The impact of innovation on the relationship between carbon emissions and corporate financial performance.

(1)

The impact of innovation on the relationship between carbon

emissions and corporate financial performance.

by Felix Grube (s1024118)

Supervisor: Prof. D. Reimsbach

Second Reader: Prof. G. J. M. Braam RA Date: 10.08.2020

Master Specialization: Corporate Finance and Control Academic Year 2019/2020

Nijmegen School of Management Radboud University Nijmegen

(2)

Abstract

As the global awareness of climate change is rising, companies are increasingly trying to adapt their corporate environmental performance to stay competitive. This thesis aims to examine the relationship between carbon emissions and corporate financial performance and to what extent innovation moderates this relationship. Panel data consisting of 238 firms in 14 European countries for the period 2006 to 2018 is used. Four sub-models are developed with different proxies for corporate financial performance. The results indicate a negative effect of carbon emissions on a firm’s financial performance using accounting-based and market-based financial performance proxies. Furthermore, the influence of innovation on the relationship between carbon emissions and a firm’s financial performance is analyzed. A significant moderation effect is indicated by the results, with different outcomes for the sub-models. The findings imply that corporate environmental performance is positively related to a firm’s financial performance and that innovation moderates this relationship.

Keywords:

(3)

I

List of Figures

Figure 1: Moderation Model ... 14

Figure 2: Moderation effect of CRDintensity for Model 2.1 ... 24

(5)

III

List of Tables

Table 1: Firms per Country ... 10

Table 2: Firms per Economic Sector ... 10

Table 3: Variable Description ... 13

Table 4: Tobin's Q per Economic Sector ... 15

Table 5: Descriptive Statistics ... 16

Table 6: Pairwise Correlations ... 17

Table 7: Results Model 1 ... 20

Table 8: Results Model 2 ... 23

Table 9: Margins of ClCO2 for Model 2.1 ... 24

Table 10: Results of Model 2 without GB ... 27

(6)

IV

List of Abbreviations

BPLM Breusch-Pagan-Lagrange-Multiplier CEP Corporate Environmental Performance CFP Corporate Financial Performance CSR Corporate Social Responsibility ESG Environmental, Social and Governance EU ETS European Union Emissions Trading System

GHG Greenhouse Gas

GB Great Britain

R&D Research and Development

(7)

1

1. Introduction

Carbon emissions, or greenhouse gas (GHG) emissions, are proven to be an important determinant of global warming leading to climate change which is one of the greatest challenges of the 21st_{century. In}

order to limit global warming to 1.5°C, carbon neutrality has to be achieved at the latest by mid-century (de Coninck, et al., 2018). The European Union Emissions Trading System (EU ETS) is a policy response that aims to reduce carbon emissions according to the Kyoto Protocol. The EU ETS operates as a carbon allocation and pricing mechanism and thereby established an EU wide market for carbon emissions trading (Zhang & Wei, 2010). Since the ETS was launched in 2005, carbon emissions are priced and companies have to report them to a regulatory authority (Comyns & Figge, 2015). Thereby, companies were incentivized to reduce GHG emissions (Fernández López, Fernández Fernández, & Olmedillas Blanco, 2018). New regulations increased the climate consciousness among firms and made the EU a leader in the field of carbon reporting. Carbon emissions1_{belong to the environmental element}

of corporate social responsibility (CSR) which increasingly gains significance, not only for the reporting companies but also for investors. The rising awareness regarding climate change impacts investment behavior and enables the market to include environmental performance in the firm valuation and also in a company’s risk profile (Matsumura, Prakash, & C. Vera-Munoz, 2014). Because carbon costs tend to increase over time (Zakeri, Dehghanian, Fahimnia, & Sarkis, 2015) and the market penalizes firms for their GHG emissions (Matsumura et al., 2014), companies need to adapt their long-term goals to sustainable production and decrease carbon emissions to stay competitive. In addition, firms can gain a competitive advantage by proactively following environmental issues. By integrating these insights into their strategic planning process, companies can enhance their corporate financial performance (Judge & Douglas, 1998).

The consciousness on climate change and carbon emissions has even increased since the adoption of the Paris Agreement in 2015. As the whole world is affected by the impact of global warming, it is crucial for firms and their stakeholders to align business strategy with sustainability practices and decrease their carbon footprint. Carbon emissions are one of the main determinants of global warming. The current global energy demand is mostly based on fossil fuels which account for the majority of worldwide carbon emissions (Howard-Grenville, Buckle, Hoskins, & George, 2014). Increasing awareness in societies and among governments regarding the pressing need to act against climate change exerts pressure on businesses to be more sustainable and reduce their GHG emissions. Hence, it is important to be aware of the relationship between corporate environmental and corporate financial performance. Environmental innovations are a way for firms to reduce their ecological footprint. This paper, therefore, aims to investigate the moderating role of innovation on the relationship between carbon emissions and a firm’s financial performance. The relationship between corporate environmental performance (CEP)

1_{In the following the terms carbon emissions, carbon dioxide emissions, and greenhouse gas emissions are used}

(8)

2

and corporate financial performance (CFP) has been discussed in recent literature. Prior studies show a positive effect of proactive environmental performance, identified by lower GHG emissions or pollution rates, on corporate financial performance. However, some studies show ambiguous outcomes. This thesis uses four different proxies for CFP to identify the relationship between CEP and CFP in European firms. To measure CEP, carbon emissions are used. Furthermore, the moderating role of innovation in this relationship will be examined. The endogenous growth theory assumes that “technological progress resulting from investment in research and development (R&D) could lead to greater efficiency in production and in the use of natural resources and energy.” (Churchill, Inekwe, Smyth, & Zhang, 2019, p. 30). Innovation is determined as the amount of R&D expenditures, which is directly linked with innovative activity (Thornhill, 2006). Through their R&D expenses, firms can steer their innovative activity and thereby possibly influence the CEP-CFP relationship. This could be the case, as investors value innovative firms more positively, which could in turn offset the effect of higher carbon emissions in the short run.

To the best of the author’s knowledge, this thesis fills a research gap, as no other study has previously used innovation as a moderator in the relationship between carbon emissions and corporate financial performance. A panel dataset is used, based on 238 firms consisting of companies from 11 different economic sectors in 14 European countries over a period of 13 years. Both accounting-based and market-based measures determine CFP. The results show that carbon emissions decrease CFP and that innovation moderates the CEP-CFP relationship. The outcomes have practical relevance as they show that firms can increase their financial performance by enhancing their environmental performance and that innovation does influence this relationship. However, the direction of the moderation effect remains unclear.

The thesis is structured as follows: in chapter two, the literature is reviewed, and the hypotheses are developed. Chapter three gives an overview of the composition of the data, the variables, and the research method. In chapter four, the results are presented before they are discussed in chapter five which also provides the conclusion.

(9)

3

2. Literature Review and Development of the Hypotheses

In the following section, the literature on corporate environmental performance (CEP) and corporate financial performance (CFP) as well as their link to innovation is reviewed. At first, the relationship between CEP and CFP is examined, followed by the impact innovation has on CFP and CEP. The literature review is conducted to present the current state of knowledge regarding the effect of carbon emissions on CFP and to derive the moderating role innovation might play in this relationship. Based on the theoretical insights, two hypotheses are developed.

2.1 Corporate Environmental Performance and Corporate Financial Performance

The relationship between CEP and CFP has already been analyzed in several studies. Often market-based variables are used as proxies for CFP or firm value. Environmental performance is either measured by using the Toxic Release Index (TRI) or by carbon emissions in the following studies. Low carbon emissions indicate a high CEP.

In the majority of the literature, a positive relationship between CEP and CFP has been found. This clarifies that the market rewards firms with a higher CEP while penalizing firms with a lower CEP. Matsumura et al. (2014), for example, identified a negative relationship between carbon emissions and firm value, which implies a positive relationship between CEP and CFP. According to their outcome, every additional metric ton of carbon emission decreases the firm value, measured as the market value of common equity. Their sample consists of US firms from 2006 to 2008. In addition, they found that firms disclosing their emissions have a higher firm value than non-disclosing firms. Firms, however, only disclose if the benefits of doing so outweigh the costs. In the same vein, Liu, Zhou, Yhang, and Hoepner (2016) discovered that carbon emissions are negatively related to the financial performance of a company by using a sample of 62 British firms from 2010 to 2012. Their proxy for financial performance consists of market-based performance measurement variables. In line with Matsumura et al. (2014), they assumed that the market responds negatively to carbon emissions. Additionally, Nishitani and Kokubu (2011) found a negative relationship between a firm’s GHG emissions and its financial performance by analyzing Japanese manufacturing firms. Accordingly, stockholders value a lower amount of carbon emissions as an intangible asset that increases firm value. This view is in line with the previous literature, emphasizing that the market rewards environmental performance. A reason for this is the lower risk of environmental liabilities and the higher profitability of a company with lower GHG emissions (Nishitani & Kokubu, 2011). With a similar approach, Aggarwal and Dow (2011) used a sample of more than 600 firms located in the US, Canada, and Europe and detected that GHG emissions decrease firm value, measured by Tobin’s Q. As their study is based on a geographically wider distribution than previous studies it enables a broader perspective on the CEP-CFP relationship. Their methods, however, contain only cross-sectional data for the year 2008 which leads to a lower validity compared with studies based on panel data including multiple years. Moreover, Lee, Min, and Yook (2015) found a positive relationship between CEP and CFP. They used Tobin’s Q to measure the

(10)

4

market reaction caused by the investors’ responses to changes in environmental performance. The authors found that the negative market reaction resulting from a low CEP is stronger than the positive market reaction resulting from a high CEP. Hence, they conclude that the market penalizes a low CEP more consistently than it rewards a strong CEP. In their study, CEP is defined by carbon emissions. The results illustrate a risk-averse behavior of the market regarding carbon emissions, which is comparable to investors’ general risk-averse behavior.

Emphasizing an alternative perspective, Clarkson, Li, Richardson, and Vasvari (2011) identified a positive relationship between a company’s environmental and financial performance, measured by the accounting-based variables return on assets, operating cash flow, and leverage. Contrary to the previous studies, Clarkson et al. (2011) measured environmental performance by pollution propensity and use it to determine the impact of it on a firm’s financial performance. It can be criticized that their variable pollution propensity is based on the TRI which does not contain carbon emissions – one of the main determinants of climate change. Moreover, their positive relationship is only valid for firms with sufficient management capabilities and financial resources (Clarkson et al., 2011). Additionally, Al-Tuwaijri, Christensen, and Hughes (2004) supported the argument, that good environmental performance is positively related to economic performance, which they measured as an industry adjusted annual return.

King and Lenox (2001) dealt with the question of whether it is profitable for firms to act sustainable, measured as having a high environmental performance. Their environmental performance measurement is based on toxic chemicals, which also indicates that carbon emissions are not included. They identified a significant relationship between pollution reduction and financial performance, measured with Tobin’s Q. Though the direction of the causality remained unclear. Based on their research on 562 US firms from 1987-1996, it remains uncertain whether profitable firms invest more in environmental performance, or whether the relationship is inverse (King & Lenox, 2001). As their sample only covers an outdated period, it is difficult to relate their results to present times.

There are various ways of how environmental performance might influence corporate financial performance. One way is through a firm’s reputation: a good reputation enables a firm to establish better relationships with stakeholders. Good relationships with suppliers, the government, and customers can be created via a higher CEP which leads to a better environmental image (Liu et al., 2016). Besides public perception, CEP also affects a company’s risk profile. Corporate environmental performance mitigates a company’s climate-related risk via lower compliance costs and less uncertainty (Matsumura et al., 2014). In addition, a company’s risk profile is also influenced by environmental liabilities and penalties. A good environmental performance can be considered an intangible value, which increases investors’ confidence and enables a higher firm value (Nishitani & Kokubu, 2011).

(11)

5

To sum up, the studies examined above illustrate that a high CEP in most cases has a positive influence on CFP. This shows that it is profitable for firms to follow environmentally proactive strategies and to decrease their carbon emissions. Supporting environmental performance can, therefore, increase financial performance, according to most of the examined literature.

Although the literature indicates a positive relationship between CEP and CFP, there is still a lack of broad consensus as previous studies are often conducted in the US. One aim of this thesis is to examine the impact of carbon emissions on a firm’s financial performance based on a broad sample of European firms. In theory, a high CEP could increase the CFP for three reasons: if a firm has a strong CEP, this provides it with a favorable public image. Through this reputation, investors and stakeholders realize that the firm can adapt to increasing sustainability requirements in the future. This also provides the company with a competitive advantage. A strong CEP furthermore minimizes possible climate-related fines and compliance costs regarding climate regulations (Matsumura et al., 2014). Also, a strong CEP could lead to a more efficient use of resources in the future and therefore lower corporate spending on those.

In line with the literature on the CEP-CFP relationship as well as theoretical considerations, the following hypothesis is derived:

H1: Carbon emissions are negatively related to a firm’s financial performance. In the next section, the influence of innovation on CFP will be introduced.

2.2 Innovation and Corporate Financial Performance

Innovation plays an important role for businesses, as an indicator of technological progress, economic development, and growth. Innovative ability is crucial for firms to survive in a competitive world with complex and dynamic markets (Assink, 2006). Recent literature assumed a positive effect of innovation on financial performance (Morgan & Berthon, 2008; Kostopoulos, Papalexandris, Papachroni, & Ioannou, 2011). Furthermore, innovation is seen as a key source for the economic growth and competitiveness of a firm (Choi & Lee, 2008), and is positively related to a company’s financial performance (Kostopoulos et al., 2011). McWilliams and Siegel (2000) ascertained that research and development (R&D) is an important determinant of CFP. They analyzed the relationship between CSR and financial performance and found that R&D is crucial in determining the relationship between CSR and financial performance.

In addition to the effect of GHG emissions on CFP, Lee, Min, and Yook (2015) also found a positive relationship between environmental R&D and a firm’s financial performance. Furthermore, they identified that investment in green technology increases a company’s financial performance and reduces carbon emissions at the same time. This aligns two important goals of a company by creating a favorable situation: Increasing the financial performance while decreasing carbon emissions presents a mutually beneficial scenario for a company.

(12)

6

Similarly, Falk (2010) found a positive and significant influence of R&D intensity on both employment and sales growth of Australian firms from 1995 to 2006. He found a decreasing impact of R&D intensity over time. Moreover, Thornhill (2006) confirmed that innovation has a positive impact on a firm’s financial performance, which was measured as revenue growth in this case. As innovation is based on knowledge, a positive relationship between R&D and CFP is in line with the resource-based view, which considers knowledge as a competitive resource for a company (Thornhill, 2006). The resource-based view implies that the rarity, non-tradability, and inimitability of an intangible resource lead to a sustainable competitive advantage. Innovation ability can be defined as an intangible asset comprising the aforementioned characteristics (Cho & Pucik, 2005). Moreover, Prajogo (2006) used the resource-based view to identify the competitive advantage generated by a firm’s innovative activity. A positive relationship between innovation performance and business performance for manufacturing and service firms was examined. Sales growth, market share, and profitability were used as proxies for business performance. To summarize, the studies examined before found an overall positive relationship between innovation and CFP.

2.3 Innovation and Corporate Environmental Performance

In the following, the literature on the relationship between innovation and environmental performance is examined. Compared to the relationship reviewed before, the literature on this relationship is relatively sparse. Specific innovation expenses, for example green R&D expenses, can be used to improve technology while ideally increasing economic performance simultaneously. This relationship is tested in the literature, and green R&D is found to be statistically significant in reducing carbon emissions and thereby improving CEP (Lee & Min, 2015). Next to its positive effect on economic growth, R&D in general is considered as a driver for sustainable development (Fernández López et al., 2018).

While the literature shows that the effect of environmental R&D on carbon emissions is significantly negative (Lee & Min, 2015), there is no consensus in the literature regarding the relation between general R&D and CFP. Blanford (2009) examined the crucial role of technology development for lowering carbon emissions and thereby preventing climate change. According to Blanford (2009), a public R&D subsidies program would be needed as a policy response to cope with climate change which emphasizes the importance of R&D at the political level. Apergis, Eleftheriou, and Panye (2013) analyzed the relationship between R&D and carbon emissions under the mandatory adoption of IFRS standards for a sample of manufacturing firms in Germany, France, and the UK. They found a negative relationship between R&D and carbon emissions after the adoption of the mandatory standards and argue that R&D leads to economic growth, technological change, and an increase in carbon dioxide abatement (Apergis et al., 2013).

Lee and Min (2015) claimed that R&D must be considered when analyzing how environmental performance affects economic performance. The authors distinguished between green R&D and general R&D and created an interaction term of both variables. They found that an increase in R&D expenditures

(13)

7

weakened the negative effect of green R&D on carbon emissions. Therefore, they identified that general R&D expenses, which also include green R&D, hampered the effectiveness of environmental R&D on carbon emissions. This illustrates that a distinction between R&D and environmental R&D could be useful to determine the impact of innovation on CEP.

Petrović and Lobanov (2020) analyzed the impact of R&D expenditures on carbon emissions with a sample of firms consisting of 16 OECD countries from 1981 to 2014. Their results present that R&D expenditures reduce carbon emissions in the long term, which does not apply for around 40% of the countries analyzed. These insights illustrate that the effect of R&D expenses on carbon emissions is inconclusive. Research-based R&D expenditures, such as medical and pharmaceutical research, may demand resources but do not directly contribute to a reduction of carbon emissions (Petrović & Lobanov, 2020).

Also, Churchill et. al (2019) examined the effect of innovation on carbon dioxide emissions. They used a wide time span from 1870 to 2014 to analyze this effect in the G7 countries. Their results showed that the relationship between R&D and GHG emissions is fluctuating over time and that R&D lowers carbon emissions, except for the period between 1955 and 1990. Their findings may help to examine the negative effect of R&D on carbon emissions. However, it is still questionable whether such a large time frame is reliable to find specific results valid for future predictions. Moreover, a period of 35 years showed a positive effect of R&D on GHG emissions. Furthermore, the effect of innovation requires a certain length of time until the desired outcomes are achieved. Additionally, spill-over effects might increase the positive impact of innovation on environmental performance through a wider dispersion into different sectors (Fernández López et al., 2018). However, rebound effects may in turn offset the positive effects of innovation on carbon emissions through a more intense usage of a product or a process. Overall, the literature estimates a predominantly negative impact of R&D on carbon emissions. The relationship between R&D expenditures and carbon emissions has been examined by different scholars. To the best of the author’s knowledge, analyzing the moderating role of innovation on the relationship between carbon emissions and a firm’s financial performance fills a research gap as it has not been studied before. From a theoretical perspective, the endogenous growth theory considers innovation as a driver for more efficient production and use of resources (Churchill et al., 2019). Nevertheless, R&D expenditures do not directly affect a firm’s financial performance, because it takes considerable time for the effects of innovation to unfold. According to the resource-based view, the innovative activity of a company can be considered as an intangible asset that is valued by investors. This results in a higher financial performance, as discussed in Chapter 2.2. Taking into account the previously mentioned literature, investors may put less emphasis on carbon emissions in highly innovative companies compared with carbon emissions in low innovative firms. This could be the case, because they expect highly innovative firms to be more profitable in the long run and to increase their technological growth opportunities. Additionally, this implies the potential for a higher CEP in the

(14)

8

future. This presents a possible explanation, why investors could value carbon emissions less negatively in highly innovative firms. Based on the negative relationship between R&D expenditures and GHG emissions from the literature and the theoretical insights, it is assumed that innovation as a moderator weakens the negative relationship between GHG emissions and CFP. If firms increase their R&D expenditures, it is expected that the negative effect of carbon emissions on CFP decreases. Therefore, the following hypothesis is formulated:

H2: Innovation activity weakens the negative relationship between carbon emissions and corporate financial performance.

(15)

9

3. Research Method

3.1 Data Sample

Panel-data has been chosen because it combines the advantages of cross-sectional and time-series components and allows to analyze multiple variables over multiple periods.

The panel dataset consists of 238 firms in 14 European countries from 11 different economic sectors from 2006 to 2018. Moreover, the dataset is balanced, however, there are missing values for some variables. The year 2006 has been chosen as the starting point because the EU ETS was established, and the Kyoto Protocol was ratified shortly before. The dataset concludes with the year 2018 to ensure that the data reflect current developments. The sample is not restricted to firms within the European Union because also non-EU countries like Switzerland and Norway are included in the data. US dollars are used to compare the financials between countries. A European sample is used because Europe plays a leading role in raising awareness and acting against climate change and because previous research has focused more on American companies. Furthermore, the emission reporting rates in Europe exceed those in the US (Matisoff, Noonan, & O'Brien, 2013).

All data is retrieved from Thomson Reuters Eikon via the Excel add-in Datastream. In Datastream, the data is retrieved from two databases, the corporate social responsibility database ASSET4 and the Worldscope database, which offers a broad range of financial statement data (Reuters, 2013).

Table 1 provides an overview of the firm observations sorted by country. As the table indicates, Great

Britain, France, and Germany are the three countries in the sample with the highest number of firms,

with shares of 33.19%, 13.03%, and 11.76%, respectively. Greece and Portugal are the countries in the sample with the lowest number of firms, with shares of 0.42%, and 1.26% of the total sample. Table 2 presents the economic sectors the firms belong to. As the table indicates, the economic sectors

Industrials and Financials are the sectors with the highest number of company observations, accounting

for 21.01% and 18.07% of the total number. The two sectors, Technology and Telecommunications

(16)

10 Table 1: Firms per Country

Country Code Number of Firms Percent Cum.

BE 5 2.10 2.10 CH 16 6.72 8.82 DE 28 11.76 20.59 DK 6 2.52 23.11 ES 18 7.56 30.67 FI 8 3.36 34.03 FR 31 13.03 47.06 GB 79 33.19 80.25 GR 1 0.42 80.67 IT 11 4.62 85.29 NL 10 4.20 89.50 NO 5 2.10 91.60 PT 3 1.26 92.86 SE 17 7.14 100.00 Total 238 100.00

Table 2: Firms per Economic Sector

Economic Sector Number of

Firms Percent Cum. Basic Materials 30 12.61 12.61 Consumer Cyclicals 33 13.87 26.47 Consumer Non-Cyclicals 19 7.98 34.45 Energy 14 5.88 40.34 Financials 43 18.07 58.40 Healthcare 16 6.72 65.13 Industrials 50 21.01 86.13 NA 1 0.42 86.55 Technology 6 2.52 89.08 Telecommunication Services 10 4.20 93.28 Utilities 16 6.72 100.00 Total 238 100.00

(17)

11

3.2 Variables

3.2.1 Dependent Variables

The dataset consists of four different dependent variables as proxies for corporate financial performance (CFP). Consequently, the dependent variables are divided into two groups, accounting-based and market-based CFP indicators. Accounting-based measures, in general, reflect internal decision-making capabilities and managerial performance (Albertini, 2013), while market-based measurements represent future expectations of CFP. For example, Tobin’s Q contains market-based measures of financial performance and reflects the market reaction, while accounting-based measures like return on assets are more related to a firm’s internal financial performance which is directly retrieved from the balance sheet (Lee, Min, & Yook, 2015). In recent literature, market-based measures are preferred over accounting-based measures, because accounting-accounting-based measures are to a greater extent influenced by managerial decisions (Wagner, 2010). Following Al-Matari, Kaid Al-Swidi and Fadzil (2014), both performance measurements will be used in this paper to reflect past performance and anticipate future performance and to distinguish between internal and external performance to enable a higher reliability of the results. 3.2.1.1 Market-based Performance Measures

Tobin’s Q is widely used as an indicator of the intangible value of a firm because it contains future expectations (Dowell, Hart, & Yeung, 2018) and it also represents long-term market-based CFP with expected future gains (King & Lenox, 2001). Tobin’s Q represents a company’s market value in relation to its replacement costs (Lindenberg & Ross, 1981). There are different methods to derive Tobin’s Q: it can, for example, be computed as the market value of equity plus the book value of liabilities, divided by the book value of assets. In the following analysis, the variable Tobin’s Q is calculated as the logarithm of the market capitalization plus total debt, divided by the total assets of a company, in line with Kuzey and Uyar (2017). The logarithmic transformation is used to provide better scalability of the results, and to bring the distribution of the variable closer to normal distribution.

3.2.1.2 Accounting-based Performance Measures

Return on Assets (ROA), return on equity (ROE), and earnings per share (EPS) are often used as measurements for a firm’s internal financial performance (Albertini, 2013). In the following, ROA2 and

ROE3 are both measured in percent, while the variable EPS is measured in US dollars. They are directly

derived as ratios from Thomson Reuters.

The accounting-based performance measures can be criticized due to their focus on the past (Al-Matari et al., 2014). Two downsides of accounting-based measures are: first, they could be manipulated, and

2_{ROA= (Net Income – Bottom Line + ((Interest Expense on Debt-Interest Capitalized) * (1-Tax Rate))) /}

Average of Last Year's and Current Year’s Total Assets * 100 (Thomson Reuters Eikon, 2020)

3_{ROE= (Net Income – Bottom Line - Preferred Dividend Requirement) / Average of Last Year's and Current}

(18)

12

only rely on a short-term period (Rowe & Morrow, Jr., 1999), and second they rely on accounting principles (Lee & Min, 2015). This way, managers can alter financial statements and mislead stakeholders about the underlying financial position, for example, via earnings management (Healy & Wahlen, 1999).

3.2.2 Independent Variables

The independent variables are carbon emissions and innovation. Carbon emissions are retrieved from the ASSET4 database from Thomson Reuters, which provides environmental, social, and governance (ESG) data. Carbon emissions are defined as the natural logarithm of total carbon dioxide equivalent emissions, measured in metric tons and scaled by net sales. This definition is in line with Busch and Hoffmann (2011). Scaling the variable by net sales allows for better comparability of GHG emissions between companies of different sizes.

To determine innovation, R&D intensity is used. It is defined as the costs of R&D, divided by total assets (Dowell et al., 2018). The R&D expenses as well as total assets are measured in 1000 US dollars. The variable R&D is retrieved from Thomson Reuters, from the Worldscope database. According to its definition, “it represents all direct and indirect costs related to creation of new processes, techniques, applications and products with commercial possibilities” (Thomson Reuters Eikon, 2020).

3.2.3 Control Variables

The following control variables are added to increase the explanatory power of the model and decrease the omitted variable bias. All control variables are retrieved from the Worldscope database of Thomson Reuters and an overview of the variables is given in Table 3.

To begin, Firmsize is a control variable that is expected to influence the dependent variable CFP as well as the independent variables. It is defined as the natural logarithm of total assets, which is measured in 1000 US dollars. The logarithmic transformation is used to bring the distribution closer to the normal distribution. According to Wagner (2010), Firmsize is negatively correlated with Tobin’s Q. Furthermore, Leverage is a predictor for CFP, which is why Leverage is included as a second control variable (Wagner, 2010). It is defined as the debt to equity ratio and measured as the total debt divided by total assets, in line with Clarkson et al. (2011). Assetgrowth is another control variable that may influence CFP. It is defined as the annual growth of total assets in percent. According to Wagner (2010), firm growth is positively related to Tobin`s Q. As a fourth control variable, Capitalintensity is included in the models because it is expected to influence a firm’s financial performance. It is measured as total assets divided by net sales and serves as a predictor for Tobin’s Q (Lee & Min, 2015).

Furthermore, the variable Assetturnover is used to control for the efficiency of a company in generating revenue from its assets (Fairfield & Yohn, 2001). It is measured as annual growth in percentage and is expected to influence a firm’s financial performance, therefore it is included in the models. The variables

(19)

13

years, countries, and economic sectors in order to prevent biased results. Country is used to control for country effects of the firm observations, while Economic sector and Year dummies are used to control for specific effects influencing year and sector (Lee, Min, & Yook, 2015).

Table 3 gives an overview of all variables used in the models, grouped in dependent, independent, and control variables.

Table 3: Variable Description

3.3 Regression Models

To test the impact of carbon emissions on CFP and answer H1 and H2, the following two models are developed.

3.3.1 Model 1

As four different indicators of CFP are used, Model 1 contains four sub-models, each with a different dependent variable of CFP : Model 1.1, with ROA, Model 1.2 with ROE, Model 1.3 with EPS and Model 1.4 with lTobin’s Q as the dependent variable that proxies corporate financial performance (CFP). When looking at the regression equation, the letter i denotes the firm, measured as the ISIN number of a company and t denotes the year. 𝐶𝐹𝑃𝑖𝑡 represents all four performance variables ROA, ROE, EPS and

Tobin’s Q.

Equation Model 1:

𝐶𝐹𝑃𝑖𝑡 = ß0+ ß1𝑙𝐶𝑂2𝑖𝑡+ ß2𝑅𝐷𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖𝑡 + ß3𝐹𝑖𝑟𝑚𝑠𝑖𝑧𝑒𝑖𝑡+ ß4𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖𝑡+

ß5𝐴𝑠𝑠𝑒𝑡𝑔𝑟𝑜𝑤𝑡ℎ𝑖𝑡+ ß6𝐴𝑠𝑠𝑒𝑡𝑡𝑢𝑟𝑛𝑜𝑣𝑒𝑟 + ß7𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖𝑡+ ß8𝑌𝑒𝑎𝑟 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ ß9𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐𝑆𝑒𝑐𝑡𝑜𝑟 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ ß10𝐶𝑜𝑢𝑛𝑡𝑟𝑦 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝜀𝑖𝑡

(20)

14

3.3.2 Model 2

Model 2 is constructed to measure the moderating effect of innovation on the CEP-CFP relationship and to answer H2. In contrast to Model 1, it contains an interaction effect consisting of the variables

CRDintensity and ClCO2 which is defined as CO2RD. Both variables in the interaction term are centered

on their mean, to allow for a better interpretation of the results. Similar to Model 1, Model 2 is divided into four sub-models and each of them contains a different CFP indicator. For reasons of simplicity, the individual sub-models of Model 2 are not shown.

Equation Model 2:

𝐶𝐹𝑃𝑖𝑡 = ß0+ ß1𝑙𝐶𝑂2𝑖𝑡+ ß2𝑅𝐷𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖𝑡 + ß3𝐶𝑂2𝑅𝐷𝑖𝑡+ ß4𝐹𝑖𝑟𝑚𝑠𝑖𝑧𝑒𝑖𝑡+ ß5𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖𝑡+ ß6𝐴𝑠𝑠𝑒𝑡𝑔𝑟𝑜𝑤𝑡ℎ𝑖𝑡+ ß7𝐴𝑠𝑠𝑒𝑡𝑡𝑢𝑟𝑛𝑜𝑣𝑒𝑟𝑖𝑡 + ß8𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖𝑡 + ß9𝑌𝑒𝑎𝑟 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ ß10𝐸𝑐𝑜𝑛𝑜𝑚𝑖𝑐𝑆𝑒𝑐𝑡𝑜𝑟 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ ß11𝐶𝑜𝑢𝑛𝑡𝑟𝑦 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖𝑡+ 𝜀𝑖𝑡

The interaction term aims at testing the moderating role of innovation on the relationship between carbon emissions and CFP. Figure 1 illustrates the moderation effect of innovation in the CEP-CFP relationship.

(21)

15

4. Results

In the following, the results are examined. First, an overview of the descriptive statistics is given, followed by the results of each regression model including tables and graphs. Finally, several robustness checks are conducted.

4.1 Descriptive Statistics

Table 4 illustrates the necessity of using dummy variables, exemplified by the Economic Sector dummies used for Tobin’s Q. The table indicates that different Economic Sectors present different values of Tobin’s Q: the sector Financials has a Median of .322, Consumer Non-Cyclicals has a median value of 1.821. Appendix A and B provide an overview of the variable Tobin’s Q per country and year. All three tables indicate the need to use EconomicSector, Year, and Country dummies as control variables. Table 4: Tobin's Q per Economic Sector

EconomicSector N mean Median sd

Basic Materials 390 1.287 1.016 .877 Consumer Cyclicals 414 1.608 1.03 1.44 Consumer Non-Cyclicals 234 1.897 1.821 .836 Energy 182 1.019 .847 .501 Financials 559 .49 .322 .687 Healthcare 208 2.404 1.642 2.194 Industrials 650 1.169 1.022 .604 NA 12 .251 .238 .104 Technology 78 1.474 1.168 .792 Telecommunications 130 1.112 1.015 .337 Utilities 208 .897 .899 .318

Table 5 gives an overview of the descriptive statistics of all variables. The EconomicSector, Year, and

Country dummies are not displayed. The variable RDintensity has the lowest number of observations

(1430 obs.), while the variable lTobin’s Q has the largest amount of observations with a total of 3065 firms. The variable Tobin’s Q shows a mean of 1.233, indicating that the firms are on average overvalued, while the mean of the logarithmic Tobin’s Q (lTobin’sQ) is -.096.

Furthermore, the four CFP proxies display a high standard deviation, compared to their mean. Specifically, the accounting-based performance variables ROA and ROE show a wide range of variation, with minimum values of -11.73 and -41.43 and a maximum of 33.67 and 90.64 percent. In order to improve the distribution, the variables containing outliers are winsorized at the 1 percent level on both tails of the distribution.

(22)

16 Table 5: Descriptive Statistics

Variables Obs Mean Std. Dev. Min Max

ROA 2990 6.538 6.881 -11.73 33.67 ROE 2938 14.855 17.705 -41.43 90.64 EPS 3003 2.397 4.357 -6.19 26.441 TobinsQ 3065 1.233 1.107 .019 12.209 lTobinsQ 3065 -.096 .834 -3.948 2.502 lCO2 3057 -3.168 1.961 -9.845 2.207 RDintensity 1430 .029 .036 0 .246 Firmsize 3068 16.963 1.871 12.519 22.285 Leverage 3068 .26 .158 0 1.672 Assetgrowth 2977 5.336 15.253 -30.24 86.02 Assetturnover 3003 .732 .558 -.1 4.77 Capitalintensity 3061 4.717 8.133 .343 41.195

Table 6 reports the Pearson correlation coefficient matrix among the variables. The asterisks denote the statistical significance of the correlation results. A correlation coefficient lower than 0.35 represents a low correlation, a value between 0.36 and 0.67 a moderate correlation, and a coefficient higher than 0.68 represents a high correlation (Taylor, 1990). Table 6 indicates that the CFP proxies are positively correlated with each other, which is also expected. To answer H1, the association between carbon emissions and the CFP proxies is important. The table indicates a significant negative correlation between lCO2 and the performance variables ROE, EPS, and lTobin’s Q. Only the correlation between

lCO2 and the performance measure, ROA, is not significant.

Regarding H2, the association between RDintensity and CFP, as well as between CO2RD and CFP is important. RDintensity is positively correlated with three of the four CFP measures at a significance level of .01. Only EPS and RDintensity are not correlated with each other. The coefficient between innovation and EPS is positive, but not significant. There is no consistent relationship between the interaction term CO2RD and the four CFP proxies. For example, ROA is negatively correlated with

CO2RD, while EPS is positively correlated with CO2RD. The performance variables, ROE, and lTobin’s Q are not correlated with the interaction term. According to the classification of Taylor (1990), no

variable is highly correlated with another one. A strong correlation would be an indicator of multicollinearity. However, to test for multicollinearity more precisely a VIF test is performed. The results of the VIF table are displayed in Appendix C. All values are below the critical value of 5.0, which indicates that there is no multicollinearity among the variables in the whole dataset.

(23)

17 Table 6: Pairwise Correlations

Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (1) ROA 1.000 (2) ROE 0.831*** 1.000 (3) EPS 0.206*** 0.297*** 1.000 (4) lTobinsQ 0.629*** 0.461*** 0.036* 1.000 (5) lCO2 -0.023 -0.048*** -0.103*** 0.231*** 1.000 (6) CO2RD -0.050* -0.037 0.089*** 0.030 -0.284*** 1.000 (7) RDintensity 0.188*** 0.123*** 0.012 0.272*** -0.361*** -0.088*** 1.000 (8) Firmsize -0.351*** -0.173*** 0.123*** -0.597*** -0.241*** -0.122*** -0.118*** 1.000 (9) Leverage -0.119*** -0.032* -0.049*** 0.136*** 0.237*** 0.030 -0.306*** 0.021 1.000 (10) Assetgrowth 0.205*** 0.231*** 0.113*** 0.064*** 0.032* -0.010 -0.046* -0.000 0.003 1.000 (11) Assetturnover 0.278*** 0.222*** -0.051*** 0.379*** 0.134*** 0.078*** 0.088*** -0.457*** -0.241*** -0.059*** 1.000 (12) Capitalintensity -0.207*** -0.205*** -0.097*** -0.473*** -0.331*** -0.152*** -0.192*** 0.474*** 0.040*** 0.047*** -0.509*** 1.000 *** p<0.01, ** p<0.05, * p<0.1

(24)

18

4.2 Regression Results

4.2.1 Model 1

In this chapter, the results of all four sub-models of Model 1 are displayed. The sub-models only differ in the dependent variable, as each of them contains a different proxy for CFP. The same theoretical approach is used in every regression model. As the variables RDintensity and lCO2 are mean centered, they are named as CRDintensity and ClCO2 in the following. First, the Hausman specification test is conducted to determine whether a fixed effects or a random effects model fits better. Based on these results the selection of the estimator is made. It is then tested for heteroskedasticity and autocorrelation of the variables. Afterward, the fixed or random effects model is compared with a Pooled OLS estimator by using either the Breusch-Pagan-Lagrange-Multiplier (BPLM) for the random effects or the F-Test if a fixed effects model has been chosen. For the sake of simplicity, not every single test result is displayed. At first, the results of the fixed effects models are explained, followed by the results of the random effects models.

The results of all four sub-models are given in Table 7. For the sub-models 1.1, 1.2, and 1.3, a fixed effects estimator is used, while for the sub-model 1.4 a random effects estimator has been chosen. A fixed effects estimator is used for Model 1.1 because the Hausman specification test is highly significant in rejecting the null hypothesis, which states that the difference in coefficients is not systematic (Appendix D). Moreover, the Wald test indicates heteroskedasticity (Appendix E), and also the Woolridge test for autocorrelation is conducted. As shown in Appendix F, the null hypothesis is rejected, and autocorrelation is found. To cope with autocorrelation and heteroskedasticity in fixed effects estimators, robust standard errors are used in all four sub-models. The BPLM is used to check whether a Pooled OLS estimator is preferred. As the F-test in Appendix G indicates that the fixed effects are non-zero, the use of a Pooled OLS model is rejected. The same procedure has been conducted with the Models 1.2 and 1.3, which use ROE and EPS as a proxy for CFP. The Hausman test indicates to choose a fixed effects model (Appendix H and I) and the F-Test rejects a Pooled OLS estimator as well. For Model 1.4, a random effects estimator has been chosen, because the Hausman test does not reject the null hypothesis (Appendix J). Moreover, the BPLM test rejects the use of a Pooled OLS estimator for Model 1.4.

In all four sub-models, ClCO2 has a statistically significant negative effect on CFP. With a significance level of .01, the highest significance of ClCO2 is given in the Models 1.2 and 1.4, which use ROE and

lTobin’s Q as a proxy for CFP. When ClCO2 increases by one unit, lTobin’s Q decreases by .110 units

in Model 1.4, which supports H1. Regarding CRDintensity, there is no significant relationship between

CRDintensity, and CFP in all four sub-models. According to the results shown in Table 7, the effect of CRDintensity on CFP is negative, except for Model 1.4.

Next, the goodness of fit among the sub-models is compared. The within R-squared, which is of interest for the fixed effects models, indicates that Model 1.1 explains 26.8 percent of the variance of the

(25)

19

dependent variable, whereas Model 1.2 explains 21.7 percent and Model 1.3 only 16.9 percent of the variance of the dependent variable. Model 1.4 has an overall R-squared of 49.0 percent. When looking at the control variables, it is noticeable that the variable Firmsize has a highly significant positive coefficient in Model 1.3, while it is negative in the other sub-models. A negative effect of Firmsize on CFP is expected. Leverage shows a significantly negative coefficient in all four sub-models except for Model 1.4. Moreover, Assetgrowth indicates a positive relationship with CFP, which is significant at the one percent level for all sub-models. Assetturnover shows a highly positive effect on CFP, except for Model 1.4, which displays no significant relationship. The effect of Capitalintensity on CFP is negative in all four sub-models. The negative relationship between Capitalintensity and Tobin’s Q was expected. In conclusion, the four regression sub-models of Model 1 support H1, which states that carbon emissions are negatively related to a firm’s financial performance.

(26)

20 Table 7: Results Model 1

(1.1)

(1.2)

(1.3)

(1.4)

Variables

ROA

(FE)

ROE

(FE)

EPS

(FE)

lTobinsQ

(RE)

ClCO2

-1.924

***

-4.823

***

-0.398

**

-0.110

***

(-3.39)

(-2.91)

(-2.07)

(-4.22)

CRDintensity

-15.670

-33.103

-2.244

0.833 (-0.61)

(-0.52)

(-0.22)

(0.62)

Firmsize

-1.699

*

-2.121

2.244

***

-0.100

**

(-1.89)

(-0.95)

(4.24)

(-2.10)

Leverage

-16.857

***

-23.409

*

-6.757

***

-0.057

(-5.31)

(-1.92)

(-3.18)

(-0.23)

Assetgrowth

0.082

***

0.244

***

0.025

***

0.002

***

(5.23)

(6.03)

(3.14)

(2.93)

Assetturnover

5.255

***

14.369

***

3.134

***

0.074 (2.91)

(2.79)

(3.19)

(0.80)

Capitalintensity

-0.379

-2.190

*

-0.524

*

-0.121

***

(-0.69)

(-1.81)

(-1.67)

(-3.60)

_cons

37.828

**

_54.189

_-34.140

***

_2.565

***

Year

EconomicSector

Country

(2.42)

Yes

No

(1.39)

Yes

No

(-3.74)

Yes

No

(3.10)

Yes

r2_w

0.268

0.217

0.169

0.336 r2_o

0.087

0.032

0.038

0.490 r2_b

0.060

0.004

0.024

0.527 rho

0.733

0.682

0.760

0.732 sigma_u

6.642

17.142

4.359

0.365 sigma_e

4.012

11.711

2.451

0.221

t statistics in parentheses

Year, EconomicSector and Country dummies were not displayed. lCO2 and RDintensity are mean centered, therefore the C is added to the variables. The level of significance is displayed with stars and the t statistics are given in parentheses. FE means fixed effects and RE means random effects.

(27)

21

4.2.2 Model 2

To answer H2, the focus is laid on the interaction term CO2RD in Model 2. The procedure of the model testing is thereby the same as in Model 1. For reasons of simplicity, not every step will be discussed in depth. The results of the four sub-models in Model 2 are displayed in Table 8. According to the Hausman test, the Models 2.1, 2.2, as well as 2.3, are estimated by a fixed effects model and for Model 2.4, a random effects estimator has been chosen. To cope with heteroskedasticity and autocorrelation, robust standard errors are used for each regression model. Moreover, for each of the four sub-models, the use of a Pooled OLS estimator is rejected.

The results of Table 8 indicate a negative relationship between ClCO2 and CFP, which is in line with the regression results of Model 1 (Table 7). The coefficients of ClCO2 for the models using ROA and

ROE as the dependent variable are, however, marginally less negative compared with Table 7.

Concerning the interaction term, CO2RD, three out of four sub-models indicate a moderating role of

CRDintensity on CFP, which is detected by a significant interaction term. The results of the Models 2.1

and 2.4 display a negative relationship, while the results of Model 2.3 indicate a positive effect between the interaction term and CFP. In Model 2.2, no significant relationship was found between CO2RD and

ROE. Hence, the results of the different sub-models show no consistent direction of the interaction

effect. In order to interpret the interaction-effect, the negative coefficient of CO2RD in Model 2.1 indicates that if the variable ClCO2 is at its mean, the variable CRDintensity strengthens the negative effect of carbon emissions on ROA. The same interpretation can be applied to Model 2.4, which uses

lTobin’s Q as a proxy for CFP. Consequently, the Models 2.1 and 2.4 contradict H2, which assumes a

positive interaction effect of the interaction term CO2RD. H2 assumes, that CRDintensity as a moderator mitigates the negative relationship between carbon emissions and CFP. To give a better explanation of the interaction effect, Table 9 represents the marginal effects of carbon emissions, exemplified in this case using ROA. It displays the coefficients of ClCO2 for three different values of CRDintensity: its mean as well as one standard deviation above and below the mean, respectively. With a coefficient of -1.895, the value of ClCO2 at CRDintensity = 0 equals the regression coefficient of ClCO2 of Model 2.1 in Table 8, as it represents the mean value. Figure 2 visualizes the moderating role of CRDintensity on the relationship between ClCO2 and ROA. It provides a graphical depiction of the moderating role of

CRDintensity. The colored lines represent the three different values of the moderating variable CRDintensity from the margins in Table 9. The y-axis represents the dependent variable, ROA, and the

x-axis displays ClCO2 values. The blue line represents the coefficient of CRDintensity one standard deviation below its mean, the red line represents the mean, and the green line represents the coefficient of CRDintensity one standard deviation above its mean. The graph indicates that a higher value of

CRDintensity strengthens the negative slope of the line because the green line has a steeper slope than

the blue line. This indicates that a higher value of CRDintensity strengthens the negative effect of ClCO2 on ROA. These findings contradict H2. Furthermore, the graph shows that the three different

(28)

22

carbon emissions on CFP. Aside from Model 2.1, the interaction effect is also negative in Model 2.4. Hence, the direction of the moderator is the same, which can be seen in Figure 3. As the direction of the relationship between carbon emissions and CFP is similar in Models 2.1 and 2.4, Figure 2 and 3 resemble each other. No graph was created for Model 2.2 because it does not contain a significant interaction effect.

Contrary to the results of Model 2.1 and 2.4, the positive coefficient of CO2RD in Model 2.3 supports H2. It assumes that CRDintensity weakens the negative effect of ClCO2 on CFP. The graphical effect of the positive relationship between ClCO2 and EPS is observed in Figure 4. In contrast to the aforementioned figures, the results indicate that CRDintensity influences the effect of carbon emissions on CFP in the opposite direction. In this particular sub-model, H2 is supported. No statistically significant coefficient of the interaction term is found in Model 2.2. When comparing the goodness of fit between the three fixed effects sub-models, Model 2.1 has the highest within R-squared, followed by Model 2.2 and 2.3. Overall, Model 2.4 offers the highest explanatory power, with an overall R-squared of 48.7 percent. To summarize, the results identify the moderating role of CRDintensity in three out of four sub-models. Nevertheless, the direction of the relationship remains unclear, when using different firm proxies. The three sub-models rejecting H2 have higher explanatory power than Model 2.3, which supports H2. Therefore, H2 is not supported.

(29)

23 Table 8: Results Model 2

(2.1)

(2.2)

(2.3)

(2.4)

Variables

ROA

(FE)

ROE

(FE)

EPS

(FE)

lTobinsQ

(RE)

ClCO2

-1.895

***

-4.748

***

-0.412

**

-0.110

***

(-3.53)

(-3.02)

(-2.08)

(-4.63)

CRDintensity

-12.419

-27.335

-3.407

1.000 (-0.55)

(-0.44)

(-0.32)

(0.86)

CO2RD

-21.585

**

-37.047

7.870

**

-1.326

***

(-2.05)

(-1.11)

(2.43)

(-2.80)

Firmsize

-1.847

**

-2.362

2.307

***

-0.114

**

(-2.01)

(-1.05)

(4.46)

(-2.34)

Leverage

-17.105

***

-23.828

**

-6.691

***

-0.067

(-5.46)

(-2.00)

(-3.15)

(-0.28)

Assetgrowth

0.081

***

0.242

***

0.026

***

0.002

***

(5.15)

(6.00)

(3.17)

(2.85)

Assetturnover

4.641

***

13.289

***

3.353

***

0.042 (2.74)

(2.73)

(3.40)

(0.48)

Capitalintensity

-0.401

-2.230

*

_-0.519

_-0.122

***

(-0.73)

(-1.87)

(-1.65)

(-3.67)

_cons

40.442

**

_58.452

_-35.237

***

_2.710

***

Year

EconomicSector

Country

(2.53)

Yes

No

(1.48)

Yes

No

(-3.95)

Yes

No

(3.22)

Yes

r2_w

0.275

0.219

0.172

0.346 r2_o

0.101

0.036

0.042

0.487 r2_b

0.073

0.005

0.028

0.521 rho

0.721

0.671

0.763

0.737 sigma_u

6.417

16.705

4.386

0.367 sigma_e

3.995

11.697

2.448

0.219

(30)

24 Table 9: Margins of ClCO2 for Model 2.1

Average marginal effects Number of obs = 1,417 Model VCE : Robust

Expression : Linear prediction, predict() dy/dx w.r.t. : ClCO2

1._at : CRDintensity = -.035516 2._at : CRDintensity = 0 3._at : CRDintensity = .035516

Delta-method

dy/dx Std.Err. z P>z [95%Conf. Interval] ClCO2

_at

1 -1.128 0.667 -1.690 0.091 -2.436 0.179 2 -1.895 0.537 -3.530 0.000 -2.948 -0.842 3 -2.662 0.642 -4.140 0.000 -3.921 -1.402

(31)

25 Figure 3: Moderation effect of CRDintensity for Model 2.4

(32)

26

4.3 Robustness Tests

To improve the external validity of the results and to test whether they stay constant under different conditions, two robustness checks are conducted. As four proxies for CFP have already been used in the analysis, no additional CFP proxy is included in the robustness checks. As indicated in Chapter 3, most of the firms are located in Great Britain (GB), with a share of 33.19 percent equaling 79 firms. In order to test the results without the influence of GB, Model 2 is performed again without this country. By doing so, 1027 observations were dropped from the sample, which then consists of the remaining 2067 observations. The results of Model 2 without GB are shown in Table 10, which indicates that ClCO2 stays significant in all four sub-models. In contrast to Table 8, the interaction variable CO2RD is less significant in Models 2.1 and 2.4 but increased in significance in Model 2.3 by p=.01. When performing the regression with GB only, the effect of CO2RD on CFP is insignificant for all four sub-models. The results are displayed in Appendix K. The effect of ClCO2 on CFP in GB is still significant and negative for all sub-models, except for Model 2.4.

A second robustness test only includes the years 2016 to 2018 in order to specifically test the influence of innovation during the most recent years. The results are displayed in Table 11. Regarding the negative effect of carbon emissions on CFP, the sub-models remained significant except for Model 2.3. Surprisingly, none of the sub-models displays a significant coefficient of the interaction term CO2RD. In Models 2.1 and 2.4, the relationship between CO2RD and CFP is negative, while it is positive in the two other sub-models. The results indicate that the time period is too short to receive significant results regarding the interaction term. This furthermore illustrates the necessity to observe the relationships over a longer period.

Altogether, the results of both robustness tests imply that ClCO2 shows a significant negative effect on CFP. No consistent relationship can be found for the moderating role of innovation, as there are contrasting results for the direction of the variable CRDintensity. Hence, the negative effect of carbon emissions on CFP is robust within all sub-models in the robustness tests. To identify a significant moderation effect of CRDintensity, a longer period and the use of multiple countries is required.

(33)

27 Table 10: Results of Model 2 without GB

(2.1)

(2.2)

(2.3)

(2.4)

Variables

ROA

(FE)

ROE

(FE)

EPS

(FE)

lTobinsQ

(RE)

ClCO2

-1.753

***

-3.555

**

-0.477

**

-0.123

***

(-2.86)

(-2.26)

(-2.08)

(-4.58)

CRDintensity

-29.300

-17.538

-2.753

0.378 (-1.15)

(-0.25)

(-0.21)

(0.31)

CO2RD

-18.789

*

-33.058

10.951

***

-1.156

**

(-1.78)

(-0.96)

(3.20)

(-2.35)

Firmsize

-1.707

*

-1.067

2.700

***

-0.109

*

(-1.70)

(-0.43)

(4.23)

(-1.87)

Leverage

-15.762

***

-22.073

-6.600

**

0.001 (-4.12)

(-1.65)

(-2.51)

(0.00)

Assetgrowth

0.078

***

0.231

***

0.026

**

0.002

**

(4.09)

(5.07)

(2.59)

(2.06)

Assetturnover

6.206

***

18.773

***

3.923

***

0.128 (3.02)

(3.10)

(3.03)

(1.11)

Capitalintensity

-0.347

-1.807

-0.817

*

_-0.104

***

(-0.50)

(-1.06)

(-1.90)

(-3.01)

_cons

36.455

**

_28.851

_-41.327

***

_2.553

**

Year

EconomicSector

Country

(2.08)

Yes

No

(0.66)

Yes

No

(-3.75)

Yes

No

(2.31)

Yes

r2_w

0.299

0.247

0.194

0.361 r2_o

0.141

0.099

0.042

0.500 r2_b

0.111

0.057

0.026

0.531 rho

0.707

0.586

0.772

0.748 sigma_u

5.989

13.215

4.912

0.385 sigma_e

3.857

11.104

2.671

0.223

(34)

28 Table 11: Results Model 2 with most recent years (2016-2018)

(2.1)

(2.2)

(2.3)

(2.4)

Variables

ROA

(FE)

ROE

(FE)

EPS

(FE)

lTobinsQ

(RE)

ClCO2

-7.812

***

-15.408

***

-0.285

-0.073

**

(-4.93)

(-2.65)

(-0.69)

(-2.49)

CRDintensity

-167.315

***

-331.981

*

10.208

1.056 (-2.62)

(-1.84)

(0.34)

(0.53)

CO2RD

-10.896

3.306

2.165 -0.867

(-0.56)

(0.07)

(0.28)

(-1.34)

Firmsize

-8.192

***

-36.043

***

2.070

**

-0.154

***

(-2.76)

(-3.54)

(2.05)

(-3.25)

Leverage

-41.997

***

-8.335

-6.238

***

0.832

***

(-3.11)

(-0.53)

(-2.65)

(4.18)

Assetgrowth

0.082

**

0.162

0.010 -0.000

(2.37)

(1.47)

(0.65)

(-0.03)

Assetturnover

-9.394

**

-11.493

0.445 -0.035

(-2.11)

(-1.08)

(0.30)

(-0.29)

Capitalintensity

-0.983

2.597 -0.153

-0.070

*

(-0.83)

(0.96)

(-0.54)

(-1.66)

_cons

169.496

***

_646.760

***

_-29.832

*

_2.682

***

Year

EconomicSector

Country

(3.56)

Yes

No

(3.67)

Yes

No

(-1.70)

Yes

No

(2.94)

Yes

r2_w

0.457

0.197

0.159

0.489 r2_o

0.110

0.018

0.053

0.605 r2_b

0.108

0.018

0.049

0.612 rho

0.948

0.958

0.880

0.915 sigma_u

15.209

49.986

4.873

0.415 sigma_e

3.546

10.460

1.800

0.126

Notes: Year, EconomicSector and Country dummies were not displayed. lCO2 and RDintensity are mean centered, therefore the C is added to the variables. The level of significance is displayed with stars and the t statistics are given in parentheses. FE means fixed effects and RE means random effects.

The impact of innovation on the relationship between carbon emissions and corporate financial performance.