Are investors sensitive to corporate social responsibility?

(1)

UNIVERSITY OF AMSTERDAM

MSC BUSINESS ECONOMICS, FINANCE TRACK

Are Investors Sensitive to Corporate Social Responsibility?

MASTER THESIS

July 2016

Author: Ying LIU

Student number: 11129476

(2)

Statement of Originality

This document is written by Student Ying LIU 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)

Acknowledgements

I am very thankful to Prof. Dr. Rafael P. Ribas for supervising me on my master thesis. His opinions are inspiring and constructive. His supervision ensured me along the right direction to pursue my research and thesis writing.

I would also like to acknowledge Mr. Yong YIN, Ms. Wei LIU and Ms. Yuhong ZHONG. Without their encouragement and support, I would not come to University of Amsterdam and finish this master program.

Furthermore I would like to thank my family, especially my parents, my husband and my lovely daughter, who give me great support and are always with me. Without them I cannot dedicate myself to study.

At last, I want to thank Meng LIU, my good partner in study during the past one year. I will remember all the moments we spend together in UvA.

(4)

i

This thesis addresses the endogeneity concern which current empirical studies suffer severely in exploring the effect of disclosing corporate social responsibility (CSR) on firm market performance, by employing a conjunction of panel regression and instrument variable regression. The instrument is constructed based on the methodology changes of KLD database. Making use of the corporate social performance (CSP, identical to CSR in this thesis) captured by newly introduced indicators in KLD, this thesis cuts off the link between lagged corporate financial performance (CFP) and the following CSP, and makes CSP exogenous. Based on a sample of 988 US firms and 3475 firm-year observations during the period of 1992 through 2013, this thesis finds no causal impact of CSP on CFP measured by Tobin’s Q, which implies investors are insensitive to CSR; those earlier positive and significant results are due to firm level heterogeneity and simultaneous causality bias. This conclusion is robust for an alternative measurement of CFP (ROA) and an expanded sample.

Keywords: Corporate Social Responsibility; Market Value; Investment; Instrumental Variable

(6)

2

1. Introduction

Corporate social responsibility (CSR) has been discussed and researched for several decades. Every year, large corporates spend a huge amount of money on social responsible activities, such as charitable contributions, community welfare, advertising on social responsibility products and so on. According to a survey by Economist in 2008, 53.5 percent of the responding firms hold the view that “corporate social responsibility is a necessary cost of doing business” (Baron et al., 2009). Advocators of CSR treat it as an intangible asset of a firm. They believe a broad range of stakeholders like employee, customer, community, environment protect NGO and other interest groups which are affected by a firm’s operation can influence a firm’s economic performance, so managers need to meet their demands in order to maximize profits and minimize potential costs of boycotts, lawsuits, and so on. At the same time, disputation and critique never stop. Opponents argue that managers take CSR for their personal benefits and it is a waste of money and time which harms firm value. The center of debate is the rationale for a firm engaging into social responsible activities: is it beneficial to shareholder’s welfare or is it at the cost of impairing corporate financial performance (CFP)?

This thesis attempts to test the effect of disclosing corporate social responsibility on corporate financial performance. Considering methodological concerns and contradictory findings in empirical studies, I hypothesize that there is no causal relation from corporate social responsibility to financial performance; in other words, investors are actually insensitive to corporate social responsibility. In order to test the null hypothesis, I exploit an exogenous change in the CSP index that is revealed to the market through methodological changes in KLD (Kinder, Lydenberg, and Domini, an independent third-party CSR evaluation agency) annual assessment. After investors learn more, if they take the disclosed CSR into account, CFP (measured by Tobin’s Q in this thesis) will increase or decrease according to theories above. Otherwise, no causal impact of CSP index on Tobin’s Q will be observed.

Actually, numerous studies have been carried on this topic, but the results are mixed and there is no consensus. Some meta-studies conclude there is a mildly

(7)

3

positive relationship between corporate social performance and financial performance (see, e.g., Orlitzky et al., 2003, and Margolis et al., 2009), while there are also increasing reports which find a negative or neutral association (see, e.g., McWilliams and Siegel, 2000, Garcia-Castro et al., 2010, and Baron et al., 2009). Just as Seifert et al. (2004, p.136) said, “the elusive link between CSP and firm financial performance is one of the most researched but least understood relationships in the field of business and society”.

In my opinion, methodological concerns are probably the main reason accounting for these inconsistent results. Accordingly, I find current empirical studies suffer greatly from the omitted variable bias and simultaneous causality bias. To isolate the causal effect of CSR on CFP, I employ a new instrument strategy, making use of CSP captured by newly introduced indicators in KLD database every year. This instrument is original in this research field.

After controlling firm level heterogeneity and the reverse causal bias, I find there is no significant effect of CSP on CFP, and thus I conclude investors are not sensitive to disclosed CSR. Using an alternative measurement of CFP (ROA) or an expanded sample yields the same conclusion. The absence of causal relation might suggest that CSR is just a signal of something else, such as the adaptability to changes, reaction capability, or other management heterogeneities. Or it might be the case as raised by Baron et al. (2009) that, firms adopting different CSR strategies can both achieve good financial performance due to various customer preferences. By comparing various regression models, I also illustrate previous ambiguous findings are mainly due to endogeneity concern. My finding emphasizes again the significance of dealing with endogeneity concern in empirical research.

Before developing the whole research, I first define CSR – the most important conception in this thesis. There are a large amount of different definitions about CSR. Here I am in support of the one proposed by Gordon and Michael (2014, p.3): “All corporate behaviors and organizational processes which directly or indirectly affect the corporation’s stakeholders – be it in a monetary or non-monetary way” can be regarded as CSR. I also follow common practices to treat corporate social

(8)

4

performance (CSP) and CSR identically, without considering the motivation of a social activity (Margolis et al., 2009).

This study proceeds as follows. Section 2 reviews mainstream theories and empirical findings, and summarizes main puzzles existing in current empirical research. Section 3 describes data and variables on CFP, CSP, the instrument, and control variables, as well as sample and descriptive statistics. The endogeneity concern in CSP-CFP link context is analyzed in section 4, followed by research design where the idea of the instrument is introduced in detail. Section 5 presents and interprets the main empirical findings and results of robustness checks; discusses limitations of this empirical work. In final section, a short conclusion closes this thesis.

2. Literature Review

2.1 Theories of the Relations between CSP and financial Performance 2.1.1 Doing Good Enables Doing Well

There are two mainstream theories in existing literature in terms of the relations between CSP and financial performance. Traditional economic and managerial theories assert all the resources should be used on the shareholder of the firm, and the only thing managers should consider is the shareholder’s self-interests. Stakeholder theory instead proposes that managers should take account of many other interest groups when making decisions and reallocating sources because they all count for firm’s operation (see, e.g., Freemen, 1984, Donaldson and Preston, 1995; and Mitchell et al., 1997). Advocates believe that “the favorable social performance is a requirement for business legitimacy and that social and financial performance tend to be positively associated over the long term” (Preston and O’Bannon, 1997, p.420); firms must pay attention to stakeholders’ interests for the purpose of maximizing their values (Jenson, 2001).

A consensus on the coverage of a corporate’s stakeholders is that it involves internal and external groups, mainly including employees, minority groups in corporate such as women, customers, communities, environmental groups, and other

(9)

5

potential political interest groups. They can be summarized from three dimensions: E (environment), S (social) and G (governance).

Various stakeholders can influence firm value through their own channels. According to “the social impact hypothesis” putted forward by Preston and O’Bannon (1997, p.421), a firm can enhance its reputation and financial performance by meeting the requirements of its main stakeholders, while the contrary CSR policy may increase “risk premium and lost profit”. In a similar way, Waddock (1997, p.307) proposes “good management theory” which argues favorable relationship with primary stakeholders helps to achieve better financial performance. For example, firms may boost employee’s morale and thus promote efficiency through good CSR policy on labor relations. They could also strive for support from local governments to enjoy policy benefits by maintaining good community relationships. Positive perceptions with respect to associations with customers, environmental activists and other stakeholders can improve competition and “reduce stakeholder management cost”.

Baron (2001) raises the conception “profit-maximizing CSR”, that is, a firm behaves socially responsibly with the purpose of increasing profit. In this sense, CSR is a strategic tool which aims to promote the competitive advantage. Baron et al. (2009) illustrate several factors which might advance firm value because of good social practices: First, “consumers could value CSP and be willing to pay a premium for the goods and services of a firm that provides social performance”, so managers could make use of CSP as “a complement to or a substitute for advertising, branding, and product quality” (p.10). Second, CSP might help recruiting and retaining high quality employees. Third, firms behaving well might obtain suppliers’ favors. Forth, some investors pay a higher price for the shares of firms who display good social responsibilities.

McWilliams and Siegel (2001) express a similar idea on how CSR enhances firm profit through consumer. They argue that some consumers hope to pay for goods with some kind of “socially responsible attributes (product innovation)”, and even favor products which are produced “in a socially responsible manner (process innovation)”,

(10)

6

and thus “consumer-oriented CSR” can serve as a label which increases “product identification and differentiation” (p.605). In addition, CSP can be treated as a reputation which indicates the firm is reliable, so consumers are prone to believe the firm’s products are also with good quality and are willing to pay more.

There are also many other literatures discussing how CSP affects firm’s profit through various stakeholders. To summarize: attract more socially conscious consumers and employees thus improve economic performance through their loyalty (e.g., Benabou and Tirole, 2010); maintain good relationships with environment, community, and other interest groups and reduce social pressures (e.g., Konar and Cohen, 2001); lower the cost of financing both in equity and debt because firms with good CSR are usually regarded with superior corporate management and less default risk (e.g.,Ghoul et al., 2011); reputational benefits contribute to extra return (Porter and Kramer, 2006).

2.1.2 CSP Damages Firm Profits

Conversely, other economists like Friedman hold the view that “the only social responsibility of business is to increase its profits” and it has no obligation to any non-shareholders (Friedman, 1970). He criticizes investing on stakeholders is at the cost of shareholder’s wealth, and it is a violation of fiduciary duty of managers. Aupperle et al. (1985) and Preston and O’Bannon (1997, p.421) propose that expenditure on charity, environmental protection, and community development may “siphon off capital and other resources from the firm, putting it at a relative disadvantage compared to firms that are less socially active”.

The root of the disagreement lies in agency cost, that is, executives are a proxy of shareholders and they have incentives to pursue their self-interests instead of shareholders’. There are many arguments which are in line with this view. Taking employee relations as an example, high compensation may raise employee’s loyalty and make routine management smoother, reduce potential conflicts with aggressive labor unions, and get a good impression from employees to enhance job safety. It is argued that CEOs take these policies for private benefits rather than the profit maximization of shareholders (see, e.g., Jensen and Meckling, 1976, and Cronqvist et

(11)

7

al., 2009).

Relatedly, Preston and O’Bannon (1997) originally raised “managerial opportunism hypothesis”, which links the available financial funds to social activities. They argue that good financial performance leads to reduced input on CSP because managers want to occupy the available profits for their welfares. However if the firm’s financial performance behaves not so good, they will increase expenditure on CSP in an attempt to conceal or make excuses for their poor performance. Obviously, investment on CSP regardless of financial conditions will damage shareholder’s interest.

Baron et al. (2009) also mentions that CSP would be “a perquisites for managers based on their own moral, warm glow, or self-interested preferences”. For example, a socially conscious CEO may keen to donate to environmental protection NGO for self- satisfaction or praise from the public. In addition to private benefit of managers, Baron (2001) puts forward another possibility that CSP may harm firm value. He distinguishes three sorts of social activity motivations: profit-maximizing CSP, altruism and threats by activists. The last one which is called “responsive CSP” is defensive and difficult to predict in advance and thus disturbs the scheduled market strategy which may impair CFP.

2.1.3 Some Alternative Theory or Speculation

There are also some alternative views apart from these two dominant theories. For example, Baron et al. (2009) raise a theory in which CSP, CFP and the social pressure imposed on firm interact, and ultimately reach an industry level equilibrium. This theory takes the stakeholder theory as a basis and it assumes there are rewards from various interest groups, like consumers, employees and investors for CSP. The difference is that it supplies a contingent but not a definite prediction. Taking the product market for example, firms may provide socially conscious customers goods with CSP attributes, and offset CSP costs through a higher price. Conversely, they can also take those who prefer a lower price as target guests and sell them products with no CSP. Two strategies can make profits due to market segmentation; Moreover, the latter might achieve higher profits under some circumstance. In this sense, CSP will not certainly lead to a better or inferior financial performance.

(12)

8

Bowman and Haire (1975) study two cases and discover an inverted U-shaped relationship between CSP and CFP. They argue that the corporate social responsibility itself is not a cause of higher profit. It is “a signal of good, sensitive, informed, balanced, modern, negotiating, coping management. For many issues it is good neither to underrespond to them, nor to overrespond to them”. Waddock and Graves (1997) also express a speculation about the importance of good management in evaluating CSP-CFP relation, although they do not hold the view that CSP-CFP relation does not have causal meaning: “It is thus entirely possible that there are direct linkages between the overall quality of management and CSP [SP]…Further, if quality of management is a critical variable in financial outcomes, then controlling for the quality of management while assessing the CSP-financial performance link might also be beneficial” (p.315).

2.2 Empirical Findings

In favor of different theories, there are extensive empirical studies investigating the association between CSP and firm financial performance. It roughly origins from 1970s and the results are very mixed.

Orlitzky et al. (2003) conduct a meta-analysis based on 52 studies with time period from 1972 through 1997. They draw a conclusion that there is a positive correlation between firm social practices and financial performance, and the effect measured by accounting-based proxies seems stronger than that measured by market-based ones. Margolis et al. (2009) summarize 251 studies from 1972 to 2007 through meta-analysis methods. They conclude that the overall effect is positive, small, and significant. Besides, according to a review made by mercer in 2011, the positive, neutral and negative findings on ESG factors and CFP or stock return are 61%, 22% and 17%, respectively, based on the period of study from 1963 to 2007.

In particular, Waddock and Graves (1997) investigate the relation between CSP and profitability based on a time horizon of consequent two years and a sample consist of 469 US companies. They conclude that better financial performance improves CSP, and superior CSP also increases profitability. Gompers et al. (2003) examine how governance affects Tobin’s Q and operating performance. Their finding is that overall

(13)

9

stronger shareholder rights lead to better Tobin’s Q and operating performance. Jiao (2010) identifies two instruments and implements a 2SLS regression. His finding supports previous conclusion that CSP is positively related with Tobin’s Q and ROA. Many other studies report a similar positive result in terms of the influence of CSR on firm economic performance (see, e.g., Preston and O’Bannon, 1997, Hillman and Keim, 2001, Bebchuk et al. 2013, and Borgers, et al. 2013).

On the contrary, McWilliams and Siegel (2000) show the positive association in previous findings is due to omitted variable bias, and actually CSP shows no significantly positive effect on profitability. Garcia-Castro et al. (2010) use firm level fixed effect and conclude the effect of CSR on ROE and MVA vary from positive to negative, and turn to be neutral on ROA and Tobin’s Q. Moreover, using the within method, Baron et al. (2009) find CSP represented by KLD strength indicators has no significant effect on Tobin’s Q, but social pressure measured by KLD concern indicators has a negative influence on Tobin’s Q.

There are also extensive literatures focusing on the effect of financial performance on CSP. In general, empirical research finds a positive and significant causal relation which supports the view of “Doing well enables doing good” (see, e.g., McGuire et al., 1988, Preston and O’Bannon, 1997, Waddock and Graves, 1997, Margolis et al., 2009, and Baron et al., 2009). The underlying theoretical explanation is that better financial performance prepares a firm the ability to invest in social responsible activities which is known as “the slack resource hypothesis”. These consistent findings emphasize an important issue in exploring influence of CSP on CFP, that is, the reverse causality problem, an obstacle which cannot be ignored in order to get a causal conclusion.

There might be several reasons to interpret these controversial results, such as measurement inconsistencies of CSR (Waddock and Graves, 1997), various study periods and samples, omitted variable bias (McWillliams and Siegel, 2000), different relationships in short-run and long-run (Ogden and Watson, 1999), and the reverse causality concerns (Garcia-Castro et al., 2010, and Jiao, 2010). In summary, there are still some puzzles to be solved in the empirical literature. As mentioned in introduction section, one issue is a large number of studies employ the pooled OLS

(14)

10

regression and do not control for firm level heterogeneity. Another concern is that the effort to deal with reverse causality issue is inadequate. In particular, little valid instrument is used in current empirical studies. As a result, they usually observe a significant and positive effect which cannot exclude the omitted variable concern and simultaneous causality bias.

3. Data, Variables and Descriptive Statistics 3.1 Data Sources

The social performance data and the instrument data come from KLD STATS database. KLD (Kinder, Lydenberg, and Domini) is a leading rating agency which specializes in evaluating corporate social performance. KLD STATS (Statistical Tool for Analyzing Trends in Social and Environmental Performance) is the data set with annual assessment result of firm social performance rated by KLD professionals (RiskMetrics Group, 2006). It is well acknowledged as one of the longest and most influential data sources available for investors and scholars (Berman et al., 1999, and Magolis and Walsh, 2003).

KLD STATS covers firms in MSCI KLD 400 Social Index and the S&P 500. It provides annual CSP report for 650 firms from 1991, the top 1,000 publicly traded US firms from 2001, and the top 3,000 publicly traded US firms from 2003. Since 2013, it began to cover non-US firms which are the constituents of series of MSCI index.

Except for its broad coverage, KLD has several advantages compared with other ESG or CSR related evaluation criteria. First, KLD evaluates multidimensional issues about CSR, not only including consequences, but also incorporating processes and management qualities. It covers a wide range of internal and external stakeholder relations (Waddock and Graves, 1997). Second, KLD uses multiple and comprehensive sources in its evaluation process. Corporate data sources include 10-k, sustainability report, proxy report as well as AGM results. External resources include academic, government and NGO datasets, 1600+ media, and other stakeholder sources. Third, KLD uses quantitative criteria (e.g., emission amount, fines, or donation figure) as much as possible, specifies descriptions for qualitative indicators,

(15)

11

and asks firms for verification to confirm accuracy (MSCI ESG RESEARCH Inc., 2015).

KLD STATS is available from WRDS. Financial data which are used to calculate Tobin’s Q and control variables are from Compustat which is also available from WRDS.

3.2 Variables Construction

3.2.1 Corporate Social Performance (CSP)

KLD STATS estimates the quality of a firm’s social performance from the perspective of environment, social and governance, in terms of seven themes which are “Environment”, “Community”, “Human rights”, “Employee relations”, “Diversity”, “Product”, and “Governance”.1_{Within each theme, there are several key issues} (indicators) which may be classified as strength or concern, denoting best practice or negative performance, respectively. The quantity of strength and/or concern differs across categories, and even within one category. It may also vary in different years due to methodology change.2

All the indicators are scored by a binary model. If a firm meets the criteria settled for an indicator (strength or concern), a “1” is assigned, and otherwise a “0” is assigned. For the data set before 2014 year, KLD provides the total scores of all strengths and all concerns for every category, respectively. Since data set in 2014 year is not included in my sample due to instrument, I then need not to calculate the corresponding total strength and concern values in that year. I add all strengths and subtract all concerns in a given year, and construct the score variable “CSP”. During this process I follow the common practice in literature and assign each indicator the equal weight (see, e.g., Hong and Kostovetsky, 2012, and Jiao, 2010).

1_{There are some changes with respect to these dimensions. For example, KLD renamed “other” category into} “governance” in 2002.

2_{Take year 2014 for example, “Environment” includes clean technique (strength), toxic emissions and waste} (strength), packaging materials and waste (strength), carbon emissions (strength), environmental management systems (strength), water stress (strength), biodiversity and land use (strength), raw material sourcing (strength), financing environmental impact (strength), green buildings (strength), renewable energy (strength), electronic waste (strength), energy efficiency (strength), product carbon footprint (strength), insuring climate change risk (strength), other strengths, regulatory compliance (concern), carbon emissions and waste (concern), energy and climate change (concern), impact of products and services (concern), biodiversity and land use (concern), operational waste (concern), supply chain management (concern), water stress (concern), and other concern. While “Community” only includes community engagement (strength) and community impact (concern).

(16)

12

Apart from the seven category indicators, KLD provides a special theme named controversial business involvement indicators, identifying companies who derive revenues from alcohol, firearms, gambling, military, nuclear power and tobacco related products or businesses. Unlike ordinary indicators, they are all concerns. There is disagreement in regard to including them or not when constructing a total CSP score. I do not incorporate them because they are mainly designed for negative screening purpose, which is a strategy of social responsible investment. Actually there is dispute whether people, except norm-constrained investors, appraise a firm negatively based on its behaviors or just belonging to sin industries.

3.2.2 CSP Scores Owing to Methodology Changes (CSP)

KLD STATS began in 1991, with 67 indicators within seven categories. Since then, there are methodological changes in the subsequent 18 years, mainly in terms of indicator increase or decrease, covering all the seven categories and accounting for 66 percent of the total indicators. The aggregate score due to indicators newly introduced in one year is identified and treated as an instrument for CSP.

Unfortunately, KLD STATS does not provide explicit clues on the track of indicator adjustments. The methodology history section in KLD manual (MSCI ESG RESEARCH Inc., 2015) is not very thorough and clear. For example, the time when some indicator is first introduced in manual cannot coincide with its first appearance in database. I also haven’t found any other materials with the motivations of methodology change which may help understanding. To be as precise as possible, I identify the newly added indicators every year in database by hand, and then double check with KLD manual and data list, excluding those just renaming or switching to other categories without any content variants.

Finally I get 70 newly introduced indicators from 1991 through 2014. In particular, these changes happen during sixteen years (1992, 1993, 1994, 1995, 1996, 1998, 1999, 2000, 2002, 2003, 2005, 2006, 2007, 2010, 2012, 2013), cover all the seven categories (Environment: 17; Community: 4; Human rights: 12; Employee relations: 10; Diversity: 5; Product: 9; and Governance: 13), and involve both strength (41) and concern (29). Following the method in constructing variable “CSP”, I create variable

(17)

13

“CSP” which captures the scores owing to newly added indicators in each year, and obtain variable “CSP*”, which is equal to CSP minus CSP.

3.2.3 Corporate Financial Performance (CFP)

Two forms of definition on corporate performance are dominant in the study of social performance and financial performance relation. One is a market-based view with Tobin’s Q or stock returns as the commonly used measure, while corporate operational performance is the accounting-based one with popular measures like return on assets, return on equity and profit margin. It is generally agreed that ROA is based on firm’s contemporaneous income, whereas Tobin’s Q is a forward-looking measure that reflects both current and expected future profitability of investment (Guenster et al., 2005). There is no consensus in prior literature on which one is better. Proponents of Tobin’s Q argue that it measures firm’s market value which exactly indicates how CSP affects shareholder wealth (Mackey et al., 2007). Guenster et al. (2005) use both accounting and market measures in that they believe social responsible activities may have both tangible and intangible impact, which are gauged by ROA and Tobin’s Q together. Considering the managerial manipulation concern about accounting-based measures, I take Tobin’ Q as the proxy for firm financial performance, and examine the effects of CSP on ROA in robustness checks.

I follow prior work and use average Q, where Q is defined as the market value of a firm’s assets divided by its replacement cost. The numerator is calculated as the market value of equity plus the difference between book value of assets and book value of equity (the debtor’ claim). The replacement cost is estimated by the book value of assets. The formula of Q is as follows：

𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄 = (𝐶𝑆𝐻𝑂 ∗ 𝑃𝑅𝐶𝐶_𝐹)+(𝐴𝑇−𝐶𝐸𝑄+𝑇𝑋𝐷𝐵 )

𝐴𝑇 (1)

3.2.4 Control Variables

This thesis refers to previous literature (see, e.g., Waddock and Graves, 1997, McWilliams and Siegel, 2000, Gompers et al. 2003, Baron et al. 2009, Jiao, 2010,

(18)

14

Bebchuk et al. 2013, and Borgers, et al. 2013) and employ the following variables related to Tobin’s Q and CSP to control the omitted variable bias:

Firm size: the natural logarithm of total assets. Large size can have positive effect on profitability through economies of scale and bargaining power versus suppliers and buyers (Porter, 1985), and big firm usually spends more on CSP; meanwhile Orlitzky (2001) empirically shows firm size has no significant effect on CFP or CSP.

Firm age: the natural logarithm of firm age in the form of number of months. Firm age is deduced by the earlier time when a firm first appeared in the Compustat database or its IPO date. It is expected to negatively affect Tobin’s Q, because older firm is subject to suffer from “organizational rigidities, costs rise, margins thin, growth slows, and rent-seeking behavior which damage its compete advantage” (Claudio and Waelchli, 2010). Older firm may positively relates to CSP if it has more slack resources.

Investment: the ratio of capital expenditures to total assets. Investment is generally considered positively associated with Tobin’s Q, yet Markus (2015) argues the effect depends on the underlying managerial motive. High investment strategy may raise more strict qualifications for employees, so firms are likely to invest more in human resources (Jiao, 2010).

Sales growth (sales): the ratio of current year’s total sales to previous year’s total sales. Sales growth indicates good profitability and more available resource for CSP, thus it is anticipated to positively influence Tobin’s Q (Mitra et al., 1991) and CSP. Return on assets (ROA): the ratio of the difference of operating income before depreciation minus depreciation and amortization to the lag total assets. Profitability is an important performance reference about firm value, so ROA is expected to have positive relation with Tobin’s Q. It may also positively affect CSP due to the slack resource.

Leverage: the ratio of debt to total assets. High leverage usually implies high default rate and leads to inferior valuation (Waddock and Graves, 1997); it may also negatively influence CSP due to more risks which increase uncertainty in the fund available to social responsibility activities (Roberts, 1992).

(19)

15

Dividend dummy: a dummy variable which equals to one if a firm pays dividend, and zero otherwise. According to corporate finance theory, firms with good investment opportunities should spend most of their earnings to business, and thus dividend payout is negative associated with Tobin’s Q. However if there are excess funds after investment expenditure, dividend payout contributes to firm value (Amold, 1988).

R&D intensity: the ratio of research and development expenditures to total sales. R&D is thought to be positively related to Tobin’s Q, since it signifies product innovation and differentiation strategy which increase firm value. On the other hand, firms with high R&D investment especially in terms of product and process innovation often accompany with CSP labels (McWilliams and Siegel, 2000, and Hall, 1999).

Advertising intensity: the ratio of advertising expenditures to total sales. Advertising intensity is “a proxy of product differentiation and entry barriers that might serve to enhance firm profitability” (McWilliams and Smart, 1993). At the same time, firms with good CSP increase their reputation through advertising champions (McWilliams and Siegel, 2000).

Delaware Incorporation: a dummy variable which equals to one if the firm is incorporated in Delaware state, and zero otherwise. Compustat provides the firm’s incorporation information. Daines (2001) finds that when a firm is incorporated in Delaware, its Q is obviously higher.

Industry dummy: Industry is a standard control variable in Tobin’s Q and CSP literature to control for competition intensity, industry-specific regulation and so on. A set of dummies based on two-digit SIC code is included in the regression.

Location dummy: Location dummy is used to control unobserved location-specific factors such as regional regulation and consumer preference variations. A firm’s location is identified through its headquarter.

(20)

16

TABLE 1 ABOUT HERE

3.3 Sample and Descriptive Statistics 3.3.1 Sample Construction

I download data when there are newly added indicators in a particular year from KLD STATS and create variables “CSP”, “CSP”, and “CSP*”; download other data from Compustat North America Fundamental Annual. Data with lag value are pre-prepared for avoiding the reduction in sample size. After creating all the variables, data from two databases are combined. One problem is KLD supplies inadequate firm identifiers, only including company name, company ID (most are missing), CUSIP code and ticker symbol. Besides, some are with both 8-digit cusip and ticker, some only with one of them, some with 6-digit or 7-digit cusip, and some even with unusual cusip or ticker like ‘ ’, ‘0’, ‘# ’, etc. Therefore, those with both cusip and ticker are merged according to year, cusip and ticker; those with 6-digit or 7-digit cusip are merged again, respectively; those with missing cusip are merged using year and ticker. Non-US firms are then excluded from sample.

I replace the value with zero if Tobin’s Q is negative, and also make sure variables “investment”, “leverage”, “sales”, “R&D intensity”, and “advertising intensity” with non-negative values. All the variables are winsored at 1 percent3_{unless they are} expressed as a natural logarithm4_{. After excluding firms with missing data, the final} sample consists of 988 firms and 3475 firm-year obs. It is an unbalanced panel whose time series is from 1992 through 2013 but with gaps in 1997, 2001, 2004, 2008, 2009 and 2011.

It deserves some explanation that firms are discretionary to report R&D and advertising information in Compustat, so there are many missing data on these two variables. This is the main reason accounting for the final sample size. Many researches (see, e.g., Himmelberg et al., 1999, Woidtke, 2002, Jiao, 2010, Bebchuk et al. 2013, and Borgers, et al. 2013) create “a dummy equal to one when the relevant

3_{Tobin’s Q and investment are high only.} 4_{Here refers firmage and firmsize.}

(21)

17

data is missing to control for the possibility that non-reporting firms are discretely different from reporting ones and to avoid a significant reduction in sample size” (Jiao, 2010). This method is first introduced byCohen and Cohen (1985) in their textbook. However, there are different opinions on the rationale of this method (see, e.g., Allison, 2001)5_{. So I keep the sample without handling these missing data in my main} regression, and implement robustness check with an expanded sample by following the method in literature mentioned above.

3.3.2 Summary Statistics

Table 2 summaries the information on total CSP scores by years and its attributes in my sample. From Panel A it can be seen, observations in the second half time period of my sample (from 2002 through 2013) account for the vast majority. During my study period, the annual average CSP score is positive in eleven of sixteen years, and it yields a mean of 0.23 in entire sample period, with minimum value of -9 and maximum value of 17. Negative CSP score occurs during 2003 to 2010. It seems the differentiation of CSP is greater after 2000. Panel B reveals in the seven dimensions of CSP, firms in my sample behave better in environment, community, employee and diversity, while worst in respect of governance. Besides, the average CSP scores on product and human are also negative.

TABLE 2 ABOUT HERE

Table 3 contains descriptive statistics for Tobin’s Q, the instrument and all the control variables. Reported in the third column of table 3, the average Tobin’s Q of firms in my sample is 2.51, varying from 0.55 to 8.40. The average firm age is about 23 years, with the youngest one is 2 years old, while the oldest has been 64. Total asset also differs greatly, which ranges from $26.58 million to $304,594 million.

5_{Allison (2001, p. 76) points: “the rationale for this method is that it makes use of all the available information} into the regression. But Jones (1996) proved that this method typically produces biased estimates of the regression coefficients, even if the data are missing completely on random. He also proved the same result for a closely related method for categorical predictors whereby an extra category is created to hold the cases with missing data. Although these methods probably produce reasonably accurate standard error estimates, the bias makes them unacceptable.”

(22)

18

These firms are with average ROA of 11.42 percent, 114.83 percent increase in sales growth, and 17.29 percent of leverage. In terms of expenses, firms spend in average 4.99 percent of their total assets on investment, 9.23 percent of their total sales on R&D, 2.8 percent of total sales on advertising, and 43 percent of them pay dividend. It can be seen that there is quite some variation among firms. For example, the respective standard deviation of total assets, leverage, sales growth and R&D intensity are 24955.56, 18.97, 25.54 and 31.42. In my sample, 67.11 percent of firms are incorporated in Delaware State. Besides, as estimated, there is not much variety in the instrument, with a mean value of -0.01 and standard deviation of 0.53.

TABLE 3 ABOUT HERE

At last, the summary statistics of pairwise correlations between variables is shown in Table 4. In my sample, CSP has a mild and significant correlation with Tobin’s Q, which seems being in line with the view that doing good enables doing well, or at least CSP won’t impair CFP. But it is just a signal rather than a causal conclusion. Aside from the instrument, all the other variables are associated with Tobin’s Q, significant at 1 percent confidence level, among which the coefficients of firm age, firm size, leverage and dividend are negative, and those of investment, ROA, sales growth, R&D intensity, advertising intensity, and Delaware incorporation are positive. It is consistent with mainstream views discussed previously, and is accordance with findings of Bebchuk and Cohen (2005), and Jiao (2010). In the meantime, most of the variables are also significantly interrelated with CSP, except investment, leverage, R&D, and Delaware. It gives some signs to incorporate them as controls; otherwise we may not get a consistent estimation due to the omitted variable bias. Finally, the instrument, the variable “CSP” is strongly related with CSP, but shows no significant association with Tobin’s Q, which is in line with my conjecture.

(23)

19

4. Research Methodology 4.1 The Issue of Endogeneity

Due to unobservable characteristics that differ across firms, some firms may take on both high social performance and high financial performance, while others may have inferior social performance and financial performance simultaneously. Thus the coefficient of CSP actually includes the effect of other factors and omitted variable bias arises. For example, McWilliams and Siegel (2000, p.606) argue “the positive and significant coefficient on CSP, as reported by Waddock and Graves (1997), could simply reflect the impact of R&D intensity on firm performance. It is impossible to isolate the impact of CSP on firm performance unless the model is properly specified”. Baron et al. (2009) propose that unobservable firm heterogeneity leads to the wrong conclusion reported in previous studies. They empirically show that when firm fixed effect is included, there is no significant correlation any more.

The second genre of endogeneity is the simultaneous causality (reverse causality). Under this circumstance, the causality between CSP and CFP exists in both directions: forward and backward, which makes the estimation of CSP to CFP is biased. Literatures supply both theoretical explanation and empirical evidence for the existence of simultaneous causality in CSP and CFP relation. Preston and O’Bannon (1997) distinguish the causal link between CSP and CFP in both directions. They subsequently develop six hypotheses, among which “available funds hypothesis” (also seeWaddock and Graves, 1997: slack resources theory) predicts a positive association between lagged CFP and the following CSP, while “managerial opportunism hypothesis” holds the opponent view. Slack resource could lead to superior social performance either if the management layer believes CSP promotes shareholder’s welfare or the CEO tries to gain personal benefits. The opposite situation may appear if managers are in pursuit of short-term benefits or make excuse for disappointing performances.

4.2 Research Design

(24)

20

several strategies trying to obtain the causal relationship, including control variables, panel regression with year and firm fixed effect, and instrument variable regression. The following equation will be estimated:

𝑪𝑭𝑷_𝒊𝒕 = 𝜷_𝟎+ 𝜷_𝟏(𝑪𝑺𝑷_𝒊𝒕− 𝑪𝑺𝑷_𝒊𝒕∗) + 𝜷_𝟐𝑪𝑺𝑷_𝒊𝒕∗+ 𝜸_𝟏𝑿_𝒊𝒕+ 𝜶_𝒊+ 𝝀_𝒕+ 𝒖_𝒊𝒕 (2)

In equation (2), 𝐶𝐹𝑃𝑖𝑡 is the dependent variable indicating financial performance of firm i in year t, 𝛽0 is a constant, 𝑋𝑖𝑡 is a set of control variables, 𝛼𝑖 and 𝜆𝑡 represent firm fixed effect and year fixed effect, respectively, and 𝑢𝑖𝑡 is the error term. 𝐶𝑆𝑃_𝑖𝑡 is the social performance measured by KLD STATS score, 𝐶𝑆𝑃_𝑖𝑡− 𝐶𝑆𝑃_𝑖𝑡∗ is used as the instrument indicating disclosed social performance variance of firm i captured by the newly introduced indicators of KLD STATS in year t, and 𝐶𝑆𝑃_𝑖𝑡∗ is the value based on the methodology of the previous year. Here I do not use the two-stage least squares (2SLS) method, because the aggregate of the values of the instrument (𝐶𝑆𝑃_𝑖𝑡− 𝐶𝑆𝑃_𝑖𝑡∗) and the controlled part (𝐶𝑆𝑃_𝑖𝑡∗_{) is actually equal to the} value of independent variable (𝐶𝑆𝑃_𝑖𝑡). And it is also why it needs not to worry about the weak instrument problem.

4.2.1 Control Variables and Fixed Effect

For those observable omitted factors, the best way is to put them into the regression model, either implementing a multiple regression or just treating them as control variables. Previous literatures provide extensive discussion on controls (I mainly refer to those that financial performance is measured as Tobin’s Q, see, e.g., Waddock and Graves, 1997, McWilliams and Siegel, 2000, Gompers et al. 2003, Fisman et al. 2008, and Baron et al. 2009), and I categorize them as three kinds: the first is about firm’s fundamentals (e.g., firm age, firm size); second is those characterizing firm’s operation (e.g., capital expenditure, sales growth, R&D intensity, advertising, ROA); the third represents firm’s risk (e.g., leverage, dividend policy). Section 3.2.4 provides a detailed discussion on how these controls affect CFP and CSP.

Aside from observable omitted variables, there are also many unobservable heterogeneities which may affect CFP and CSP: some are macroeconomic variations

(25)

21

(Baron et al., 2009) such as general market conditions and mood, and technological and policy changes; some are industry specific factors like various supply and demand conditions (Baron et al., 2009), and competition intensity; some are region specific factors, like different regional regulation and consumer preference (Belu and Manescu, 2011); and some are firm specific characteristics, like management style, firm culture, bargain power of stakeholders and so on. For these factors difficult to capture and quantify, fixed effect is often used to control their effects. For instance, year fixed effect can account for those entity-invariant but time-variant factors; entity fixed effect can absorb the effect of those time-invariant but entity-variant ones. There are different views in literature concerning entity fixed effect. Some researches prefer to employ industry fixed effect with or without region fixed effect (Gompers et al. 2003, Jiao, 2010, Bebchuk et al. 2013, and Borgers, et al. 2013). However, just as Baron et al. (2009) prove, firm fixed effect is far more thorough than any other level of entity fixed effect.

4.2.2 Instrument Variables Regression

Although a large set of control variables along with year fixed effect and firm fixed effect account for a great deal of omitted factors, there may still exist time-variant, firm-level omitted variables which cannot be observed and eliminated. More important, reverse causality issue cannot be dealt with by control variables and panel regression. Here calls for the instrument variables regression. Its principle is to isolate the part which is uncorrected with the error term from the interested independent variable and use these variations as tools to get consistent estimation of the regression coefficient. The key point of instrumental variables regression is to find at least one valid instrument which is related to CSP but not directly related to CFP (affects CFP only through CSP). These are called instrument relevance and instrument exogeneity principle, respectively (Stock and Watson, 2010, p. 419, p. 434).

It is not an easy thing to find a suitable instrument; especially the exogeneity standard is often hard to meet. It is probably the main reason why there are not so many researches employing instrument in CSP-CFP field, even though there are piles of papers on this topic and the endogeneity concern is still an open question by now.

(26)

22

As showed in literature review section, Jiao (2010) makes use of two instruments to investigate the association of stakeholder welfare and firm performance: one is an indicator for positive earnings in the previous year; the other is equity ownership by activist public pension funds in the previous year. Although the author proposes arguments attempting to prove these instruments and Tobin’s Q are not interrelated directly, opposite opinions can be raised. For instance, one can question positive earning indicates good operation, and thus contributes to better Q. Woidtke (2002) empirically proves that activist institutional investors is negatively associated with firm value.

This thesis will introduce a new instrument based on the methodology changes of KLD STATS database (see section 3.2.2 for details on construction of the instrument variable). This instrument meets relevance and exogeneity requirements. On one hand, score induced by newly increased indicators is a composition of total CSP score in that year, thus it is definitely related with CSP score (the interested independent variable). On the other hand, it is not decided or influenced by firm’s previous financial conditions in a large part, since it is newly introduced.

It is worthwhile explaining in more detail why the instrument is in line with the exogeneity principle. As mentioned previously, one main concern about exogeneity of social performance is the reverse causality issue. Since the instrument is a constituent part of CSP score, it is necessary to illustrate it is not the case for the CSP captured by those newly introduced indicators. The starting point is to understand the motivation of corporate social activities. Baron (2001) once summarized three kinds of incentives: profit maximization, altruism, and threats by the interest groups and activists. For those firms who take CSP mainly driven by profit maximization or potential pressures, they pay great attention to their reputations and stakeholders’ attitudes, which can be severely influenced by KLD rating (scores). KLD rating is widely acknowledged by the public and investors. KLD disclosed in 2006 that “15 of the top 25 institutional financial managers in the world use its research and more than $10 billion fund investing is based on its ratings” (Chatterji et al., 2009, p.8). Just as inferior credit ratings, poor KLD social ratings can also shame firms, harm their reputations, weaken

(27)

23

their competitiveness, and even threaten their legitimacy. There are empirical findings which prove that firms rated poorly by KLD subsequently improved their social performance (see, e.g., Aaron et al. 2008, 2009). Therefore, it is reasonable to infer that many firms, especially large firms which are exactly included in KLD database, are likely to cater KLD’s criteria, and engage in social activities in line with KLD’s standard when there is slack resource. Nevertheless, a newly introduced indicator is not available before, meaning that firms cannot invest in it before hand, and thus the link from lagged CFP to subsequent CSP is broken. Of course, one can still call into question that altruists are not sensitive to KLD’s criteria, and they may have involved in social activities before the criteria or indicators are included in KLD STATS. Another concern is if there are still omitted factors after controlling observable variables and fixed effects, our instrument probably cannot solve the omitted variable bias issue. But at least, it can settle the reverse causality problem largely.

5. Empirical Results and Discussion 5.1 Regression Analysis

In this section I will compare the empirical results with various regression specifications, discuss their statistical and economic meanings, and try to discover the causal link between firm’s social performance and financial performance.

The empirical study starts with the simplest estimation on the CSP and CFP relationship which mainly coincides with work of Gompers et al. (2003). In Column (i) of Table 5, I employ a pooled OLS regression to estimate the effect of CSP on Tobin’s Q. Only firm age, firm size and Delaware incorporation serve as control variables, along with fixed effect on industry, state, and year level, to control observable and unobservable firm specific heterogeneity. The sign of my finding is consistent with that of Gompers et al. (2003). The baseline coefficient is 0.062, significant at the 1 percent confidence level. It means that one more score of CSP increases Tobin’s Q by 0.062. And a one standard deviation increase in CSP promotes by 11 percent of

(28)

24

Tobin’s Q’s standard deviation.6

In column (ii), I add more control variables which are frequently used in earlier studies, including investment, leverage, ROA, sales growth and dividend policy (paying dividend or not). As we can see, the adjusted R squared is dramatically raised from 0.261 to 0.408, indicating the improvement of the model specification to some extent. The coefficient of CSP drops from 0.062 to 0.048, due to other explanatory power coming from the increased controls. Although the magnitude of the coefficient is even smaller, it is still positive and strongly significant in statistical meaning.

According to McWilliams and Siegel (2000), Column (iii) adds R&D intensity which represents product differentiation strategy and reliable enterprise image, and advertising intensity which serves as a proxy for product differentiation and entry barriers. After including these two control variables, McWilliams and Siegel (2000) find the positive and significant result concluded by Waddock and Graves (1997) vanished. Whereas my finding is different: the sensitivity of the coefficient of CSP only changes slightly from 0.048 to 0.040. This significantly positive effect is in compliance with many previous studies which also consist of these two controls, like Jiao (2010), Bebchuk et al. (2013), Borgers, et al. (2013), and Dowell et al. (2000).

Until now we consistently observe a positive link between CSP and Tobin’s Q, and the effect is robust at the 1 percent confidence level. The coefficients of control variables basically agree with predictions by theory: firm age, firm size and leverage negatively associate with Tobin’s Q, and investment, ROA, sales growth, R&D intensity and advertising intensity display a positive effect on Tobin’s Q. As for Delaware incorporation and dividend payout, no significant results are found on these variables, which might imply an ambiguous influence on Tobin’s Q. It seems everything goes well. If the conclusion above is reliable, it confirms that investors appreciate better social performance and doing good brings financial benefit to firms, even though this benefit is very small. At least, it won’t damage firm’s economic function. Reward from stakeholders offsets or even exceeds cost involving in CSP. It

6_{The result is calculated as follows: One standard deviation (2.7773) increase on CSP raises Tobin’s Q by 0.062 *} 2.7773 = 0.1722, accounting for 11 percent of Tobin’s Q’s standard deviation (0.1722 / 1.54 = 0.11).

(29)

25

may come from the success of product differentiation strategy which identifies socially conscious consumers and investors, benefit from more harmonious employee relationship which attracts the talent and inspires productivity, avoid social pressures from community, environmental NGO and other interest groups, and/or reduce costly penalty due to violating regulations.

Nevertheless, endogeneity concern makes it far from being a causal conclusion. As Bebchuk et al (2013, p.343) wrote, “Our findings do not resolve the causality questions - which the literature has generally been unable to resolve - concerning the extent to which governance provisions directly cause or merely signal the worse performance of the firms having them”. Although their research emphasis is whether investing firms with superior CSP yields excess stock return, it is still important to verify the association between CSP and corporate performance, for it rationales the original source of excess return. If the foundation is not solid, their conclusion will lose robustness, since CSP is only a proxy of something else and this correlation may change someday unpredictably.

In order to deal with the endogeneity issue, I continue my research by taking into account firm fixed effect and employing panel regression using within method. The variable Delaware incorporation, industry fixed effect and state fixed effect are excluded, for otherwise there will be perfect multicollinearity.7_{A different picture} emerges. In column (iv), we observe a negative coefficient of -0.022, significant at the 5 percent confidence level. It indicates that one more score of CSP drops Tobin’s Q by 0.022 and a one standard deviation increase in CSP decreases by 4 percent of Tobin’s Q’s standard deviation. This different result compared with that deriving from previous OLS estimation can be explained by unobservable firm level heterogeneity. For instance, some management idiosyncrasy or CEO preference may lead to high social responsible behavior and financial performance simultaneously, vice versa. If these firms dominate the sample and the correlation persists during the whole study

7_{There is some problem with my sample when employing panel within method. My sample is an unbalanced} panel, and there are 732 firms with only one period observation. It means these observations are actually not used in within regression process. I alternatively estimate it using “least squares with dummy variables” method along with clustered standard errors. It is identical to within method in econometric. The regression results almost remain the same.

(30)

26

horizon, we may observe a spurious positive or negative relationship. The converse conclusion with and without firm fixed effect confirms the view discussed in section 4: observable omitted issue can be solved by adding control variables into empirical model, yet the unobservable firm level heterogeneity may still be there and bias the estimation. Industry specific fixed effect is not enough. That is one reason why previous studies report mixed findings.8

Despite the negative coefficient is very small, the point is the sign. A negative effect on Tobin’s Q may support “the managerial opportunism hypothesis” that manager pursues personal benefits from investing stakeholder relationship, such as personal reputation, job safety, less management pressure and so on. Investors regard CSP as the result of agency cost which damages firm value. It may also support the “responsive CSP hypothesis” raised by Baron et al. (2009) who argue CSP due to social pressure reduces CFP. If it is the case, firm may need to re-estimate its CSP strategy. But again, this result might still suffer from endogeneity concern. In addition to unobservable but time-variant omitted variables, another possible threat may be the reverse causality. In contrast to the negative coefficient in column (iv), Baron et al. (2009)9_{and Garcia-Castro et al. (2010) find no significant association between CSP} and CFP after controlling firm level heterogeneity. These different findings call for further research.

I continue my exploration by adding instrument into panel regression. As described in section 3.2.2, the instrument is the CSP score due to newly introduced indicators in a particular year (CSP). The most advantage is that it can be logically proved not to be decided by lagged CFP (see section 4.2.2 for details), so it won’t have the simultaneous causality problem. Since CSP is a part of CSP, there is not the weak instrument problem. Column (v) shows the final result. After controlling CSP score based on the unadjusted methodology which is used in the previous year (CSP*), with full set of control variables in column (iv), the effect on Tobin’s Q identified by the instrument is still negative, but far from being significant (t statistics is -0.65). This is

8_{Take the example above, with the same model specification and control variables like R&D intensity and} advertising intensity, McWilliams and Siegel (2000) and Jiao (2010), among others, reach different results. 9_{Baron et al. (2009) use KLD strength indicators to represent CSP.}

(31)

27

a striking finding for it indicates the direction and magnitude of the coefficient do not have any explanatory power, because no causal relationship exists between CSP and CFP. It might be the case that my sample is different with that of Garcia-Castro et al. (2010)’s. Nevertheless, there is still endogeneity concern in their research mainly due to reverse causality issue, and thus I argue their result is not reliable, although we reach the same conclusion at last. With respect to control variables, coefficients of firm age, firm size, ROA, sales growth and R&D intensity are significant and have the same sign with those estimated from OLS regression. Investment, leverage and advertising intensity are not significant any more.

The series of tests conclude with a neutral result which supports my hypothesis: there is no causal correlation from corporate social performance to financial performance, that is, investors are actually insensitive to CSP. All the previous significant findings may be just a mistake due to imperfect identification strategies. Although it may disappoint both advocators of stakeholder theory and shareholder theory, the truth is that CSP might merely be a substitute of something else which actually decides firm profit.

TABLE 5 ABOUT HERE 5.2 Robustness Checks

By far I have examined the effect of CSP on CFP through OLS, Panel regression and instrument variable regression. The conclusion is that I find no evidence of causal impact of CSP on CFP with Tobin’s Q as proxy. In this section I implement alternative checks to test the robustness of my finding.

Firstly, I replicate regression analysis in section 5.1 but make use of an expanded sample. As pointed in section 3.3.1, I drop the observations with missing data on “R&D intensity” and “advertising intensity” to obtain the final sample. It is the safest way to deal with missing data, but the defect is that it dramatically decreases the sample size, from 13595 observations with 3203 firms to 3475 observations with 988 firms (Fortunately, it does not waste the instrument, for all the sixteen years with

(32)

28

newly introduced LKD indicators are still in the reduced sample). I then follow a large number of literatures like Jiao (2010), replace all the missing data on “R&D intensity” and “advertising intensity” with zero, and include two respective dummies equal to one when data is missing into regression model. Table 6 reports the result. Consistent with the main regression, the influence of CSP on CFP is still negative (-0.005), and are far from being significant (t statistics is -0.36). Again, I conclude that there is no causal correlation between CSP and CFP. All the other regression results display the same sign with previous findings, except that the positive association disappears even in the panel regression (Column v). In conclusion, table 6 indicates that my findings are robust to an expanded sample.

TABLE 6 ABOUT HERE

Next, I further examine my results by using alternative measurer of CFP. It has been discussed in section 3.2.3 of the pro and con about the two commonly used measures: Tobin’s Q and ROA. Although existing literature shows more preference to Tobin’s Q, there are still many researches explore how CSP affects the profitability/operating performance with ROA as the proxy. In table 7, I repeat all the tests with ROA as the independent regressor. Moreover, considering ROA is an accounting-based measure, I follow previous studies, like Waddock and Graves (1997), Gompers et al. (2003), Jiao (2010) and Borgers, et al. (2013), and also test the relation of CSP and ROA of year t+1.

Table 7 presents the regression results. We can see that the influences of CSP on ROA calculated in the current year and the following year, respectively, are very similar. In column (i) and column (v), the coefficients of CSP are both positive (0.003) under the baseline OLS regression model. It is even smaller than that in Tobin’s Q regression. They are both significant, at the 5 percent and 1 percent confidence levels, respectively. After adding more controls, the coefficients in column (ii) and (vi) change slightly, still positive and significant. But when the firm fixed effect is

(33)

29

considered under panel regression, both coefficients become negative and no significant results are observed in column (iii) and (vii). At last, with the instrument variables regression, coefficients of CSP in column (iv) and (viii) keep the negative values but not significant, implying there is no causal link from CSP to ROA. I get the same empirical conclusion with that in Tobin’s Q regression. In addition, I find ROA seems more sensitive to investment, leverage, dividend and advertising intensity than Tobin’s Q, but shows no significant response to firm age.

TABLE 7 ABOUT HERE

5.3 Discussion

5.3.1 Interpretation, Inference and Implication

Considering the different results with and without firm fixed effect in my estimation, it can be inferred that some of the true factors which influence CFP and relate with CSP simultaneously are time-invariant during my study horizon, such as corporate culture, core values and so on. They are got rid of by employing the within method. However, it cannot be excluded the possibility that some other factors which change with time during my study period may also account for financial performance, like CEO’s preference, management style, some coping management strategy, and so on. They might be the true factors investors actually value. Neither panel regression nor the instrument in this study can totally deal with them. The main reason is my instrument is a part of the concurrent CSP, so it might also suffer the same problem with CSP in terms of omitted variable bias. But it is valid to the reverse causal issue, just as my final result indicates. Some scholars such as Bowman and Haire (1975), and Waddock and Graves (1997) speculate management heterogeneity as well.

From the perspective of “between” level, it might also be the case that, both firms adopting high or low CSP strategies can achieve good financial performance due to product differentiation strategy. This is in line with Baron et al. (2009). High-income consumers are more likely to pay a premium for CSP products, and firms make profit

Are investors sensitive to corporate social responsibility?

UNIVERSITY OF AMSTERDAM

MSC BUSINESS ECONOMICS, FINANCE TRACK