How did the financial crisis affect the correlation between CSR performance and corporate financial performance in the financial sector?

(1)

BSC BETA GAMMA BEDRIJFSKUNDE

How did the financial

crisis affect the

correlation between CSR

performance and

corporate financial

performance in the

financial sector?

Supervisor: Katinka Quintelier

Secondary supervisor: Daniel Wäger

Jia Din Cai 6168914 7/15/2015

Firms are encouraged to engage in Corporate Social Responsibility (CSR) activities. Theoretically, conducting CSR activities improves stakeholder relations. This leads to increased Corporate Financial Performance (CFP) and a certain buffer of goodwill during times of crisis. CSR activities can also restore stakeholder relations. However, previous research has criticized these stated outcomes of CSR. In this research, I analyzed how this theory applies in the financial sector, which is responsible for the financial crisis in 2008. An amount of 58 American banks were researched with several regression models. Firstly, it was concluded that firms did not raise their CSR to restore stakeholder relations. Secondly, no significant relation between CSR and CFP was found. Thirdly, it was concluded that the crisis did not have a significant effect on the CSR-CFP relation.

(2)

1

This document is written by Student Jia Din Cai, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

2

Introduction

In the past decade, media and news have been dominated by the financial crisis and its effects on the economy, politics and the society (Schifferes & Roberts, 2015). Business models were forced to be revised and adapted immediately to the urgent economic circumstances (Jakob, 2012). Firms were forced to reevaluate all operations and internal structures, such as human resources, strategy and marketing.

The sector that was “hit” hard by the crisis is “financial services”. Ironically, this sector is also seen (by public, political and academic opinion) as the guilty party of the financial crisis around the year 2008 (Van Engelen, 2011). With the reputation severely damaged and the financial performance dropping, banks now have a need for reputational risk management in order to restore both trust and profitability. Corporate Social Responsibility activities are a form of reputational risk management in e.g. the financial services industry (Raithel & Wilczynsk et al., 2010).

However, while CSR can indeed increase financial performance, studies suggest that it can also backfire for firms that already have a bad reputation (Yoon & Gurhan-Canli, 2006). This research therefore tries to analyze Corporate Social Responsibility activities as a specific part of marketing, and its effect on the financial performances of banks. Therefore, the main research question this thesis tries to answer is: How did the financial crisis affect the correlation between CSR performance and corporate financial performance in the financial sector?

This paper starts with explaining the definition, goal and motivations of conducting CSR activities. Secondly, I present several similar empirical studies about this case. Thirdly, a review of several relevant theories is described in this paper. Both these theories and the conclusions of the empirical researches are applied to the financial sector industry. I then conduct a statistical analysis in order to test the hypotheses. Surprisingly, there were no significant results found in the analyses. I discuss the results in the light of research on the ‘business case’ for CSR.

(5)

4

Literature Review

Since a couple of decades, firms are being encouraged by stakeholders (e.g. customers, employees, suppliers and governments) to engage in Corporate Social Responsibility (CSR) activities. McWilliams and Siegel (2001) define CSR activities as “actions that appear to do

some social good, beyond the interests of the firm and that which is required by law”. These

activities focus on societal interest and concerns, such as environmental, poverty or public health issues. For example, 90% of 34,000 McDonalds restaurants uses recycled cooking oil in order to reduce its environmental implications (McDonald’s, 2015).

Why do most firms engage in CSR activities? First of all, the idea of giving back, engaging costly activities for societal interest and concerns, and raising its CSR-performance (CSRP) might sound ideological. However, many firms do benefit from such activities according to most academic researchers. CSR-activities improve corporate image, trustworthiness (McWilliams & Siegel, 2007) and the attitude from important stakeholders. This improved relationship leads to a decrease of transactions cost, which results in a financial

gain (Henisz & Dorobantu, 2014). But does CSR always contribute positively for every firm?

According to Barnett (2007), CSR has a varying effect on CFP and no definitive conclusion can be drawn from empirical studies. Equal investments in CSR can lead to different financial results, depending on the time, sector or firm. Hull, Tang and Rothenberg (2012) try to this varying effect by explain that the CSR-CSP relationship is metafactorial, making the effect of CSR of CSP hardly predictable. The authors claim that for example, not the amount of CSR activities, but rather the strategy behind it improves the CFP. It is also proposed that CSR is just an indicator of innovation and research investments, which is the true underlying driver of financial performance (Hull & Rothenberg, 2008). Brammer and Millington (2008) even claim that firms with extreme CSR performances (high and low) can see an improved financial performance, indicating that the relative CSR performance with other firms is a considerable factor for CFP.

The empirical studies that are described above question a direct positive relationship between CSR activities and CFP. However, Yoon, Gurhan-Canli and Norbert Schwarz (2006) stated that positive results from CSR-activities depend on the reputation of the firm and the activity itself. Firms with a bad reputation are more sensitive to marketing blowbacks, as is

(6)

5

shown in the research. When the sincerity is questioned, the attitude toward the firm might drop. In this case, the CSR-performance doesn’t add any value to the company but can even subtract value.

It is interesting how the conclusion from Yoon et al. contradicts with the conclusion of Henisz et al. (2014). Also, previous described research has also indicated a small consensus about the general effect of CSR on CFP. However, beside as a direct CFP driver, CSR is also proposed as a “financial damage controller”. Godfrey, Merril and Hansen (2008) concluded that CSR might create a “reservoir of goodwill” during times of crisis with stakeholder relations. It is notable that this empirical research about the “goodwill concept” took place in the short period before the financial crisis started. According to the three researchers, shareholder value decreases less in the context of a negative event if the firm engages in CSR activities compared to firms that have no such activities. Also, CSR activities that are aimed at secondary stakeholders (parties that has high interests in firms, but cannot enforce any claim or decision) provides an ‘insurance-like’ benefit.

Despite the article from Godfrey et al. (2008), firms still reacts differently after the crisis started. A study in 2011 conducted by Giannarakis concludes that a lot of firms have raised their CSR-performance crisis in 2009 to re-establish the trust between companies and their stakeholders, sustaining brand name and regaining customer trust.

However, according to a research about the implications of financial distress on CSR-performance (Jacob, 2012), companies with financial problems have made huge expenditure cuts on community involvement programs, such as budgets for labor related CSR-activities.

The research of Jacob and Giannarakis et al. show us that firms react differently during the crisis. However, both studies did not focus on the financial sector, which is considered the guilty party in the crisis, damaging their reputation (Van Engelen, 2011). Secondly, the companies in the sample group didn’t have a mentionable bad reputation, while the banking sector does have. In order to fill this academic gap, I try to investigate how the financial crisis has affected the CSR activities of financial firms.

(7)

6

Theoretical framework

To gain a deeper insight in this case, several theoretical frameworks are presented.

Perrini, Russo, Tencati et al. (2012) tried to provide a guide, “useful for understanding

better the mechanisms by which certain activities may translate into improvements of firm performance, levering on specific drivers of market and operational results”. In this

framework, they propose 5 different efforts that each have their own impact or drive on the CSP performance. Each effort and the effect are briefly described with several examples. More examples can be found in model 1 below.

The authors begin with the Internal Organization driver, which consists of activities that improves the job circumstances. By e.g.: putting effort in the safety or job environment, and design employees are proved to be more loyal and motivated. Firms that inspire employees’ trust and cooperation would outperform competing firms.

Firstly, the CSR-Related Customer Drivers are described. These programs focus mainly on the needs and wants of the consumer, such as improving product safety or providing extra beneficial services for them. Perrini et al. propose this drives have a positive impact on both consumer attitudes toward the firm and the perceived quality of the product, leading to a long-lasting firm-customer-relationship. Market studies conclude that intangible assets, such as trust and market reciprocity are affected by this driver. Finally, CSR practices inspire firms to analyze, understand and communicate the customers, gaining useful market information and fundaments for innovative products.

Secondly, the CSR-Related Supply Chain Drivers are presented. CSR activities that improve processes within the production and distribution processes will result in a long-term buyer-supplier-relationship. This makes knowledge exchange easier, improves coordination and increases the innovation potential and the quality of the products. Several examples of this driver are more ethical operations in the factories in less developed countries.

Third are the Society drivers. The community remains an important stakeholder. By collaborating with the (local) community, firms can improve their social capital, an intangible resource that describes the relationship between firm and the local actors. Several ways to improve this social capital is by increasing transparency and good citizenships. This will

(8)

7

stimulate employee loyalty and a service-oriented mentality within the company, leading to lower human resource costs and higher quality of the services.

The fourth driver consists of CSR-related Natural environment factors. By focusing on pollution or waste limitation, firms can identify and cooperate with new emerging green markets and markets. Reducing waste can lead to cost reductions by being thriftier of consuming materials and energy. Also, paying regard to environmental issues can help prevent big governmental fines.

CSR-Related Governance drivers are the fifth domain that is described. These activities consist of e.g. diversity management and the democracy within a company. CSR engagement can be interpreted as attempts to meet stakeholder expectations, which lowers the threshold for capital markets. Some forms of CSR, such as governal codes and guidelines, stimulate managers and board members to act in the best interests of share- and stakeholders. As said in the descriptions of the drivers, CSR can contribute to CFP in both revenue-related and cost-related outcomes. This framework is summarized in the model below.

(9)

8

Another theoretical framework that is relevant for this thesis is written by Janssen, Sen and Battacharya (2015). The theoretical framework is shown in model 2 and 3 below. According to the authors, many firms “believe that CSR acts as a reservoir of goodwill, insulating the

firm from the negative impacts of a crisis”. However, the writers state that this “believe” is

too simplistic, and the reality is much more complex. They propose a contingent framework which describes 4 roles CSR plays in times of (reputational) crisis.

First of all, CSR increases the attention of stakeholders on company activities. People pay more attention to companies that conduct CSR activities than companies without a notable CSR visibility.

Secondly, CSR affects blame attributions, which refers to the “causal reasoning stakeholders engage” when analyzing certain negative or out of the ordinary events or occurrences. Due to the (almost) unpredictable character of corporate crises, it is likely to garner extensive attributional thinking. Stakeholders spontaneously try to find out who is to blame during negative events. The more they hold the firm responsible for the crisis, the more their attitude or reactions toward the firm tend to be negative. However, CSR can either moderate or even eliminate this effect.

Stakeholders may either see firms who conduct CSR as either extrinsic (in which firms act out of self-interests to increase its profit) or intrinsic (in which firms genuinely act out of concern for societal issues) (Janssen & Sen, 2015). This assessment of stakeholders depend on different factors, such as the perceived congruence of the firm’s CSR activities with its core business, the method of communicating CSR to the public, and the extent to which stakeholders see the effectiveness of CSR-activities on solving or improving societal issues. Stakeholders have (theoretically) a more positive inference and attitude towards intrinsic firms. Firms with (in the perception of stakeholders) extrinsic motives create more skepticism, which might lead to a (more) negative image, making the company more vulnerable for reputational crises.

The third role CSR plays in times of crisis is by raising stakeholder expectations. In general, CSR activities reveal the true ethical identity of the company, which can be described as more fundamental, enduring and distinctive by nature than other corporate information. According to a model from Reeder and Brewer (1979), a hierarchal restrictive scheme influences attributions pertaining to moral traits. According to their theory, “people intuitively

(10)

9

(...) is expected to exhibit both dishonest and honest behaviors, depending on the situation”.

Stakeholders thus might hold CSR companies to higher standards than companies who doesn’t actively communicate or conduct CSR activities.

The last role CSR plays in the model of Janssen, Sen and Bhattacharya (2015) is that it changes stakeholders’ evaluations of crisis situations. According to the writers, managers of researchers have to make three assessments before the outcome of this last role can be predicted.

The first assessment is the type of crisis that the company faces. An important criterion is the extent to which the company can be hold responsible or accountable for the crisis. The overall logic is that stakeholders put less blame towards a responsible company when it conducts CSR activities. However, this logic is only applicable when the evidence suggests a mild to low responsibility towards the company.

The second assessment is the domain in which the crisis occurs. The writers distinguish two types of corporate crises: matters of product performance or matters involving ethical matters. Failures of mistakes with product performances have a smaller impact on stakeholder attitudes than ethical or social blunders.

Third assessment is the severity of the crisis, which can be defined (or quantified) as financial, human and environmental damage. The perceived negativity of the situation is directly related to this severity.

The final (fourth) assessment to be made is the extent of the stakeholder identification with the company. This refers to the perceived amount of mutual facets between the company and the stakeholder. Research show that individuals who has a psychological attachment to the firm have more trust in its intentions, engage more supportive behavior, and tend to be more resilient to negative information. However, this resilience has its limits and can be negatively influenced by other factors such as severity and crisis domain.

The first three assessments are easily applicable in the financial crisis. First of all, mass media, politics and academic world hold banks primarily responsible for the financial crisis in 2008. Secondly, the crisis involves ethical matters, as the high risk financial operations led to the economic meltdown. This global scale economic adversity led to national austerity measures and high unemployment rates. Thus, we can conclude that the severity of the crisis is at least substantial.

(11)

10

The roles CSR plays in crisis, together with the crisis characteristics, can lead to two different results. An assimilation (or insulation effect) is the most favorable outcome, containing both the relational damage with the stakeholders and limiting any financial harm following a crisis. For this to happen, the firm has favorable scores on the assessments.

Additionally, CSR can also have a blowback effect, creating a contrast (amplification) effect, which ironically damages the relation with stakeholders (and CFP) even more, compared to firms without CSR visibility. For this outcome, firms must have unfavorable scores in the assessments.

According to this model, banks should expect a contrasting or amplifying effect from CSR, since these scored unfavorable on the assessments concerning crisis characteristics. This effect should damage the relation with stakeholders and thus decrease the CFP of the financial sector.

Model 3: Framework of the outcomes of the crisis characteristics on the roll of CSR, as proposed by Janssen, Sen & Battaraya (2015)

(12)

11

Hypotheses

There has been very limited research conducted about the impact of the financial crisis on the relation between CSR performance and CFP in the financial sector. Previous literature primarily focused on the effect of CSR and crises on individual firms. However, after the crisis the general attitude towards the entire financial sector deteriorated, and the culture and ethics within the entire sector became revealed (Van Engelen, 2011). The “cowboy style” of investing and high risk financial operations for financial gain of the banks and bankers attracted broad media attention and caused economic adversity on global scale. According to the literature, banks may feel forced to conduct more CSR activities (and raise their CSR score) after 2008 in order to restore their image, reputation and stakeholder relation (Giannarakis, 2011). This causes a peak in the CSR increase. The first hypotheses will be formulated as:

H1: There will be a peak in the CSR increase in 2009 in the financial sector.

According to most literature, CSR performance has a positive correlation with CFP. However, most of the research didn’t specifically include the financial sector. For this research, it is important to know whether or not the correlation is positive in general before measuring the moderation. Based on the research of Barnett (2007), this leads to the second hypothesis.

H2: There is a positive correlation between CSR performance in year x and CFP in year x+1 in the financial sector.

Unfortunately for managers, CSR doesn’t always have a favorable outcome during and after crises. As described in the framework of Janssen, Sen and Battacharya (2011) and the empirical study of Yoon et al. (2006), CSR can also have a blowback effect, damaging the stakeholder relationship rather than improving it. According to the framework of Janssen et al. (2011), CSR has a negative amplifying effect when firms are highly responsible for a severe crisis on a fundamental social or ethical matter. Yoon et al. (2006) state that bad reputations in general can have a negative impact on the relation between CSR performance and CFP. This leads to the third hypothesis.

(13)

12

H3: The crisis will have a negative moderating effect on the relationship between CSR performance in year x and CFR in year x+1 in the financial sector.

(14)

13

Research method

In order to test all hypotheses, I will conduct several regression analyses with SPSS, a computer program for statistical data analyses. The input will primarily consist of quantified variables.

The CSR scores will be derived from the Kinder Lydenberg Domini (KLD) database, constructed by Morgan Stanley Capital International. This database contains separate scores about the CSR performance of firms on several factors, such as environmental, community & society, employees & supply chain, customers, governance & ethic (MSCI, 2015). The “CSR-scores” in this database “evaluate companies against peer industries on triple bottom performance” (Turner, 2013).

Unfortunately, this KLD database consists only of US based financial firms. Therefore, only these American firms are included in this research. Secondly, the database doesn’t contain the KLD from all periods. For this reason, only CSR scores in the period 2004- 2012 are used. Not all financial reports from 2014 are published or approved by the time this research is conducted, so the CSR-performance of 2013 and the financial results of 2014 are excluded.

Another problem is the range of the binary scores of these CSR scores. Both strengths and weaknesses have a score of 1 on the criteria, while no notable strengths or weaknesses score a 0 (RiskMetricGroup, 2010). This means the average score will not have a clear indication whether the CSR activities of a company are evaluated positively of negatively. As solution, the scores of CSR weaknesses are multiplied by -1, creating a new range for negative factors between -1 and 0. This way, firms with a negative CSR score can be distinguished from firms with a good CSR sheet.

Based on the researches of McWillliams and Siegel (2000) and Lo and Sheu (2007), the Market value and Risk of investments of financial firms are taken into account as a control variable to analyze any possible covariance and to improve the validation of the correlation. These variables will be derived from Compustat.

(15)

14

The Corporate Financial Performance will be measured as the Return of Assets (RoA), as this is an often used ratio for shareholders to assess the financial performance of firms (Marshal, McManus & Viele, 2011). The data will be derived from Compustat. Another possible indicator for CFP is Return on Equity (RoE). However, RoA has less data gaps in Compustat compared to RoE. Because the effect of CSR is measured on the following year’s CFP, only the financial reports from 2005-2013 are to be used. As stated earlier, not all financial reports from 2014 are published or approved at the time this research was conducted, so the CSR-performance of 2013 and the financial results of 2014 are excluded. Before the hypotheses are tested, several descriptive and normality analyses are done.

The research method for all three hypotheses will consist of a linear regression analysis. For the first and latter hypothesis, a moderator will be added.

For hypotheses 1, the moderating effect of the crisis on CSR changes is measured. I created a new variable “CSR-change”, which was calculated by subtracting the CSR in year

x+1 minus CSR in year x. In order to measure a change in CSR score, this variable will be the

input as a dependent variable, with “CSR year” as independent. The regression will show us if and how the CSR strategy and activities have changed over time. In order to focus on a possible CSR-change in or after the financial crisis, an ordinal dummy variable “dummy-2009” is created, with 0=period before and after 2009, and 1=period 2009.

With the multiplication of the standardized Z-score of this dummy variable and the standardized Z-score of CSR year, I created a new moderator variable, which will be used as an independent variable for the linear regression analysis. This moderator is the product or multiplication of both standardized z-scores of “dummy-2009” and the CSR-years.

For the second hypothesis, the relation between CSR and CFP in general is researched. The CSR performance of financial firms shall be used as an independent variable in a regression analysis, whereas the Return on Assets will be the dependent variable. The above mentioned control variables (firm) will also be included in the input of SPSS. In the dataset, each CSR score of a certain year and firm was paired with each CFP score of that firm in the next year.

For the third hypothesis, the moderating effect of the crisis on the CSR-CFP relation is tested. For this hypothesis, a similar linear regression analyses is conducted as the testing for hypothesis 2. However, two new variables are created to measure the moderating effect of the

(16)

15

crisis. First, another binary dummy variable is created, with 0=before the crisis and 1=after the crisis. Secondly, the moderator variable is created by multiplying the standardized Z-score with the standardized Z-score of the KLD-scores.

(17)

16

Results

In this paragraph, the results are presented of this research. For the statistical analyses and data processing, a combination was used of Microsoft Excel and IBM SPSS. First of all, some unexpected or unpredictable problems around the data collection and its solutions are explained, such as missing data. Secondly, the quantitative values of the variables are

introduced and analyzed. Several descriptive statistics, normality tests and

correlation/collinearity test are presented. And lastly, the results of the three linear regression analyses are explained. To improve the legibility of this paragraph, all values in text are rounded up to 3 decimals. The original values can be found in the tables in each paragraph. Due to some technical limitations of SPSS, the names variables in the tables differ from the names in the text. CSR score is called “CSR-averageSPSS”, Risk is called “RISK percentual”

and the dummy variable for the crisis period is called

“DUMMYVARIABLEBEFORAFTER” Missing data

Although undesirable, it is not very uncommon for statistical researches to have a certain amount of missing data (Field, 2009). However, these gaps can have an effect on the analyses and conclusions. Therefore, researchers often adjust their dataset to solve the missing data. There are several techniques to deal with this problem. The first method is the deletion of the entire individual case. Other methods try to estimate the missing value by using the data of similar cases (e.g. hot deck simulation).

To solve the problem with the missing data, I choose to delete all cases with missing data. It was not possible to estimate the data gaps due the lack of similar data. One example was the case of Citigroup, wherein all market values were absent.

Descriptive statistics

In this paragraph, all descriptive values are presented. For this thesis, an amount of 58 financial firms in the United States were researched. Five variables have been taken into this thesis. The table containing the values is shown below in table 1.

The average CSR score of the financial firms is around 1,803, with an Standard Deviation of 4,157 (Min.=-6,557; Max.=23,636). During the analyses, a technical problem

(18)

17

occurred. After the data import from Excel, SPSS didn’t recognize the exponent symbol “E” as a mathematical symbol. This turned the CSR variable into a string variable. By multiplying this number with 100 in Excel, the E symbol disappeared without changing the numerical value much. On a scale of scale (-100,100), the average score of 1,803 does not seem high. This insinuates that banks don’t conduct CSR activities much in general. However, the relatively high standard deviation indicates that there are huge differences between banks. This possibly compensated for each other in the calculation of the mean.

The second variable (CSR change) represented the change of CSR scores from year x+1 and year x. This variable had a mean of -0,02, with a standard deviation of 3,58 (Min.=18,31;Max.-18,59). I purposely did not use the percentual change, but the absolute change due to the “division by zero problem”. Some firms have their CSR score increased or decreased from 0, leading to an invalid percentual change. With absolute values, this problem does not occur. It is notable that this variable too has high deviations with a low mean.

Both CSR variables are used as independent variables for different hypotheses. The main dependent variable (Return on Assets) indicates the corporate financial performance. The average CFP of all firms during the years was 0,007, ranging from -0,06 to 0,062 (Std. Dev.=0,01). The return on assets was calculated by the fiscal net income with the book value of assets.

Based on the literature, Risk of investments and the firms’ size were used as control variables. The Risk has a percentual mean of 11,92%, ranging from 6,75% and 23,20% (Std. Dev=0,028). The size of the firm was determined by the total market value of the firm, with a mean of 12732,174. Notable are the huge values of the standard deviation (35130,682), minimum (54,615) and maximum (238675,2).

Another possible explanation for the high standard deviations of Return on Assets and Market Value is that the data contained several years, including the instable crisis years, wherein a lot of fluctuations occurred in the financial performances of the financial sector.

The last factor to be noted is the difference in sample sizes. All variables had a N-value of 487, while CSR change has a N N-value of 429. This difference is caused by the calculation method of the variable itself. CSR change is calculated by subtracting CSR score in year x+1 with CSR score in year x. However, in the first year of each bank, there is no previous value to subtract it with, leaving a blank in the dataset. Not coincidently, the

(19)

18

difference between the N-value of CSR change and the other variables is the amount of banks that is analyzed: 58.

Statistics

CSR-averageSPSS

CSR change Return on Assets &

RISK percentual Market Value - Total - Fiscal N Valid 487 429 487 487 487 Missing 0 58 0 0 0 Mean 1,803243 -,020313 ,007028 ,119201 12732,173547 Std. Deviation 4,1574607 3,5789450 ,0096655 ,0283722 35130,6822677 Minimum -6,5574 -18,3060 -,0600 ,0675 54,6150 Maximum 23,6364 18,5924 ,0619 ,2320 238675,2002 Table 1: descriptive statistics

Correlations

Correlations are a statistical phenomenon which measures how much 2 variables depend on each other. To test the variables for correlations, a Pearson Bivariate Correlation test is conducted. The table can be found below in below in table 2. The correlation matrix shows us there are several significant correlations between variables. The CSR score of financial firms seem to correlate with almost all variables, except with the dummy variables and Return on Assets. First of all, CSR score correlates with CSR change (Corr.= 0,416; Sig.=0,00), which doesn’t seem much of a surprise since the CSR change is derived from the CSR score. Also, CSR score correlates with Risk (Corr.=0,134; sig.0,003) and Market Value (Cor.=0,248; sig. 0,000).

The CSR change variable correlates with no other variable besides CSR score. Return on assets has no significant correlation besides the dummy variables. The Risk variable correlates with Market Value and the dummy variables. This doesn’t seem much surprising as big firms have more capital and can afford more risky investments. (Berk & Demarzo, 2011). And lastly, the Market value only correlates with the dummy variables.

The correlation matrix shows a limited amount of significant correlations between values. However, the variables that do correlate significantly do not have a notable large correlation coefficient. Based on this information, no variables need to be excluded or adapted.

(20)

19

Table 2: Correlation Matrix

Correlations

CSR-averageSPSS CSR change Return on Assets &

RISK percentual Market Value - Total - Fiscal

dumyvariable 2009 DUMMYVARIABL EBEFOREAFTER

CSR-averageSPSS Pearson Correlation 1

Sig. (2-tailed) N

CSR change Pearson Correlation -,416** 1 Sig. (2-tailed) ,000

N 429 429

Return on Assets & Pearson Correlation ,051 -,003 1 Sig. (2-tailed) ,257 ,946

N 487 429 487

RISK percentual Pearson Correlation ,134** -,030 -,008 1 Sig. (2-tailed) ,003 ,533 ,863

N 487 429 487 487

Market Value - Total - Fiscal Pearson Correlation ,284** -,087 ,076 -,150** 1 Sig. (2-tailed) ,000 ,072 ,095 ,001

N 487 429 487 487 487

dumyvariable 2009 Pearson Correlation ,006 ,009 -,117** ,174** ,000 1

Sig. (2-tailed) ,889 ,853 ,010 ,000 ,997 N 487 429 487 487 487 487 DUMMYVARIABLEBEFOREAF TER Pearson Correlation ,117** -,046 -,177** ,464** ,027 ,328** 1 Sig. (2-tailed) ,010 ,338 ,000 ,000 ,550 ,000 N 486 429 486 486 486 486

(21)

20

Multicollinearity

Before a linear regression analysis can be conducted, the data has to be tested for multicollinearity. This statistical problem explains the correlation of predictor variables in a multiple regression analysis. In order to measure this collinearity, the Variance Inflation Factor (VIF) test is done on all variables. The relevant tables 17-23 are shown in the appendices. When the value of VIF is below 10, it can be assumed that there is no multicollinearity (Myer, 1990). The VIF values vary between 1,011 and 1,392. This means all VIF-values are below the limit of 10 that is proposed by Myer. Based on the assumption of Myer and the VIF-values, I can assume there is no multicollinearity between the values.

Normality tests

Before conducting the linear regression analysis, the variables need to be tested for normality problems, such as kurtosis and skewness issues. Any possible skewness in the data is detected through e.g. the Kolmogorov-Smirnov test, descriptive table and a visual examination on the histograms and Q-Q plots. These graphs can be found in graph 1-20 in the appendices.

According to the data analyses, all five variables do have skewness and kurtosis issues. The CSR score has a positives skewness of 1,541 with a kurtosis of 4,776. This (slight) skewness is also visible in the histogram and Q-Q plots and boxplot.

The CSR change has a negative skewness of -0,547 with a kurtosis of 5,416. This skewness is notable and visible in the graph 5 in the appendices.

Return of assets did have a more negative skewness with a value of -2,231. However, most notable is the high kurtosis value of 14,030.

The skewness of Risk had a value of 0,881, while the kurtosis of Risk has a value of 1,103. Market value has the most extreme values, with a skewness of 4,197 and a kurtosis of 17,790. These normality issues are visible in graph 17-20.

(22)

21 Statistics CSR-averageSPSS CSR change Return on Assets & RISK percentual Market Value - Total - Fiscal N Valid 487 429 487 487 487 Missing 0 58 0 0 0 Skewness 1,541 -,547 -2,231 ,881 4,197 Std. Error of Skewness ,111 ,118 ,111 ,111 ,111 Kurtosis 4,776 5,416 14,030 1,103 17,790 Std. Error of Kurtosis ,221 ,235 ,221 ,221 ,221 Table 3

The kolmogorov-smirnov and the Shapiro-Wilk tests measure whether the distribution of the sample deviates from a normal distribution (Field, 2009). When the significance of the test is below 0,05, then the normality of the sample size is not significant, indicating a possible skewness. The results of the Kolmogorov-Smirnov and Shaprio-Wilk shows a significance value of 0,000 (df.=429) for all five variables. This means the data is probably skewed.

Tests of Normality

Kolmogorov-Smirnova Shapiro-Wilk Statistic df Sig. Statistic df Sig. CSR-averageSPSS ,144 429 ,000 ,888 429 ,000 CSR change ,211 429 ,000 ,880 429 ,000 Return on Assets & ,192 429 ,000 ,731 429 ,000 RISK percentual ,069 429 ,000 ,963 429 ,000 Market Value - Total - Fiscal ,360 429 ,000 ,363 429 ,000 a. Lilliefors Significance Correction

Table 4

A fallacy of the Kolmogorov-Smirnov and Shapiro-Wilk test is that both methods have low power when the sample size is small (Field, 2009). So even though the significance is low (which indicates skewness), this detection is weak considering the sample size contains only 58 banks. However, even though the power of the tests is weak, it is still an indication of skewness. Secondly, these normality issues are also visually visible in the graphs and plots, so I still assume the data is skewed and has kurtosis problems.

(23)

22

Transformation and removal of outliers

Based on the indicated skewness, it can be considered to transform the data. Four conventional methods are mostly used when normality problems are encountered (Field, 2009).

The first method is the “log transformation”, wherein the logarithm is taken over all data. The second method is the “square root transformation”, which is quite similar to the log transformation, except the square root is taken instead of the logarithm. The third option is the reciprocal method, in which we divide 1 by each single value. The last method is the reverse score transformation, in which each score is subtracted by the highest score obtained. The first three methods are suitable for positive skewed data, while the latter is suitable for negative skews.

However, I have decided not to transform the data. First of all, the academic world disagrees whether or not transforming data would be a positive contribution to the research. For example, Glass, Peckham and Sanders (1971) claim: “the payoff of normalizing

transformations in terms of more valid probability statements is low, and they are seldom considered to be worth the effort”.

The second reason to reject any transformation is the lack of appropriate transformation methods. As described in the previous paragraph, the variables consist of 3 positive and 2 negative skewed variables. Unfortunately, none of the transformation methods are appropriate for both categories. Since all variables have to be treated equally and consistent, none of the transformation methods can be chosen.

Thirdly, the data cause mathematical problems in most of the transformation methods. The variables CSR score, CSR change and RoA contain negative values. It is not possible to derive the logarithm or square root from negative values. The dataset also contains “zeroes”, which causes a “division by zero problem” when the reciprocal method is chosen. Even though the values will not cause mathematical problems when used for the reverse score transformation, this method is still not suitable because it is designed for negative skews, while the majority of variables is positively skewed.

Because the data cannot be transformed into a dataset with a significant normality, I continue to work with the skewed data for the regression analyses in the next paragraphs. The implications of this issue will be addressed in the discussion part of this thesis.

(24)

23

Another consideration to be made is whether or not to delete the outliers in the sample. Mostly, outliers are removed or changed because it is assumed these values are an error in the data set (Field, 2009). However, the data in this research are mostly American fiscal data that is checked and validated by several independent institutions, such as accountancy firms, IRS and other financial departments of the United States government. Based on the controlling mechanism, we can assume all values in the dataset are solid, which gives no valid argument to remove the outliers.

(25)

24

Testing the hypotheses.

After the testing for multicollinearity and skewness, the hypotheses are to be tested. The three hypotheses were tested with linear regression models. Based on the literature, we proposed hypothesis 1: There will be a peak in the CSR increase in 2009 in the financial

sector. Testing this hypothesis should show whether or not the banks reacted to the financial

crisis with an increase of CSR activities.

For this analysis, a moderator variable was created by multiplying the standardized scores of CSR year and the dummy variable “dummy-2009”. This new variable is named “moderator_2009”. A linear regression model was created where model 1 contained the control variables Risk and Market value and model 2 contained CSR year and the moderator variable. The CSR change is the dependent variable.

According to this linear regression model (N=429) shown in table 5-8, the Adjusted R2

of model 1 has value of 0,005 (F=2,030; sig.=0,133), while the R2 of model 2 had the same

value of 0,005 (F=1,485, sig.=0,206). This means model 2 has no contribution to the explanatory factor of model 1. None of the values in both models had a beta with a corresponding significance value below 0,05, meaning hypothesis 1 is rejected.

Descriptive Statistics

Mean Std. Deviation N CSR change -,020313 3,5789450 429 RISK percentual ,121157 ,0278899 429 Market Value - Total - Fiscal 12821,144153 35515,7148289 429

CSR Year 2008,44 2,307 429

moderator2009 ,0977 ,51765 429 Table 5: Descriptive statistics of the linear regression model

Model Summary Mod el R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change df1 df2 Sig. F Change 1 ,097a ,009 ,005 3,5703624 ,009 2,030 2 426 ,133 2 ,118b ,014 ,005 3,5708598 ,004 ,941 2 424 ,391 a. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual

b. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual, moderator2009, CSR Year Table 6: Model summary of the linear regression model

(26)

25

ANOVAa

Model Sum of Squares df Mean Square F Sig.

1 Regression 51,757 2 25,879 2,030 ,133b Residual 5430,430 426 12,747 Total 5482,187 428 2 Regression 75,746 4 18,937 1,485 ,206c Residual 5406,441 424 12,751 Total 5482,187 428 a. Dependent Variable: CSR change

b. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual

c. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual, moderator2009, CSR Year Table 7: ANOVA of the linear regression model

Coefficientsa

Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) ,779 ,788 ,989 ,323 RISK percentual -5,606 6,254 -,044 -,896 ,371 Market Value - Total –

Fiscal -9,407E-006 ,000 -,093 -1,915 ,056

2

(Constant) 179,999 197,022 ,914 ,361 RISK percentual -2,081 6,980 -,016 -,298 ,766 Market Value - Total –

Fiscal -8,840E-006 ,000 -,088 -1,792 ,074 CSR Year -,089 ,098 -,058 -,910 ,363 moderator2009 ,149 ,399 ,021 ,372 ,710 a. Dependent Variable: CSR change

(27)

26

Hypothesis 2 There is a positive correlation between CSR pe rformance in year x and CFP in year x+1 in the financial sector

For this hypothesis, a linear regression model was created with CSR score in year x as dependent variable and RoA in year x+1. This hypothesis determines whether or not the overall CSR assumptions in the literature apply to the financial sector.

A linear regression model (N=487) was conducted. Model 1 remained unchanged and still contained the control variables. Model 2 now contains the CSR scores. The predictor variable is the Return on Assets.

The adjusted R2 is 0,2% with an F-value of 1,404 and a significance of 0,247. By adding the predictor variable CSR score, the adjusted R2 dropped to 0,1%. The beta of CSR score has a negligible value of almost 0, with a t-value of 0,685 and a significance of 0,494. The lack of a notable beta and the high significance both lead to the rejection of hypothesis 2. The relevant tables 9-12 are shown below.

Mean Std. Deviation N

Return on Assets & ,007028 ,0096655 487

RISK percentual ,119201 ,0283722 487

Market Value - Total - Fiscal 12732,173547 35130,6822677 487 CSR-averageSPSS 1,803243 4,1574607 487 Table 9: Descriptive statistics of the linear regression model

Model Summary Mod el R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change df1 df2 Sig. F Change 1 ,076a ,006 ,002 ,0096575 ,006 1,404 2 484 ,247 2 ,082b ,007 ,001 ,0096628 ,001 ,469 1 483 ,494 a. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual

b. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual, CSR-averageSPSS Table 10: Model Summary of the linear regression model

(28)

27

ANOVAa

1 Regression ,000 2 ,000 1,404 ,247b Residual ,045 484 ,000 Total ,045 486 2 Regression ,000 3 ,000 1,091 ,353c Residual ,045 483 ,000 Total ,045 486

a. Dependent Variable: Return on Assets;

c. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual, CSR-averageSPSS Table 11: ANOVA of the linear regression model

Coefficientsa

Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) ,007 ,002 3,404 ,001 RISK percentual ,001 ,016 ,004 ,080 ,937 Market Value - Total - Fiscal 2,102E-008 ,000 ,076 1,666 ,096

2

(Constant) ,007 ,002 3,455 ,001

RISK percentual -,001 ,016 -,002 -,049 ,961 Market Value - Total - Fiscal 1,820E-008 ,000 ,066 1,371 ,171 CSR-averageSPSS 7,664E-005 ,000 ,033 ,685 ,494 a. Dependent Variable: Return on Assets &

(29)

28

Hypothesis 3 The crisis will have a negative moder ation of the relationship between CSR performance in year x and CFR in year x+1 in the financial sector The third test is to analyze whether or not the crisis had a moderating effect on the CSR-CFP relationship. For this regression model, a moderator variable was created by multiplying the standardized scores of CSR score and “dummy variable crisis”. This new variable was added in model 2 with the CSR score variable. Model 1 remained unchanged with the control

variables. Surprisingly, after adding the moderator, the adjusted R2 changed to -0,001, with a

F-value of 0,863 and a significance of 0,468. The beta of the moderator is zero, with a t-value of -0,431 and a significance of 0,667. Due to the low R2, the low beta and the high significance, we reject hypothesis 3. Table 13-16 below show the regression output.

Mean Std. Deviation N

Return on Assets & ,007028 ,0096655 487

RISK percentual ,119201 ,0283722 487

Market Value - Total - Fiscal 12732,173547 35130,6822677 487 CSR-averageSPSS 1,803243 4,1574607 487

moderator_crisis ,1143 ,96063 487

Table 13: Descriptive statistics of the linear regression model

Model Summary Mod el R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change df1 df2 Sig. F Change 1 ,076a ,006 ,002 ,0096575 ,006 1,404 2 484 ,247 2 ,084b _,007 _-,001 _,0096710 _,001 _,327 ₂ ₄₈₂ _,721

a. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual

b. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual, moderator_crisis, CSR-averageSPSS Table 14 Model Summary of the linear regression model

(30)

29

ANOVAa

1 Regression ,000 2 ,000 1,404 ,247b Residual ,045 484 ,000 Total ,045 486 2 Regression ,000 4 ,000 ,863 ,486c Residual ,045 482 ,000 Total ,045 486

a. Dependent Variable: Return on Assets &

c. Predictors: (Constant), Market Value - Total - Fiscal, RISK percentual, moderator_crisis, CSR-averageSPSS

Table 15: ANOVA of the linear regression model

Coefficientsa

Model Unstandardized Coefficients Standardized Coefficients t Sig. B Std. Error Beta 1 (Constant) ,007 ,002 3,404 ,001 RISK percentual ,001 ,016 ,004 ,080 ,937 Market Value - Total –

Fiscal 2,102E-008 ,000 ,076 1,666 ,096

2

(Constant) ,007 ,002 3,465 ,001

RISK percentual -,001 ,016 -,003 -,070 ,945 Market Value - Total –

Fiscal 1,818E-008 ,000 ,066 1,368 ,172 CSR-averageSPSS 9,585E-005 ,000 ,041 ,795 ,427 moderator_crisis ,000 ,000 -,021 -,431 ,667 a. Dependent Variable: Return on Assets &

(31)

30

Conclusion

For decades, the overall logic was to encourage firms to conduct CSR activities, as this would improve the financial performance (McWilliams & Siegel, 2000). By conducting CSR activities, the relation with stakeholders would improve, making it easier to access resources, such as knowledge and capital.

However, there is no broad consensus on this positive relationship, as there is limited empirical proof (Barnett, 2007). Others claim that CSR doesn’t directly improve CFP, but rather plays a role as indicator of other CFP drivers (Hull & Rothenberg, 2008; Brammer & Millington, 2008). Yoon, Gürhan-Canli and Schwarz (2006) even concluded that CSR can have a negative effect when the firm has a bad reputation.

Another possible (positive) effect is that CSR can reduce damage during times of reputational crisis (Janssen & Sen 2015). CSR activities can create a certain reservoir or buffer of goodwill when the good relationships with stakeholders are at stake. CSR can also act as a way to repair the damaged relation, as Giannarakis (2011) proposed. However, according to Jacob (2012), a huge group of firms have cut CSR budgets due to financial distress. Most of the sample in the empirical studies found does not specifically focus on the financial sector that created the global financial crisis in 2008 (Van Engelen, 2011).It is not clear how financial firms reacted to or are affected by the crisis they started, and whether they benefited from the proposed reservoir of goodwill. In this thesis, I try to fill this knowledge gap, with the following main research question: How did the financial crisis affect the

correlation between CSR performance and corporate financial performance in the financial sector?

For this research, two theoretical frameworks were described. The first framework is written by Perrini, Russo, Tencati et al. (2012). They explained 6 types of CSR target efforts that contribute to CFP: internal organizational, customer, supply chain, society, environmental and governal drivers. These efforts reduce costs and raise revenue by e.g. improving stakeholder loyalty, lowering threshold for capital, lowering material consumption and improving product quality and innovation.

The second theoretical framework is proposed by Janssen, Sen and Battaraya (2015). This theory explains how CSR affects the stakeholder relationships during times of crisis.

(32)

31

According to the article, CSR plays 3 general roles during stable times. First, it increases the attention of stakeholders on company activities. Secondly, it influences the blame attributions of stakeholders. Thirdly, it raises stakeholder expectations of the firm. The outcome of CSR activities depends on several assessments that are to be made: the type of crisis, the domain in which the crisis occurs, the severity and how much the stakeholders identify themselves with the company. When the firm scores favorable on all four assessments, CSR will play a damage insulating role, reducing the relationship damage. However, when the firm scores unfavorable on the assessments, CSR will have an amplifying effect, making the damage even worse.

In this thesis, we tried to understand how the CSR activities and its outcomes were affected by the financial crisis. The CSR scores from the period 2004-2012 and fiscal data from 2005-2013 of 58 American banks were analyzed. For this research, several linear regression models were made. The CSR scores and financial data were derived from the KLD and Compustat database. For all regressions, Risk of investments and Market value were used as control variables.

Hypothesis 1 stated: There will be a peak in the CSR increase in 2009 in the financial

sector. A linear regression model was created with CSR year and a moderator variable as

independent variables, and the change of CSR as dependent. However, no significant results were found, so hypothesis 1 is rejected. This contradicts the article of Giannarakis. However, the lack of significance can be explained by Jacob (2012), who stated that firms made huge budget cuts in their CSR activities.

Hypothesis 2 stated: There is a positive correlation between CSR performance in year

x and CFP in year x+1 in the financial sector. For this test, a linear regression was made with

CSR score year x as independent variable, and CFP in year x+1 as dependent variable. No significant results were found, so hypothesis 2 is rejected. This is surprising, since the proposed theoretical framework proposed a positive relationship. However, as Barnett et al. (2007) already claimed: statistical CSR researches can lead to different financial results, depending on the time, sector or firm.

Hypothesis 3 stated: The crisis will have a negative moderation of the relationship

between CSR performance in year x and CFR in year x+1 in the financial sector. For this test,

(33)

32

significant results were found, so hypothesis 3 is rejected. This contradicts the theoretical framework of Janssen, Sen and Battaraya (2015).

Based on the three rejections, I conclude that the financial crisis did not affect the correlation between the CSR performance and the corporate financial performance in the financial sector.

(34)

33

Discussion

Limitations

Hopefully, the results and conclusions of this thesis can contribute to new managerial and academic insights about the effect of the financial crisis on the relation between Corporate Social Responsibility activities and Corporate Financial Performance of the financial services sector (e.g. banks). However, there are several factors to be considered that limit the reliability or the practical usefulness of this study.

Limited data

First of all, due to the limited availability of data, the sample size has been (arguably) small. The database primarily banks which were enlisted on the New York Stock Exchange. Smaller firms were not included. Secondly, no CSR research has been conducted on banks outside the United States. These data can have a significant contribution to this (or similar) research due to the global character of the crisis. Also, it is to be noted that although KLD did contain CSR scores of most NYSE enlisted banks, not all data were complete.

Location sample

This study is performed on 58 US based banks. The results are therefore partly generalizable to the financial sectors of other firms. This is because of the legislative and cultural differences between the continents. However, according to the literature, the high risk banking culture seemed to be international, making the other financials sectors partly similar.

Method CSR scores

Another limitation of this thesis is the transparency of the database. The KLD only showed the criteria that financial firms are being benchmarked on, but it is not clear how the KLD scores have been calculated. The scoring of different criteria might have been very subjective.

Voluntary base

It can also be questioned whether the CSR criteria that the banks met were the result of voluntary CSR-activities, coincidences or were mandatory by law. The voluntary deliberate factor is crucial for labeling corporate operations as “CSR activities”. It made this limitation

(35)

34

even more complex when realized that states have different laws concerning corporate activities. A CSR activity might be mandatory in one state, but voluntary in another, making CSR scores from different firms difficult to compare.

The weights of CSR scores

In this thesis, all separate CSR scores in the KLD database have been calculated with an equal weight when computing the overall CSR score. However, it can be argued that some CSR criteria should have a higher weighting than other variables. Some might claim that avoiding lucrative but ethically controversial operations (e.g. weapon investments) leading to global concerns or issues might be more “noble” than providing (for example) education for employees about the implications of drug abuse.

Manipulated CSR scores

Unfortunately, CSR scoring can be manipulated by using legal loopholes or spurious corporate structures to hide controversial operations.

Skewed data

As described earlier in this thesis, de data is quite skewed, while the outliers haven’t been deleted, nor have the data been transformed. Using skewed data do cause issues for the interpretations of linear regression analyses, reducing the validity and generalizability of the output and conclusions.

Recommendations for future research

The rejection of hypotheses and the main research question both indicate that it is not feasible to conduct CSR activities in the financial sector. However, as it is stated earlier, future scientific research could conclude both a bigger sample size and another geographical location of the financial firms. As suggested in the limitations paragraph, the sample and skewness are huge limitations for the generalizability of this research. This could possibly be solved with a bigger sample size. Analyzing other banks at other continents might give new useful insights.

Also, a stakeholder analysis was not conducted. However, it is still interesting to take the intentions of stakeholders into account when doing similar research, as these intentions are crucial for the CSR-CFP relation. All previous theoretical frameworks assumed that

(36)

35

stakeholders would react positively on CSR activities, which would result in financial benefits. It is possible that stakeholders don’t have any interest in CSR activities, or even rejects CSR. When this is the case, all theoretical assumptions are to be rejected and the research could take another new angle.

Managerial implications

As this research doesn’t confirm a (positive) relationship between CSR performance and CFP, it can be considered whether or not it is feasible for financial firms to conduct CSR activities. As CSR activities contribute to societal issues, it does not contribute to the financial performance. Marketing departments should also reevaluate the CSR activities in their

(37)

36

Literature

Barnett M. (2009) Stakeholder influence capacity and the varability of financial returns to corporate social responsibility. The academy of management review Vol 32 No 3 (July 2007) pp 794-816 published by academy of management

Berk, J., & DeMarzo, P. (2011). Common versus Independent Risk. In Corporate Finance (2nd ed., p. 308). Harlow, England: Pearson.

Brammer, S. and Millington, A. (2008), Does it pay to be different? An analysis of the relationship between corporate social and financial performance. Strat. Mgmt. J., 29: 1325– 1343. doi: 10.1002/smj.714

Field, Andy. (a) "Exploring Assumptions." Exploring SPSS. 3rd ed. London: Sage, 2009. 144-155. Print.

Giannarakis G. (2011) The effect of financial cris in corporate social responsibility performance International Journal of marketing studies Vol 3 (1) februari 2011

Glass, G. V., Peckham, P.D., & Sanders,. J. R. (1972). Consequentes of failure to meet sasumptions underlying the fixed effects analyses of variance and covariance. Review of

Educational Research, 42(3), 237-288

Godfrey P. Merril C, & Hansen J. (2009) The relationship between corporate social responsibility and shareholder value: an empirical test of the risk management hypothesis. Strategic management journal vol 30 pp 425-445 published by Wiley InterSience.

Henisz, W. J., Dorobantu, S. and Nartey, L. J. (2014), Spinning gold: The financial returns to stakeholder engagement. Strat. Mgmt. J., 35: 1727–1748. doi: 10.1002/smj.2180

Hull, C. E. and Rothenberg, S. (2008), Firm performance: the interactions of corporate social performance with innovation and industry differentiation. Strat. Mgmt. J., 29: 781–789. doi: 10.1002/smj.675

Jacob C (2012) The impact of financial crisis on corporate social responsibility and its implications for reputation risk management. Journal of Management and Sustainability Vol 2 (2) august 2012

(38)

37

Janssen C, Sankar S and Bhattachyrya (2015) Corporate crises in the age of corporate social responsibility. Business Horizons, March-April 2015 Vol 58 (2). Pp 182-192.

Lo, S.F., Sheu, H.J. (2007), Is Corporate Sustainability a Value-Increasing Strategy for

Business?,

Corporate Governance: An International Review, Volume15, pp 345-358

Marshall, D., McManus, W., & Viele, D. (2011). Financial Statement Analysis. In Accounting What the numbers mean (11th ed., p. 433). New York, New York: McGrawHill.

Mcwilliams, A & Siegel, D Corporate Social Responsibility: A Theory of the Firm Perspective The Academy of Management Review, 1 January 2001, Vol.26(1), pp.117-127

McWilliams, A. and Siegel, D. (2000), Corporate Social Responsibility and Financial Performance: Correlation or Misspecification?, Strategic Management Journal, Volume. 21, pp. 603-609

MSCI KLD 400 Social Index. (2015, June 30). Retrieved July 15, 2015, from

https://www.msci.com/www/fact-sheet/msci-kld-400-social-index/07239644

Myers, R. (1990). Classical and modern regression with applications (2nd. Ed.). Bostons, MAL Duxbury

Perrini, F, Russo, A, Tencati, A, Vurro, C (2011) Deconstructing the Relationship Between Corporate Social and Financial Performance; Journal of Business Ethics, Vol.102(1), pp.59-76

Raithel S, Wilczynski P., Schloderer M.P., Schwaiger M., (2010) "The value‐relevance of corporate reputation during the financial crisis", Journal of Product & Brand Management, Vol. 19 Iss: 6, pp.389 – 400

Tang, Z., Hull, C. E. and Rothenberg, S. (2012), How Corporate Social Responsibility Engagement Strategy Moderates the CSR–Financial Performance Relationship. Journal of Management Studies, 49: 1274–1303. doi: 10.1111/j.1467-6486.2012.01068.x

Turner, L. (2013, July 15). KLD Scores. Retrieved June 1, 2015, from http://nbs.net/knowledge/kld-scores/

Roberts, R., & Schifferes, S. (2015). The Media and Financial Crises : Comparative and Historical Perspectives. Abingdon, Oxon: Routledge.

(39)

38

Van Engelen, E. (2011). After the great Complacence (1st ed.). Oxford: Oxford University Press.

Yoon Y, Gürhan-Canli Z & Schwarz N (2006)The effect of corporate Social responsibility (CSR) activities on companies with bad reputations Journal of Consumer Psychology, 2006 vol. 16 (4) pp 377-390 published by ScieVerse ScienceDirect Journals

(40)

39

Appendix

VIF tables

Coefficientsa

Model Collinearity Statistics Tolerance VIF

1

CSR change ,989 1,011 Return on Assets & ,971 1,030 RISK percentual ,759 1,318 Market Value - Total - Fiscal ,953 1,049 dumyvariable 2009 ,903 1,108 DUMMYVARIABLEBEFOR

EAFTER ,718 1,392

a. Dependent Variable: CSR-averageSPSS Table 17: VIF analysis

Coefficientsa

1

Return on Assets & ,970 1,031 RISK percentual ,742 1,348 Market Value - Total - Fiscal ,874 1,145 dumyvariable 2009 ,902 1,109 DUMMYVARIABLEBEFOR

EAFTER

,718 1,392

CSR-averageSPSS ,887 1,127 a. Dependent Variable: CSR change

Table 18: VIF analysis

Coefficientsa

1

RISK percentual ,744 1,344 Market Value - Total - Fiscal ,879 1,138 dumyvariable 2009 ,906 1,104 DUMMYVARIABLEBEFOR

EAFTER ,727 1,376

CSR-averageSPSS ,740 1,351 CSR change ,825 1,213

(41)

40

a. Dependent Variable: Return on Assets & Table 19: VIF analysis

Coefficientsa

1

Market Value - Total - Fiscal ,914 1,094 dumyvariable 2009 ,903 1,107 DUMMYVARIABLEBEFOR

EAFTER ,890 1,123

CSR-averageSPSS ,758 1,319 CSR change ,826 1,211 Return on Assets & ,974 1,026 a. Dependent Variable: RISK percentual

Coefficientsa

Model Collinearity Statistics Tolerance VIF 1 dumyvariable 2009 ,902 1,109 DUMMYVARIABLEBEFOR EAFTER ,723 1,382 CSR-averageSPSS ,808 1,237 CSR change ,826 1,211 Return on Assets & ,977 1,023 RISK percentual ,776 1,289 a. Dependent Variable: Market Value - Total - Fiscal

Coefficientsa

Model Collinearity Statistics Tolerance VIF 1 DUMMYVARIABLEBEFOR EAFTER ,766 1,305 CSR-averageSPSS ,740 1,351 CSR change ,825 1,213 Return on Assets & ,974 1,026 RISK percentual ,742 1,349

(42)

41

Market Value - Total - Fiscal ,872 1,146 a. Dependent Variable: dumyvariable 2009

Coefficientsa

1

CSR-averageSPSS ,740 1,351 CSR change ,825 1,212 Return on Assets & ,982 1,019 RISK percentual ,918 1,089 Market Value - Total - Fiscal ,879 1,138 dumyvariable 2009 ,962 1,039 a. Dependent Variable: DUMMYVARIABLEBEFOREAFTER Table 23: VIF analysis

(43)

42

HISTOGRAMS & Q-Q plots

(44)

43 graph 2

(45)

44 graph 3

(46)

45 graph 4

(47)

46

CSR change

(48)

47 graph 6

(49)

48 graph 7

(50)

49 graph 8

(51)

50

Return on Assets &

(52)

51 graph 10

(53)

52 graph 11

(54)

53 graph 12

(55)

54

RISK percentual

How did the financial crisis affect the correlation between CSR performance and corporate financial performance in the financial sector?