are clustered at the firm level. Although not tabulated, the model is estimated with an intercept.

(1) (2) (3) (4)

VARIABLES ROA Tobin’s Q ROA Tobin’s Q CSR 0.010*** 0.003*** 0.017 0.005 (0.004) (0.001) (0.022) (0.006) CSR2 _{-0.000 -0.000} (0.000) (0.000) B*CSR -0.013*** -0.004*** -0.049* -0.014 (0.005) (0.001) (0.030) (0.009) B*CSR2 _{0.000 0.000} (0.000) (0.000) SIZ -1.411*** -0.434*** -1.413*** -0.434*** (0.329) (0.133) (0.329) (0.133) LEV -7.329*** 0.020 -7.325*** 0.021 (1.445) (0.401) (1.444) (0.401) LIQ 25.724 7.501 25.734 7.504 (16.721) (4.750) (16.744) (4.753) RISK -0.141*** -0.015** -0.141*** -0.015** (0.024) (0.006) (0.024) (0.006) INT -3.082** -0.242 -3.091** -0.246 (1.498) (0.362) (1.499) (0.362) Observations 5,496 6,133 5,496 6,133 Number of companies 557 565 557 565 Company fixed effects Yes Yes Yes Yes Time fixed effects Yes Yes Yes Yes Hausman p-value 0.000 0.000 0.000 0.000 Redundant fixed effects p-value 0.000 0.000 0.000 0.000

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

19

outperformance. For the social performance pillar the coefficient is statistically different from zero at 5% for both measures of performance. A strong component of the social pillar score is workforce treatment. This suggests that companies which are better able to manage its workforce engagement have a better performance. Besides, firms outperforming in responsible environmental practices do only outperform in terms of Tobin’s Q. This might indicate environmental business practices do not directly result in better accounting performance, but given the significance for Tobin’s Q as dependent variable it is valued by investors. Tobin’s Q not only incorporates current performance, but perceived future growth as well. Hence, investors perceive doing business in an environmental sound manner is the way to go and ultimately will improve business performance. The insensitivity of performance to the governance pillar indicates that investments in corporate governance are not worthwhile if the sole goal is performance enhancement. Secondly, the question remains what the driver of the insensitivity to CSR is for banks. Table 4 shows only a significant negative effect is found for the social pillar score with Tobin’s Q as dependent variable. Nevertheless, almost all coefficients on the interaction terms show a negative sign. This suggests the lower sensitivity of performance to CSR found for banks compared to non-banks is really a combination of all the different pillars instead of traced back to one specific pillar.

Table 4. Relationship between the respective pillars and performance.

Estimation of the regression (1) for the different pillars of CSR. The dependent variables are ROA and Q and are stated in the different column headings. All explanatory variables are in the rows. For the definition of all variables I refer to Appendix A. B is a dummy taking the value 1 if the firm is a bank and 0 otherwise. The models are estimated by fixed effects instead of random effects due to significance of the Hausman test. There is unobservable heterogeneity across firms due to significance of the redundant fixed effects test. This necessitates to estimate by fixed effects instead of pooled OLS. Standard errors are clustered at the firm level. Although not tabulated, the model is estimated with an intercept.

20 LEV -7.396*** 0.001 -7.362*** 0.019 -7.395*** 0.007 (1.448) (0.400) (1.445) (0.400) (1.457) (0.402) LIQ 25.313 7.334 25.631 7.552 25.681 7.515 (16.812) (4.817) (16.651) (4.760) (16.745) (4.772) RISK -0.143*** -0.016** -0.141*** -0.015** -0.144*** -0.016*** (0.024) (0.006) (0.024) (0.006) (0.024) (0.006) INT -3.020** -0.228 -3.061** -0.242 -3.046** -0.223 (1.505) (0.362) (1.498) (0.363) (1.499) (0.364) Observations 5,496 6,133 5,496 6,133 5,496 6,133 Number of companies 557 565 557 565 557 565 Company fixed effects

Yes Yes Yes Yes Yes Yes Time fixed effects Yes Yes Yes Yes Yes Yes Hausman p-value 0.000 0.000 0.000 0.000 0.000 0.000 Redundant fixed

effects p-value

0.000 0.000 0.000 0.000 0.000 0.000 Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

5.2 CSR and risk-taking

21

for non-banks. Secondly, the coefficients of CSR for banks are obtained in the same fashion as a in the previous subsection. The stepwise procedure has been outlined in footnote 3 and 4. This allows to test if the effect of CSR on bank risk-taking is significantly different from 0 (hypothesis 4). The results found are consistent across the different proxies of risk-taking. There is no relationship found between CSR and risk-taking of banks. None of coefficients is found to be significantly different from zero. Therefore, no evidence supporting hypothesis 4 is found in the banking industry. This finding is in line with Bolton (2013), where also no significant relationship between total CSR and bank risk-taking has been found. However, it is inconsistent with Scholtens and van ‘t Klooster (2019). The main difference between these two studies is the choice of variables. Bolton (2013), uses general control variables similar to my study, while Scholtens and van ‘t Klooster (2019) tailor their control variables to the specifics of the banking industry. This might explain the heterogeneity in findings.

Table 5. Relationship between CSR and risk-taking.

Estimation of regression (4). The dependent variables are the Z-Score, Risk and Intangibles and are stated in the different column headings. All explanatory variables are in the rows. For the definition of all variables I refer to Appendix A. B is a dummy taking the value 1 if the firm is a bank and 0 otherwise. The models are estimated by fixed effects instead of random effects due to significance of the Hausman test. There is unobservable

heterogeneity across firms due to significance of the redundant fixed effects test. This necessitates to estimate by fixed effects instead of pooled OLS. Standard errors are clustered at the firm level. Although not tabulated, the model is estimated with an intercept.

(1) (2) (3)

VARIABLES Z RISK INT

CSR 0.023** -0.018*** 0.000 (0.011) (0.003) (0.000) B*CSR -0.053** 0.009 0.000 (0.022) (0.019) (0.000) ROAt-1 0.094*** -0.092*** -0.000 (0.022) (0.018) (0.000) SIZ -2.585 0.282 0.028*** (1.675) (0.199) (0.006) LEV -27.990*** 3.373*** 0.048** (3.320) (0.975) (0.021) LIQ 26.128 14.416 -0.455** (17.379) (10.053) (0.209) Observations 5,077 5,383 5,113 Number of companies 539 552 538 Company fixed effects Yes Yes Yes

Time fixed effects Yes Yes Yes

Hausman p-value 0.000 0.000 0.000 Redundant fixed effects p-value 0.000 0.000 0.000

22

To give more insight into the way CSR can reduce risk-taking there is differentiated between the different components of CSR in table 6. As pointed out and illustrated in the previous subsection, the various components of CSR can have heterogeneous effects. I am mainly interested in the regressions with the Z-Score and Risk as dependent variable because Intangibles as risk-taking proxy did not provide valuable insights both for banks and non-banks previously. Table 6 shows that the results found for non-banks using the Z-Score and Risk variable have different drivers. The positive relationship found between CSR and the Z-Score is driven by corporate governance outperformance. On the other hand, the negative relationship between CSR and the risk variable is due to differences in environmental and social performance. This leads to following conclusions. First of all, a strong governance structure is valuable for companies to reduce insolvency risk. Secondarily, investors base the riskiness of the businesses mainly on the environmental and social component of CSR. Therefore, for the drivers of the negative relationship between CSR and risk-taking there seems to be a mismatch between accounting-based and market-based measures of risk-taking. Furthermore, investigation of the interaction variable for banks does not provide an unambiguous answer regarding the driver of the neutral relationship between CSR and risk-taking for banks. For the Z-Score as dependent variable no significance is found for the interaction term and the signs do also not provide convincing evidence. For the risk variable the cause of the insignificant result for banks found earlier seems to be caused by the environmental pillar. This finding can be due to the difference in operations of banks compared to non-banks. For instance, manufacturing firms have a strong public pressure to reduce CO2 emissions, while for a service provider such

23 Table 6. Relationship between the respective pillars and risk-taking.

Estimation of regression (4) for the different pillars of CSR. The dependent variables are the Z-Score, Risk and Intangibles and are stated in the different column headings. All explanatory variables are in the rows. For the definition of all variables I refer to Appendix A. B is a dummy taking the value 1 if the firm is a bank and 0 otherwise. The models are estimated by fixed effects instead of random effects due to significance of the Hausman test. There is unobservable heterogeneity across firms due to significance of the redundant fixed effects test. This necessitates to estimate by fixed effects instead of pooled OLS. Standard errors are clustered at the firm level. Although not tabulated, the model is estimated with an intercept.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

VARIABLES Z RISK INT Z RISK INT Z RISK INT

EPS -0.011 -0.015*** -0.000 (0.019) (0.005) (0.000) B*EPS 0.033 0.054*** -0.000*** (0.032) (0.013) (0.000) SPS 0.020 -0.019*** 0.000 (0.015) (0.005) (0.000) B*SPS -0.043 0.030 -0.000* (0.031) (0.024) (0.000) GPS 0.021* -0.002 0.000 (0.011) (0.004) (0.000) B*GPS -0.009 -0.014 -0.000 (0.023) (0.023) (0.000) ROAt-1 0.099*** -0.093*** -0.000 0.094*** -0.091*** -0.000 0.098*** -0.095*** -0.000 (0.023) (0.019) (0.000) (0.023) (0.019) (0.000) (0.022) (0.019) (0.000) SIZ -2.505 0.285 0.029*** -2.604 0.306 0.028*** -2.595 0.252 0.028*** (1.689) (0.199) (0.006) (1.695) (0.199) (0.006) (1.679) (0.200) (0.006) LEV -28.125*** 3.486*** 0.048** -28.054*** 3.446*** 0.048** -28.092*** 3.497*** 0.048** (3.322) (0.986) (0.021) (3.345) (0.974) (0.021) (3.320) (0.989) (0.021) LIQ 24.902 13.708 -0.438** 26.454 14.510 -0.452** 26.667 15.702 -0.450** (17.481) (10.276) (0.209) (17.562) (10.119) (0.209) (17.380) (10.219) (0.209) Observations 5,077 5,383 5,113 5,077 5,383 5,113 5,077 5,383 5,113 Number of companies 539 552 538 539 552 538 539 552 538

Company fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes

Time fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes

Hausman p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Redundant fixed effects p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Robust standard errors in parentheses

24

Regarding the difference in findings for banks compared to non-banks three plausible interpretations are proposed. First of all, banks are really different than non-banks and this is reflected in the results. In the banking industry CSR does not enhance performance and is unable to reduce risk-taking behaviour. Overall, the role CSR can play in the banking industry is just very limited. Secondly, CSR can improve performance and reduce risk-taking in the banking industry too, but real engagement to CSR is lower for banks. CSR scores are mainly based on public reports of the company and companies are well aware of this. Banks might exploit these reports to show their engagement to responsible business practices to the public, while the real resources used to carry out the responsible activities are limited. This is in accordance with the greenwashing hypothesis of Baron (2001). Thirdly, given that banks fulfil a very different role in the economy than other firms, using general variables might not fully capture the specifics of the banking industry. As a consequence, insensitivity of performance and risk-taking to CSR could have been found for banks.

5.3 Robustness

The previous results have been obtained by winsorizing the data at the 1st and 99th percentile on a yearly basis. The data in these tails of the distribution had been set equal to the value at the percentile cut-offs. In this way, the effect of outliers on the results is reduced. This ensures the results are not driven by potential erroneous outliers. On the other hand, outliers might also contain valuable information and should therefore be kept in. In order to assess the impact of the decision to winsorize, the models are re-estimated while including all data. The results of this robustness check are presented in appendix D. Comparing the robustness results with the winsorized results shows that winsorizing does not change the results significantly. All CSR coefficient remain the same sign. The significance of the CSR variables does not change, except in two cases. In the non-linear performance regression (table D1, column 3), the linear interaction variable changes from being significant to being insignificant. Furthermore, in the relationship between CSR and risk-taking significance of CSR changed from 5% significance to 10% significance with Z-Score as dependent variable (table D2, column 1). Overall, the conclusion can be drawn that the results are fairly robust to outliers.

25

the null hypothesis of non-causality. There is evidence that CSR “granger” causes performance. Additionally, in column 3 and 4 the coefficients of lagged performance determine if there could be simultaneity. This provides mixed evidence. The Granger causality test rules out reverse causality as an alternative explanation of the results found with ROA. For Tobin’s Q simultaneity cannot be ruled out completely, but column 4 shows only a weakly significant effect from Tobin’s Q to CSR, which indicates causality is stronger from CSR to Tobin’s Q than in the reversed direction. Overall, there is convincing evidence that CSR causes performance, while the evidence for simultaneity is limited.

Table 7. Granger causality test

Estimation of the Granger causality regression (3). The columns show the different dependent variables. All explanatory variables are in the rows. For the definition of all variables I refer to Appendix A The models have been estimated with the same control variables as in table 3, but lagged one year. Standard errors are clustered at the firm level. Although not tabulated, the model is estimated with an intercept.

(1) (2) (3) (4) VARIABLES ROA Tobin’s Q CSR CSR

ROAt-1 0.506*** 0.093 (0.043) (0.064) Tobin’s Qt-1 0.584*** 0.689* (0.048) (0.394) CSRt-1 0.006** 0.001*** 0.164*** 0.167*** (0.003) (0.000) (0.020) (0.019) Observations 4,739 5,533 5,034 5,577 Control variables Yes Yes Yes Yes Company fixed effects Yes Yes Yes Yes Time fixed effects Yes Yes Yes Yes

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

26

27

6. Conclusion

In the past years the interest in CSR practices in the banking industry has increased significantly. Nevertheless, the effects of CSR are far from being undisputed. Therefore, this paper revisited the relationship between CSR and financial performance and additionally investigated if CSR can be a valuable tool to reduce risk-taking behaviour. The period 2003-2018 is investigated for European firms with a special focus on European banks. This is highly relevant because the global financial crisis can be traced back to financial institutions taking too much risk. A focus on more responsible business practices could have been adequate to keep risks manageable. The sole reason for inadequate CSR adoption could be reluctance of banks to spend sufficient resources on social business practices because they believe these resources are not put in productive use.

28

proxied by the CSR reports of the companies itself. Without generally accepted standards and independent auditing of the CSR reports it remains doubtful that banks with a higher score also actually engage in more socially responsible business practices.

29

7. References

Adjei, F. (2013). The effects of cash holdings on corporate performance during a credit crunch: evidence from the sub-prime mortgage crisis. Journal of Economics and Finance, 37(2), 188-199.

Altunbas, Y., Gambacorta, L., & Marques-Ibanez, D. (2010). Does monetary policy affect bank risk-taking? BIS Working Papers No. 298.

Aupperle, K. E., Carroll, A. B., & Hatfield, J. D. (1985). An empirical examination of the relationship between corporate social responsibility and profitability. Academy of

management Journal, 28(2), 446-463.

Barnett, M. L., & Salomon, R. M. (2012). Does it pay to be really good? Addressing the shape of the relationship between social and financial performance. Strategic Management

Journal, 33(11), 1304-1320.

Baron, D. (2001). Private politics, corporate social responsibility and integrated strategy. Journal of Economics and Management Strategy 10, 7-45.

Bhagat, S., Bolton, B., & Lu, J. (2012). Size, leverage, and risk-taking of financial institutions. Journal of Banking & Finance, 59, 520-537.

Boddy, C. R. (2011). The corporate psychopaths theory of the global financial crisis. Journal of Business Ethics, 102(2), 255-259.

Bolton, B. J. (2013). Corporate social responsibility and bank performance. Available at: http://ssrn.com/ abstract=2277912.

Bound, J., Jaeger, D. A., & Baker, R. (1993). The cure can be worse than the disease: A cautionary tale regarding instrumental variables. NBER Technical Working Paper No. 137. Cambridge, MA: National Bureau of Economic Research.

Bromiley, P. (1991). Testing a causal model of corporate risk taking and performance. Academy of Management journal, 34(1), 37-59.

Cho, S. J., Chung, C. Y., & Young, J. (2019). Study on the relationship between CSR and financial performance. Sustainability, 11(2), 343.

30

Cornett, M. M., Erhemjamts, O., & Tehranian, H. (2016). Greed or good deeds: An examination of the relation between corporate social responsibility and the financial performance of US commercial banks around the financial crisis. Journal of Banking & Finance, 70, 137-159.

Dahlsrud, A. (2008). How corporate social responsibility is defined: an analysis of 37 definitions. Corporate social responsibility and environmental management, 15(1), 1-13. Dam, L., Koetter, M., & Scholtens, B. (2009). Why do firms do good? Evidence from managerial efficiency. Evidence from Managerial Efficiency.

Delis, M. D., & Kouretas, G. P. (2011). Interest rates and bank risk-taking. Journal of Banking & Finance, 35(4), 840-855.

Dyck, A., Lins, K. V., Roth, L., & Wagner, H. F. (2019). Do institutional investors drive corporate social responsibility? International evidence. Journal of Financial

Economics, 131(3), 693-714.

Echambadi, R., & Hess, J. D. (2007). Mean-centering does not alleviate collinearity problems in moderated multiple regression models. Marketing Science, 26(3), 438-445.

European Commission. Directorate-General for Employment. (2001). Promoting a European framework for corporate social responsibility: Green paper. Office for Official Publications of the European Communities.

Fijałkowska, J., Zyznarska-Dworczak, B., & Garsztka, P. (2018). Corporate social-environmental performance versus financial performance of banks in Central and Eastern European countries. Sustainability, 10(3), 772.

Freeman, R.E. (1984). Strategic management: a stakeholder approach. London: Pitman.

Friede, G., Busch, T., & Bassen, A. (2015). ESG and financial performance: aggregated evidence from more than 2000 empirical studies. Journal of Sustainable Finance & Investment, 5(4), 210-233.

Friedman, M. (1970). The Social Responsibility of Business is to Increase its Profits. New York Times Magazine.

31

Griffin, J. J., & Mahon, J. F. (1997). The corporate social performance and corporate financial performance debate: Twenty-five years of incomparable research. Business & society, 36(1), 5-31.

Habib, A., & Hasan, M. M. (2017). Firm life cycle, corporate risk‐ taking and investor sentiment. Accounting & Finance, 57(2), 465-497.

Harjoto, M., Jo, H., & Kim, Y. (2017). Is institutional ownership related to corporate social responsibility? The nonlinear relation and its implication for stock return volatility. Journal of Business Ethics, 146(1), 77-109.

Harjoto, M., & Laksmana, I. (2018). The impact of corporate social responsibility on risk taking and firm value. Journal of Business Ethics, 151(2), 353-373.

Harjoto, M. A., Laksmana, I., & Yang, Y. W. (2018). Board diversity and corporate risk taking. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2412634.

Holtz-Eakin, D., Newey, W., & Rosen, H. S. (1988). Estimating vector autoregressions with panel data. Econometrica: Journal of the Econometric Society, 1371-1395.

Houston, J. F., Lin, C., Lin, P., & Ma, Y. (2010). Creditor rights, information sharing, and bank risk taking. Journal of Financial Economics, 96(3), 485-512.

Hull, C. E., & Rothenberg, S. (2008). Firm performance: The interactions of corporate social performance with innovation and industry differentiation. Strategic Management

Journal, 29(7), 781-789.

Islam, Z. M., Ahmed, S. U., & Hasan, I. (2012). Corporate social responsibility and financial performance linkage: Evidence from the banking sector of Bangladesh. Journal of

Organizational Management, 1(1), 14-21.

Jiménez, G., Lopez, J. A., & Saurina, J. (2013). How does competition affect bank risk-taking? Journal of Financial stability, 9(2), 185-195.

Liang, H., & Renneboog, L. (2017). On the foundations of corporate social responsibility. The Journal of Finance, 72(2), 853-910.

McWilliams, A., & Siegel, D. (2000). Corporate social responsibility and financial

32

McWilliams, A., & Siegel, D. (2001). Corporate social responsibility: A theory of the firm perspective. Academy of management review, 26(1), 117-127.

Mishra, S., & Suar, D. (2010). Does corporate social responsibility influence firm performance of Indian companies? Journal of business ethics, 95(4), 571-601.

Nelling, E., & Webb, E. (2009). Corporate social responsibility and financial performance: the “virtuous circle” revisited. Review of Quantitative Finance and Accounting, 32(2), 197-209.

Nguyen, B. T. N., Tran, H. T. T., Le, O. H., Nguyen, P. T., Trinh, T. H., & Le, V. (2015). Association between corporate social responsibility disclosures and firm value–Empirical evidence from Vietnam. International Journal of Accounting and Financial Reporting, 5(1), 212-228.

Nollet, J., Filis, G., & Mitrokostas, E. (2016). Corporate social responsibility and financial performance: A non-linear and disaggregated approach. Economic Modelling, 52, 400-407. Orlitzky, M., Schmidt, F. L., & Rynes, S. L. (2003). Corporate social and financial

performance: A meta-analysis. Organization studies, 24(3), 403-441.

Preston, L. E., & O'bannon, D. P. (1997). The corporate social-financial performance relationship: A typology and analysis. Business & Society, 36(4), 419-429.

Saeidi, S. P., Sofian, S., Saeidi, P., Saeidi, S. P., & Saaeidi, S. A. (2015). How does corporate social responsibility contribute to firm financial performance? The mediating role of

competitive advantage, reputation, and customer satisfaction. Journal of business research, 68(2), 341-350.

Scholtens, B. (2009). Corporate social responsibility in the international banking industry. Journal of Business Ethics, 86(2), 159-175.

Scholtens, B., & van’t Klooster, S. (2019). Sustainability and bank risk. Palgrave Communications, 5(1), 1-8.

Scholtens, B., & Zhou, Y. (2008). Stakeholder relations and financial performance. Sustainable Development, 16(3), 213-232.

33

Staiger, D., & J.H. Stock (1997). Instrumental Variables Regression with Weak Instruments. Econometrica, 65, 557-586.

Waddock, S A., & Graves B. S. (1997). The corporate social performance-financial performance link. Strategic Management Journal 18 (4), 303-319.

Wright, P., & Ferris, S. P. (1997). Agency conflict and corporate strategy: The effect of divestment on corporate value. Strategic management journal, 18(1), 77-83.

34

Appendix A: list of variables. Table A1. Variables

All variables (left column), abbreviations (middle column) and its calculation or proxy (right column).

VARIABLES Abbreviation Definition

Return on assets ROA Net income

Total assets∗ 100

Tobin’s Q - Market capitalization + book value of debt

Total assets Corporate social

responsibility

CSR ESG Combined Score

Corporate social responsibility2

CSR2 (ESG Combined Score)2

Environmental performance

EPS Environmental pillar score

Social performance SPS Social pillar score

Governance performance

GPS Governance pillar score

Z-Score Z

Return on assets + capital to assets σ_{return on assets}

Leverage LEV Debt

Total assets

Liquidity LIQ Cash

Total assets

Size SIZ ln(total assets)

Intangibles INT Intangible assets

Total assets Volatility of monthly

stock returns over the year

RISK σ_{monthly stock returns over the year}

Bank dummy B = 1 if firm is a bank

35

Appendix B: measurement of CSR

The main variable of interest in this study is CSR. Consequently, it is crucial to measure CSR as precise as possible. The key measure used for CSR is constructed by Thomson Reuters in the following way:

𝐸𝑆𝐺 𝐶𝑜𝑚𝑏𝑖𝑛𝑒𝑑 𝑆𝑐𝑜𝑟𝑒 = 𝐸𝑆𝐺 𝑆𝑐𝑜𝑟𝑒 𝑖𝑓 𝐶𝑜𝑛𝑡𝑟𝑜𝑣𝑒𝑟𝑠𝑦 𝑆𝑐𝑜𝑟𝑒 > 50

𝐸𝑆𝐺 𝐶𝑜𝑚𝑏𝑖𝑛𝑒𝑑 𝑆𝑐𝑜𝑟𝑒 = 𝐸𝑆𝐺 𝑆𝑐𝑜𝑟𝑒 𝑖𝑓 𝐸𝑆𝐺 𝑆𝑐𝑜𝑟𝑒 < 𝐶𝑜𝑛𝑡𝑟𝑜𝑣𝑒𝑟𝑠𝑦 𝑆𝑐𝑜𝑟𝑒 < 50

𝐸𝑆𝐺 𝐶𝑜𝑚𝑏𝑖𝑛𝑒𝑑 𝑆𝑐𝑜𝑟𝑒 = (𝐸𝑆𝐺 𝑆𝑐𝑜𝑟𝑒 + 𝐶𝑜𝑛𝑡𝑟𝑜𝑣𝑒𝑟𝑠𝑦 𝑆𝑐𝑜𝑟𝑒)/2 𝑖𝑓 𝐶𝑜𝑛𝑡𝑟𝑜𝑣𝑒𝑟𝑠𝑦𝑆𝑐𝑜𝑟𝑒 < 𝐸𝑆𝐺 𝑆𝑐𝑜𝑟𝑒 𝑎𝑛𝑑 𝐶𝑜𝑛𝑡𝑟𝑜𝑣𝑒𝑟𝑠𝑦 𝑆𝑐𝑜𝑟𝑒 < 50

Both the ESG score and the controversy score are measured on a scale from 1 to 100. A higher ESG score means better ESG performance and a higher controversies score means a firm has been engaged in a lower number of controversies in the past fiscal year.

The ESG Score is further disentangled as follows:

𝐸𝑆𝐺 𝑆𝑐𝑜𝑟𝑒 = 𝑥₁∗ 𝐸𝑛𝑣𝑖𝑟𝑜𝑛𝑚𝑒𝑛𝑡𝑎𝑙 𝑝𝑖𝑙𝑙𝑎𝑟 + 𝑥₂∗ 𝑆𝑜𝑐𝑖𝑎𝑙 𝑝𝑖𝑙𝑙𝑎𝑟 + 𝑥₃∗ 𝐺𝑜𝑣𝑒𝑟𝑛𝑎𝑛𝑐𝑒 𝑝𝑖𝑙𝑙𝑎𝑟 In this formula 𝑥1, 𝑥2 𝑎𝑛𝑑 𝑥3 are the weights of the respective pillars. All weights sum up to 1

and exact weights are 𝑥₁ = 0.34, 𝑥₂ = 0.355 𝑎𝑛𝑑 𝑥₃ = 0.305. The pillars can be

36

Appendix C: observations per industry and per year

Table C1. Observations per industry

This table shows the number of observations (N) for the variable CSR per industry. Industries are according to NAICS industry classification. In order to separate banks, the finance and insurance industry has been

subdivided into banks and other finance and insurance companies according to TRBC classifications.

Industry N

Accommodation and food services 97 Administrative and support and waste management and remediation services 172 Arts, entertainment and recreation 19

Banks 584

Construction 181

Health care and social assistance 35

Information 450

Manufacturing 2724

Mining, quarrying, and oil and gas extraction 169 Other finance and insurance 777 Other services (except public administration) 14 Professional, scientific and technical services 313 Real estate and rental and leasing 363

Retail trade 335

Transportation and warehousing 270

Utilities 315

Wholesale trade 130

Table C2. Observations per year

This table shows the number of observations (N) for the variable CSR per year.

37

Appendix D: robustness tables of full dataset

Table D1. Robustness of the relationship between CSR and performance.

This table corresponds to table 3 in the main text. All further information about the table is presented above table 3 in the main text. Compared to table 3, the results have been re-estimated by including all data instead of winsorizing at the 1st _{and 99}th _{percentile as a robustness check. For the definition of all variables I refer to}

Appendix A.

(1) (2) (3) (4)

VARIABLES ROA Tobin’s Q ROA Tobin’s Q

CSR 0.008** 0.004** 0.012 0.001 (0.004) (0.002) (0.023) (0.010) CSR2 _{-0.000 0.000} (0.000) (0.000) B*CSR -0.011** -0.004** -0.034 -0.008 (0.005) (0.002) (0.034) (0.012) B*CSR2 _{0.000 0.000} (0.000) (0.000) SIZ -1.410*** -0.564** -1.412*** -0.564** (0.340) (0.225) (0.339) (0.225) LEV -7.469*** 0.741 -7.468*** 0.739 (1.474) (0.998) (1.474) (1.000) LIQ 25.629*** 18.981*** 25.636*** 18.964*** (8.121) (6.366) (8.129) (6.381) RISK -0.139*** -0.016** -0.139*** -0.017** (0.025) (0.008) (0.025) (0.008) INT -3.151** -0.256 -3.155** -0.250 (1.447) (0.429) (1.447) (0.426) Observations 5,496 6,133 5,496 6,133 Number of companies 557 565 557 565 Company fixed effects Yes Yes Yes Yes Time fixed effects Yes Yes Yes Yes

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table D2. Robustness of the relationship between CSR and risk-taking.

This table corresponds to table 5 in the main text. All further information about the table is presented above table 5 in the main text. Compared to table 5, the results have been re-estimated by including all data instead of winsorizing at the 1st _{and 99}th _{percentile as a robustness check. For the definition of all variables I refer to}

Appendix A.

(1) (2) (3)

VARIABLES Z RISK INT

38 LEV -28.161*** 3.619*** 0.045** (3.926) (1.050) (0.022) LIQ 5.805 1.668 -0.228 (11.474) (11.976) (0.179) Observations 5,077 5,383 5,113 Number of companies 539 552 538

Company fixed effects Yes Yes Yes

Time fixed effects Yes Yes Yes

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Appendix E: estimations with lagged CSR

Table E1. Robustness CSR-performance relationship to lagged CSR

This table corresponds to table 3 in the main text. All further information about the table is presented above table 3 in the main text. Compared to table 3, the results have been re-estimated by lagging CSR and control variables as a robustness check. For the definition of all variables I refer to Appendix A.

(1) (2) (3) (4)

VARIABLES ROA Tobin’s Q ROA Tobin’s Q CSRt-1 0.010** 0.003*** 0.046** 0.006 (0.004) (0.001) (0.023) (0.005) CSR2 t-1 -0.000* -0.000 (0.000) (0.000) B* CSRt-1 -0.012** -0.004*** -0.084*** -0.016 (0.005) (0.001) (0.031) (0.010) B* CSR2 t-1 0.001** 0.000 (0.000) (0.000) SIZt-1 -2.150*** -0.349*** -2.160*** -0.349*** (0.322) (0.109) (0.322) (0.109) LEVt-1 -5.321*** -0.296 -5.293*** -0.295 (1.397) (0.335) (1.395) (0.334) LIQt-1 12.246 7.205 12.352 7.201 (15.162) (4.930) (15.167) (4.931) RISKt-1 -0.082*** -0.007 -0.081*** -0.007 (0.017) (0.005) (0.017) (0.005) INTt-1 -1.299 -0.035 -1.356 -0.040 (1.585) (0.336) (1.583) (0.336) Observations 5,000 5,551 5,000 5,551 Number of companies 535 547 535 547 Company fixed effects Yes Yes Yes Yes Time fixed effects Yes Yes Yes Yes

39

Table E2. Robustness CSR-risk taking relationship to lagged CSR

This table corresponds to table 5 in the main text. All further information about the table is presented above table 5 in the main text. Compared to table 5, the results have been re-estimated by lagging CSR and control variables as a robustness check. For the definition of all variables I refer to Appendix A.

(1) (2) (3)

VARIABLES Z RISK INT

CSRt-1 0.031** -0.008** 0.000 (0.013) (0.004) (0.000) B*CSRt-1 -0.052** 0.016 -0.000* (0.026) (0.017) (0.000) ROAt-1 0.074*** -0.084*** -0.000 (0.023) (0.018) (0.000) SIZt-1 -0.982 0.587*** 0.015*** (0.654) (0.201) (0.004) LEVt-1 -18.453*** 2.714*** 0.004 (2.449) (0.889) (0.020) LIQt-1 18.989 7.260 0.009 (19.455) (8.698) (0.186) Observations 5,061 5,372 5,100 Number of companies 537 553 539

Company fixed effects Yes Yes Yes

Time fixed effects Yes Yes Yes