The effect of environmental sustainability on the cost of equity: Evidence from the utility sector1.

The effect of environmental sustainability on the cost of equity:

Evidence from the utility sector

.

Martijn van Oosten

Supervisor: Dr. R.O.S. Zaal MBA University of Groningen

June 2020

This study examines the effect of environmental sustainability on the cost of equity using an international sample of utility firms in the period 2009-2019. Holding a cleaner energy mix is linked with a higher ex-ante cost of equity. Yet, setting the right policies that should reduce emissions in time reduces the cost of equity. These results are risk-driven. Therefore, investors perceive utility firms with more renewable energy sources in the timeframe of this study as riskier which hinders a swift energy transition.

JEL classifications: G32; M14; Q41

Keywords: Environmental sustainability; cost of equity; energy transition

1. Introduction

The consequences of climate change as a result of greenhouse gas (GHG) emissions are materializing. On a global level, the five warmest years, since the start of recording in 1880, occurred from 2015-2019 (NOAA, 2020). In a similar vein, extreme events of precipitation and droughts took place over the world in 2019. Some other examples of the direct effects of climate change are the reduction of agriculture productivity, flooding, and even heat-related morbidity and mortality (IPCC, 2018). To combat these effects the well-known Paris Agreement (UNFCCC, 2015) is made and signed by a vast amount of countries. It is agreed to cut greenhouse gas emissions with a phase that increases average global temperatures to a maximum of two degrees Celsius above pre-industrial levels. Following this initiative, a carbon budget is established2. With this carbon budget, it becomes evident that a swift reduction in GHG emissions is necessary. When analyzing GHG emissions per industry the utility sector is responsible for the lion share of all GHG emissions3. Hence, for utility firms the transition toward renewable energy sources is inevitable.

The ambition of governments and the preferences of a firm’s stakeholders to cut GHG emissions forms indirect risks related to climate change. These risks relate to the transition towards a carbon-neutral economy. Stricter regulation, such as creating and increasing the level of carbon taxes, the risk of stranded assets, litigation risk, and reputational risk what can lead to divestment campaigns of carbon-intensive assets are all examples of these risks. Especially for the utility sector, due to their high contribution to global GHG emissions, these risks are likely to be severe. That utility firms are aware of these risk can be illustrated with the following example. Consider the company E.ON. This company is an energy utility firm from Germany that decided in 2016 to focus more on renewable energy and intelligent networks. E.ON spun off its conventional energy sources in 2016 by creating the company Uniper. In 2018 the stake of E.ON in this new company was sold which completed the transition towards a cleaner utility firm4_{. This transition is also visible in the data where the carbon intensity of E.ON largely} reduced in 2016. The spun-off activities are still operating under a different company and, therefore, there exists no net reduction in GHG emissions. However, with the sale of these activities E.ON mitigated risks related to the transition towards a carbon-neutral economy. In contrast to E.ON, another utility giant from Germany, RWE, is still heavily dependent on coal. As a result, their production process is more carbon-intense, which suggests that RWE faces more risks related to the transition to a carbon-neutral economy.

Utility firms with more commitment towards the energy transition have a higher share of renewable and a lower share of conventional energy sources. If this is a risk-mitigating process, the utility firms with more renewable sources should have a lower cost of equity. The cost of equity is a financial reward that investors get for investing in a firm. With higher levels of risk, 2 _{The carbon budget indicates the number of greenhouse gases that, on a global level, can still be} emitted to reach the maximum temperature increase of two degrees Celsius. Several scenarios to reduce emissions are calculated with different confidence intervals to stay below a temperature increase of two degrees Celsius. All scenarios point in the direction that, on a global level, economies should become carbon-neutral in the second half of this century.

investors face more financial uncertainty which must be compensated with a higher financial reward. As this higher financial reward must be paid by the firm, it increases the cost of equity financing. Hence, from the risk perspective, this may offer the first channel explaining why more environmentally sustainable utility firms have lower costs to attract equity. A second channel, that may answer if environmental sustainability leads to a lower cost of equity, is built on individual preferences and is explained in a model developed by Heinkel et al. (2001). According to this model, ethical screening by investors can influence stock prices, the cost of equity, and even stimulate a firm to adapt to a cleaner production process. Due to the effect of screening, ‘polluting’ firms hold a narrower investor base. Lower demand drops the stock price and increases the cost of equity. Firms with a ‘cleaner’ process have a larger investor base which has an increasing effect on the stock price and decreases the cost of equity. The transformation from a ‘polluting’ firm to a ‘cleaner’ firm is, therefore, a beneficial process due to the increased size of the investor base. If the benefit from transforming to a cleaner production process exceeds the costs of this transformation a firm is financially incentivized to make this transition. This study aims to shed light on the incentive of utility firms, with these two discussed channels, to participate in the energy transition and become more environmentally sustainable. For the risk channel, previous studies find that a higher level of corporate responsibility (El Ghoul et al., 2011; Jo and Na, 2012; Albuquerque et al., 2019) and environmental responsibility (Sharfman and Fernando, 2008; El Ghoul et al., 2018) leads to a lower cost of equity. In a different study, Hong and Kacperczyk (2009) find that stocks related to sin industries are subject to ethical screening and outperformed the market. The latter finding supports the channel explained by Heinkel et al. (2001).

In the next section, related literature is summarized from which the hypotheses are created. Next, the variables are explained, the empirical model is discussed, and summary statistics are given. Then, the results are shown with additional robustness checks. Thereafter, this study and its results are discussed and the conclusion remains.

2. Literature review

As this study links environmental sustainability with the cost of equity financing, it is necessary to explain corporate responsibility first. The theory and empirical evidence linking corporate responsibility with the cost of equity is applicable for environmental responsibility and sustainability as well.

2.1 Corporate responsibility

A simple definition of corporate responsibility would be sacrificing profits for the common good. This immediately launches a debate if firms should engage in corporate responsibility at all because sacrificing profits is value-destroying. Indeed, the conventional view is built by neoliberal thinkers and states that firms should opt for profit maximization on behalf of the shareholders (Friedman, 1970). In the search for efficiency, firms should behave in a self-interested way. Whenever negative externalities affect moral standards, shareholders who benefited from corporate profits could engage in charities to deal with these externalities. Yet, this is not a task for firms. Firms that do decide to act more responsible are serving the interest of managers. These managers fulfill their own interests as they are looking for self-esteem or have a personal affection with responsible entrepreneurship. This behavior is a costly diversion of the firm’s resources and is paid by the shareholders. Given these arguments, the conventional view is that any form of corporate responsibility destroys firm value.

Empirical evidence is found that higher corporate responsible firms are less risky what leads to higher market values (Jo and Na, 2012; Albuquerque et al., 2019). Thus, corporate responsibility is not simply a waste of a firm’s resources but can be value-enhancing. Yet, this does not explain why the level of corporate responsibility is heterogeneous among firms. If corporate responsibility is value-enhancing, every firm should opt for the level of responsibility at the point where benefits of risk reduction equal the costs of acting responsibly. At this level, a firm likely meets most regulations and stakeholder values so that the chances of negative events occurring are mostly reduced. This reasoning implies that the risk channel cannot explain why some firms act more responsible than others. However, this can be explained by heterogeneous preferences of individuals. It is shown that the political preferences of individuals play a major role in the question to what extent firms should engage in corporate responsibility (Di Giuli and Kostovetsky, 2014). From the investor’s perspective, the demand for corporate responsibility should influence a firm’s decision to become more or less responsible as it affects the impact on firm value (Mackey et al., 2007; Dam and Scholtens, 2015). Evidence explains that investor’s preferences do affect stock prices (Fama and French, 2007) and stock returns (Hong and Kacperczyk, 2009). In this latter study, stock related to the sin industries alcohol, tobacco and, gambling outperformed the market during the period 1962– 2003. This latter finding states that ethical screening by investors affects the cost of equity. More controversial firms experience higher costs to attract equity financing. A study that directly links the cost of equity financing with corporate responsibility finds a negative relationship (El Ghoul et al., 2011).

2.2 Environmental sustainability

If risk is the driver of the link between environmental sustainability and firm value, more environmental utility firms should hold a lower cost of equity. Studies linking environmental performance with the cost of equity studies using realized returns to capture the cost of equity find a negative relationship (Sharfman and Fernando, 2008; Trinks et al., 2017). Trinks et al. (2017) use GHG intensity (GHG emissions relative to sales) to capture the actual reduction of GHG pollution. Sharfman and Fernando (2008) focus on U.S. firms and use data of environmental concerns and strengths from KLD. Additionally, no link has been found between the level of environmental risk management of a firm and the number of institutional investors. Similarly, evidence exists that firms with a higher level of environmental responsibility experience a lower ex-ante cost of equity (Chava, 2014; El Ghoul et al., 2018; Gupta, 2018). Chava (2014) links the cost of equity and the cost of debt, using bank loan data, to environmental strengths and concerns from KLD. Firms with environmental strengths cannot expect to attract funds via equity or debt for lower costs. Also, some evidence is found that firms with more environmental concerns have a smaller investor base. Interestingly, this also holds for firms with clean energy. As a robustness check, Chava (2014) shows that environmental concern does not affect the cost of equity via the channel of increased default risk. El Ghoul et al. (2018) use a dataset of environmental costs of a firm and show a positive relationship between these costs and the cost of equity for manufacturing firms in 30 countries. This relationship is significant before and after the financial crisis. During the financial crisis (2007 and 2008) the effect between environmental costs and the cost of equity is positive but insignificant. When moving more towards environmental sustainability, Kim et al. (2015) exploit changed regulation by the South Korean government to find a positive effect between carbon intensity and the cost of equity of South Korean firms from the period of 2007 to 2011. This change in regulation required designated firms to disclose their carbon emission. Additionally, a lower effect on the cost of equity is found for high carbon-emitting firms. This is explained due to the less severe regulation by the South Korean government for heavily polluting industries to make them more competitive to foreign industries.

An additional driver for the negative relationship between environmental responsibility or sustainability and the cost of equity of a firm can be explained by the force of the ‘investor base’ (Heinkel et al., 2001). Similar to corporate responsibility, investors hold heterogeneous preferences regarding a firm’s level of environmental responsibility. Individual investors can engage in ethical screening of polluting firms when this matches their personal believes. Also, institutional investors may have constraints to financing polluting activities. If the share of these investors is significant enough the demand for the shares of ‘greener’ utility firms rises and of ‘polluting’ firms drops. Hence, ‘greener’ firms have a larger investor base because they don’t experience ethical screening. The investor base of ‘polluting’ firms is smaller because some investors withhold from buying these stocks. As a result, this demand effect results in lower costs of equity financing for the ‘greener’ firms and higher costs for the ‘polluting’ firms. From the above-discussed topics, it is suggested that utility firms that act more environmentally sustainable reduce their risks, and have access to a larger investor base. Therefore, the first hypothesis that this study tests is:

2.3 Effect over time

As discussed, utility firms face the challenge to transform their production processes more sustainably. To realize the Paris Agreement (UNFCCC, 2015), this directly means cutting greenhouse gases. To accomplish the targets set in this agreement, politics move forward with setting concrete goals. For example, the European Commission developed the ‘Green deal’ which states that European economies should be carbon neutral in 2050 (European Commission, 2019). As a consequence, utility firms experience risks related to this mandatory transition. This requires utility firms to set pollution reduction targets, reports annual emissions, and invest in renewable energy sources. Hence, within the timeframe of this study (2009-2019) most risks can be mitigated by setting the right policies. In time, these policies should materialize in a cleaner production process. As a result, a mix of policies which reflect ‘future environmental sustainability’ mostly affect the risk channel:

(2a) The negative link between future environmental sustainability and the cost of equity is stronger than the link between current pollution levels and the cost of equity.

Additionally, it is expected that the risks related to the mandatory energy transition have risen during 2009-2019. The signature of the Paris agreement and its implementation by governments are key moments in the period of this study. This advocates that the regulatory framework stimulates, increasingly more, renewable energy sources and penalizes conventional sources. To test whether environmental sustainability increased over time the following sub-hypothesis is developed:

(2b) Over time the negative relationship between environmental sustainability and the cost of equity has strengthened.

As mentioned above, more risks related to the energy transition came existent in the period of this study. Therefore, backward looking data do not fully incorporate these risks. To be more precise, when estimating the cost of equity using realized returns risks related to the energy transition are not incorporated. When using an ex-ante cost of equity based on analysts’ forecasts, energy transition-related risks are absorbed in the cost of equity. If this holds a mismatch between the ex-post and ex-ante cost of equity is visible, and the effect of environmental sustainability on the cost of equity is stronger when using the cost of equity based on analysts’ forecasts:

(2c) The effect of environmental sustainability on the ex-ante cost of equity is stronger than the effect of environmental sustainability on the ex-post cost of equity.

2.4 Mitigating risk

‘polluting’ firms, are more environmentally sustainable, less risky, and have higher market values. The choice to become a ‘green’ firm and more environmentally sustainable than ‘normal’ firms mitigates less risk than the process of going from a ‘polluting’ to ‘normal’ firm. This is for the reason that most uncertainties regarding the transition to a carbon-neutral economy are already dealt with. This train of thought is supported by the findings that no relationship is found between a firm’s environmental strengths and the cost of equity while it does for a firm’s environmental weaknesses (Chava, 2014). Also, institutional investors seem to avoid both ‘polluting’ and ‘green’ firms (Fernando et al., 2010). Similarly, Sharfman and Fernando (2008) find no relationship with environmental risk management and the share of institutional ownership. For corporate responsibility similar results arise. Oikonomon (2010) finds that for responsible firms there exists only a weak link between corporate responsibility, for irresponsible firms this effect is much stronger.

Given these results, it is expected that the effect between environmental sustainability and the cost of equity is stronger for the dirtier utility firms than for the greener utility firms.

(3) The effect of environmental sustainability on the cost of equity is stronger for utility firms with a low environmental sustainable profile than for utility firms with a high environmental

sustainable profile.

2.5 Long-term investors

So far, the discussion on the link between environmental sustainability and the cost of equity focused on the risk channel. As theorized by Heinkel et al. (2001) the demand for ‘greenness’ by investors can also affect the cost of equity. It suggests that green utility firms can attract a larger investor base which reduces the cost of equity. However, green investors are hard to identify. Previous studies show that the share of institutional ownership is not higher and may even be lower for ‘green’ firms (Sharfman and Fernando, 2008; Fernando et al., 2010; Chava, 2014). Also, divestment from ‘polluting’ stock is not a popular strategy among institutional investors to address concerns related to climate risks, but they prefer a dialogue (Krueger et al., 2020). Similarly, Bushee (1998) shows that a significant share of institutional investors behaves myopic and sacrifices long-term profits to meet short-term goals. These findings point in the direction that institutional investors cannot be seen as the green investors from Heinkel’s (2001) model. Starks et al. (2018) find that long-term orientated investors are more likely to hold stock with a high Environmental, Social, and Governance (ESG) profile. And, these stocks are sold less after a negative earnings report. Hence, long-term orientated investors can function as a proxy for green investors to test if the share of green investors has an additional effect on the cost of equity. It is predicted that:

3. Data and methodology

3.1 Sample construction

To construct the sample linking environmental sustainability to the cost of equity financing several databases are employed. (1) Sustainalytics provided data on GHG intensity and a scoring matrix indicating how clean the energy mix of a utility firm is. From (2) Asset4 data on GHG intensity and the score that captures a firm’s current environmental sustainability plus policies to become more environmentally sustainable are collected. (3) Institutional Brokers Earnings Services (I/B/E/S) provides share prices, earnings forecasts, and growth forecasts. At last, from (4) Eikon’s datastream all other financial data are collected. This study uses an international sample of utility firms in the period 2009-2019 with 1580 firm-year observations. Table 1 summarizes the composition of the sample based on the different years and the different sub-industries within the utility sector. For space issues, the breakdown of the sample by country is reported in Table B.1 of Appendix B.

Table 1. Sample breakdown by year and sub-industries

Year Observations Percentage

2009 130 8.23 2010 138 8.73 2011 145 9.18 2012 154 9.75 2013 167 10.57 2014 179 11.33 2015 190 12.03 2016 202 12.78 2017 188 11.90 2018 24 1.52 2019 63 3.99 Total 1580 100

Sub-industry Observations Percentage

Electric utilities 650 41.14

Gas utilities 182 11.52

Water utilities 122 7.72

Multi-utilities 341 21.58

Renewable power production 112 7.09

Independent power production and traders

173 10.95

3.1 Cost of equity 3.1.1 Ex-ante

To test hypotheses 1, 2a, 2b, 3, and 4 the ex-ante cost of equity is used as the dependent variable. In estimating the ex-ante cost of equity this study closely follows El Ghoul et al. (2011, 2018). This technique takes the average ex-ante cost of equity from four different models based on analysts’ forecasts of earnings per share and growth rates. Two of these models are based on residual income valuation (Gebhardt et al., 2001; Claus and Thomas, 2001) and two models based on abnormal earnings growth (Ohlson and Juettner-Naurorth, 2005; Easton, 2004). The ex-ante cost of equity is found by using numerical techniques that search for the rate that sets the valuation equation equal to the current share price. All values below 0 and above 1 are discarded to handle outlier problems and to deal with unrealistic values. Average values are only taken when at least two out of four models have values to prevent that some data points are based on one model. Appendix A provides the technical overview of each model and the assumptions made to calculate the ex-ante cost of equity in each of these models.

Using the ex-ante cost of equity holds several advantages compared to the cost of equity based on realized returns and is, for this reason, used to test most hypotheses. The ex-post cost of equity is built on the assumption that newly provided information has no net effect on the expected returns (Elton, 1999). Especially since this study covers a period where the regulatory framework regarding environmental sustainability has become stricter, this is an unrealistic assumption. Additionally, it is shown that realized returns are a poor proxy for expected returns, and hence the cost of equity (Fama and French, 1997). Table B.2 in Appendix B displays the correlation between the four different models and the average ex-ante cost of equity.

3.1.2 Ex-post

As an alternative to the ex-ante cost of equity, two different approaches are used to estimate the ex-post cost of equity. The ex-post cost of equity is used to test hypothesis 2c. This method is similar to the studies of Sharfman and Fernando (2008) and Trinks et al. (2017). To estimate the ex-post of equity monthly data in the prior 10 years of return indices are used. These return indices are used to compute the monthly returns of all utility firms in this study. The risk-free rate, which is the yield on a 10 year U.S. government bond, is subtracted to create monthly excess returns. With the first approach, the prior 10 years of monthly returns are explained with one factor. This factor is the difference between the return on the market portfolio and the risk-free rate. This process results in a beta for each firm for each month. The ex-post cost of equity is calculated using this prior 10 years of monthly betas, excess market returns, and the risk-free rate. With the second approach, the prior 10 years of excess returns are explained by the five factors of Fama and French (2015). Next to a firm’s beta, the size of a firm, it’s market-to-book value, profitability, and investment patterns are employed5.

In Appendix B, Table B.3 shows the correlation between the ex-ante cost of equity and the two approaches estimating the ex-post cost of equity. The correlation between the average ex-ante cost of equity and the Fama and French (2015) five-factor model is 0.0587, for the one-factor

model this is 0.1119. These low coefficients raise the argument that including the asset pricing models is not simply a robustness check. But, it can answer whether results differ when using a backward-looking model instead of forward-looking models.

3.2 Environmental sustainability

As stated, this study investigates the link between both the realized environmental sustainability and the ambition to become more sustainable with the cost of equity. Hence, variables are required that reflect the sustainability of the current energy mix and variables that reflect a set of policies that show the ambition to become more sustainable in time.

3.2.1 GHG Intensity

The first variable to capture the current environmental sustainability is the number of greenhouse gases emitted over the sales of a firm. This is the so-called ‘GHG intensity’ of a firm and is even more relevant for utility firms due to the urgency of creating low-carbon production processes. In the process of creating this variable, data of self-reported GHG emissions are collected from Sustainalytics and the Asset4 database. Both data sources supply scope 1 and scope 2 emissions. Scope 1 emissions are direct emissions from activities controlled by the firm. Scope 2 are indirect emissions from energy or electricity purchased by the firm. Scope 3 emissions are excluded in this study. Scope 3 emissions are indirect emissions from downstream activities and tend to be harder to measure. Also, they are mostly outside the control of a firm. In most industries scope 3 GHG emissions are the greatest share of the carbon footprint. However, due to the characteristics of the utility sector scope 1 and 2 GHG emissions are more important than scope 3 emissions. Thus, the focus on the utility sector offers more reliable results than the scenario where all industries are included. The following gases are included; CO2, CH4, N2O, HFCS, PFCS, SF6, and NF3. Emissions are expressed in tons of CO2 equivalents.

3.2.2 Energy mix

A second estimate for the environmental sustainability of the current energy mix, the energy mix score, is provided by Sustainalytics. This data is unique as it is only relevant for (energy) utility firms. This data is a scoring matrix ranging from 0 to 100. A value of 0 indicates that a utility firm has the most ‘dirty’ process to generate energy. When a firm uses clean renewable energy only to generate energy this score is 100. The data on the generation mix ranges from 2009 till 2019.

3.2.3 Emission score

Next to these two variables, this study uses the emission score retrieved from the Asset4 database. The emission score is one of the elements of the environmental pillar score. The environmental pillar score is a widely used matrix to determine a firm’s level of risk related to environmental issues. One of the components creating the emission score is the GHG intensity, which is the same estimate as this study uses. The other components are waste generated divided by sales, biodiversity, and environmental management systems. The score ranges from 0 to 100 where a score of 100 indicates that a firm is highly environmentally sustainable6. The score is built with data going beyond emissions intensity and takes into account many policies set, related to environmental sustainability, such as how a firm deals with carbon-related risks. Hence, it includes current emission intensities and a mix of policies that reflect future environmental sustainability.

3.3 Long-term investors

To see if long-term orientated behave like the ‘green’ investors in the model of Heinkel et al. (2001) a proxy variable for the share of long-term investors is needed. This variable is self-created by taking, on a daily basis, the volume traded on the stock market divided by the market capitalization of the corresponding firm. Then, all values within a year are averaged. This variable is created with the assumption that long-term investors trade less frequently and is named LongInv. As this is an imperfect way to capture long-term investors, two additional proxies are created for robustness. The second variable is called LongInv2 and is created as follows; for every firm in every year, the three worst trading days are located. On these days and the day after, the volume traded over the market capitalization is taken. The average of these six days is used to create this variable. The last method to capture long-term investors is similar to the previous method but corrects for the size of the drop in share price on these worst trading days. This last variable is named LongInv3. The variables LongInv2 and LongInv3 are created with the assumption that long-term investors are less likely to start trading after the firm experienced a drop in market value.

6 _{For more information on the methodology of the Emission score:}

3.4 Control variables

To control for a list of known factors affecting the cost of equity, several control variables are included. These are the volatility of stock returns, the natural logarithm of the assets, market over book value, leverage ratio, the inflation rate, and the bias in analysts’ earnings per share forecast. The volatility is a proxy for a firm’s level of systematic and non-systematic risk. Higher volatility levels relate to a more risky firm. As investors need to be compensated for higher levels of risk, volatility increases the cost of equity. Fama and French (1992) find a negative relationship between the size of a firm and its market-to-book value. The natural logarithm of the total assets accounts for the size of a firm. The market-to-book value is the market value of a firm divided by its book value. It is expected that higher levels of debt are linked with more risk. Again, additional risk increases the cost of equity. The leverage ratio, measured as total debt over total assets, is included to control for this. The inflation rate is included since the earnings per share forecasts are expressed in domestic currency and nominal terms. Easton and Sommers (2007) show the presence of optimism in analysts’ forecasts. To control for this noise, an additional control variable is added. This is the forecasted earnings per share minus the realized earnings per share, then, this value is divided by the share price on the 30th of June of the corresponding year.

3.5 Empirical model

To test the hypotheses 1,2a-c, and 3, the following model is employed:

(1) 𝐶𝑜𝐸𝑖𝑡 = 𝛼 + 𝛽 𝐺𝐻𝐺 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦𝑖𝑡+ 𝛾 𝐸𝑛𝑒𝑟𝑔𝑦 𝑚𝑖𝑥𝑖𝑡+ 𝛿 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛 𝑠𝑐𝑜𝑟𝑒𝑖𝑡+ 𝜃 𝑋𝑖𝑡+ 𝑐𝑜𝑢𝑛𝑡𝑟𝑦, 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦, 𝑎𝑛𝑑 𝑦𝑒𝑎𝑟 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜀𝑖𝑡

Here i indexes the different utility firms, t represents the different years. When testing hypotheses 1, 2a, 2b, and 3 the CoE reflects the ex-ante cost of equity based on analysts’ forecasts. For hypothesis 2c, CoE is the cost of equity based on realized returns. GHG Intensity is the greenhouse gases emitted over sales, and if the hypothesis 1 holds, should have a positive sign. An alternative measure for the sustainability of a utility firm’s production process, Energy mix is used which in case hypothesis 1 holds should have a negative sign. Similarly, the Emission score, which reflects both current environmental sustainability and future environmental sustainability, is predicted to have a negative sign. To test hypothesis 2a, the parameters 𝛽 and 𝛾 are compared with the parameter 𝛿. For hypothesis 2b, Eq. (1) is tested using two subsamples. The first subsample ranges from 2009-2014, the second from 2015-2019. For hypothesis 3, the sample is divided by the median values of GHG Intensity, Energy mix, and Emission score.

this is negative, for Leverage this is positive, for Market-to-book this is negative, for Inflation positive, and for Bias a positive sign is predicted. Country, sub-industry, and year dummies are employed to account for unobserved heterogeneity across these entities and over time. Including clustered errors corrects for correlation between errors on the firm-level which is likely given the structure of the panel data. For this reason, robust standard errors clustered at the firm level are used following El Ghoul et al. (2011, 2018).

To test hypothesis 3 and answer whether the effect of environmental sustainability on the cost of equity increases when a firm has more long-term orientated shareholders the following models are deployed:

(2𝑎) 𝐶𝑜𝐸_𝑖𝑡 = 𝛼 + 𝛽 𝐺𝐻𝐺 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦_𝑖𝑡+ 𝛾 𝐿𝑜𝑛𝑔𝐼𝑛𝑣_𝑖𝑡+ 𝛿 𝐿𝑜𝑛𝑔𝐼𝑛𝑣_𝑖𝑡 ∗ 𝐺𝐻𝐺 𝐼𝑛𝑡𝑒𝑛𝑠𝑖𝑡𝑦_𝑖𝑡 + 𝜃 𝑋_𝑖𝑡+ 𝑐𝑜𝑢𝑛𝑡𝑟𝑦, 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦, 𝑎𝑛𝑑 𝑦𝑒𝑎𝑟 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜀_𝑖𝑡 (2𝑏) 𝐶𝑜𝐸𝑖𝑡 = 𝛼 + 𝛽 𝐸𝑛𝑒𝑟𝑔𝑦 𝑚𝑖𝑥𝑖𝑡+ 𝛾 𝐿𝑜𝑛𝑔𝐼𝑛𝑣𝑖𝑡+ 𝛿 𝐿𝑜𝑛𝑔𝐼𝑛𝑣𝑖𝑡 ∗ 𝐸𝑛𝑒𝑟𝑔𝑦 𝑚𝑖𝑥𝑖𝑡 + 𝜃 𝑋𝑖𝑡+ 𝑐𝑜𝑢𝑛𝑡𝑟𝑦, 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦, 𝑎𝑛𝑑 𝑦𝑒𝑎𝑟 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜀𝑖𝑡 (2𝑐) 𝐶𝑜𝐸𝑖𝑡 = 𝛼 + 𝛽 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛 𝑠𝑐𝑜𝑟𝑒𝑖𝑡+ 𝛾 𝐿𝑜𝑛𝑔𝐼𝑛𝑣𝑖𝑡+ 𝛿 𝐿𝑜𝑛𝑔𝐼𝑛𝑣𝑖𝑡 ∗ 𝐸𝑚𝑖𝑠𝑠𝑖𝑜𝑛 𝑠𝑐𝑜𝑟𝑒𝑖𝑡+ 𝜃 𝑋𝑖𝑡 + 𝑐𝑜𝑢𝑛𝑡𝑟𝑦, 𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦, 𝑎𝑛𝑑 𝑦𝑒𝑎𝑟 𝑓𝑖𝑥𝑒𝑑 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 + 𝜀𝑖𝑡

Eq. (2a-c) each use a different variable for environmental sustainability. This allows testing if the share of long-term investors affects the link of any of the environmental sustainability estimates with the cost of equity. The variable LongInv is a proxy for the share of long-term investors. Hypothesis 3 predicts that, for Eq (2a), the parameter 𝛿 is negative. For the Eq. (2b) and Eq. (2c) this parameter is predicted to be positive.

4.5 Descriptive statistics

positive correlation between Emission score and Size indicates that larger utility firms are more environmentally sustainable with their current production process and policies they have set. None of the estimates of environmental sustainability hold a significant correlation with the self-created proxy for the share of long-term investors. This is an indication that long-term investors and ‘green’ investors are not similar.

Table 2. Descriptive statistics

Variable N Mean Median Standard

deviation Min. Max. GHG Intensity 1073 2.8121 1.3320 3.9141 5.45 * 10^-5 36.5054 Energy mix 982 28.1161 0 35.7395 0 100 Emission score 1580 54.3422 55.245 28.3174 0.38 99.72 CoE ex-ante 1580 0.1036 0.0920 0.0533 5.72 * 10^-6 0.5704 CoE 1f ex-post 1257 0.0381 0.0312 0.0377 -0.0380 0.2597 CoE 5f ex-post 1257 0.0707 0.0619 0.0498 -0.1509 0.3397 Volatility 1580 0.2500 0.2143 0.1211 0.0789 1.0114 Size 1580 16.3747 16.3863 1.2132 11.9091 19.7099 Leverage 1580 0.3831 0.3691 0.1484 0.0000 0.7958 Market-to-book 1580 1.6809 1.4800 0.9148 0.1200 6.7700 Inflation 1580 0.0233 0.0179 0.0212 -0.0174 0.1553 Bias 1580 0.0027 0.0003 0.0472 -0.3355 0.3921 LongInv 1244 0.0029 0.0022 0.0034 9.76 * 10^-7 0.0761 LongInv2 1214 0.0050 0.0034 0.0082 4.56 * 10^-7 0.2044 LongInv3 1208 0.4411 0.3192 0.4823 2.91 * 10^-5 7.6351 LongInv * GHG intensity 823 0.0069 0.0022 0.0130 1.80 * 10^-8 0.1793

LongInv * Energy mix 752 0.0731 0.0000 0.1669 0 3.0817

LongInv * Emission score 1244 0.1586 0.1080 0.1724 2.49 * 10^-5 2.7208 LongInv2 * GHG Intensity 803 0.0104 0.0031 0.0239 2.03 * 10^-8 0.4816

LongInv2 * Energy mix 735 0.1334 0 0.4665 0 9.7532

LongInv2 * Emission score 1214 0.2578 0.1596 0.3868 3.58 * 10^-5 7.3105 LongInv3 * GHG Intensity 799 1.0590 0.3086 2.2506 1.41 * 10^-6 36.3259

LongInv3 * Energy mix 730 10.7708 0 22.2166 0 234.6942

LongInv3 * Emission score 1208 24.4278 14.1954 29.0331 0.0013 273.1084

GHG Intensity is the ratio of a firm’s emitted greenhouse gases to sales, Energy mix is a matrix reflecting

the sustainability of the energy mix, Emission score reflects current pollution levels and policies set to reduce them, CoE ex-ante is the ex-ante cost of equity, CoE 1f ex-post is the cost of equity based on realized returns and the beta, CoE 5f ex-post is the cost of equity based on realized returns and the five factors of Fama and French (2015), Volatility is measured using past five years of stock returns, Size is the logarithm of assets, Leverage is the ratio of debt to assets, Market-to-book is the ratio of market value to book value, Inflation is the realized inflation level, Bias is the error in analysts’ forecast relative to the share price, LongInv, LongInv2, and LongInv3 are all self-created proxies for the share of long-term investors. LongInv * GHG Intensity, LongInv * Energy mix, LongInv * Emission score, LongInv2

Table 3. Correlation matrix of the variables of interest and the control variables. Bold coefficient indicate a significant correlation on the 1% level. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (1) GHG Intensity 1.0000 (2) Energy mix -0.3405 1.0000 (3) Emission score -0.2343 -0.0064 1.0000 (4) Volatility -0.0002 0.0460 -0.2637 1.0000 (5) Size -0.0252 -0.1488 0.4129 -0.3280 1.0000 (6) Leverage -0.0401 -0.0097 -0.0053 0.1913 0.1166 1.0000 (7) Market-to-book -0.1044 0.0839 0.0230 -0.0897 -0.1579 0.1010 1.0000 (8) Inflation -0.0100 0.1170 -0.1231 0.1887 -0.0672 -0.0510 -0.0609 1.0000 (9) Bias -0.0094 -0.0688 0.0672 -0.0069 0.0869 -0.0304 -0.0580 0.0070 1.0000 (10) LongInv -0.0210 -0.0267 -0.0508 0.1764 -0.0484 0.1132 0.1571 -0.0880 -0.0334 1.0000 (11) GHG intensity * LongInv 0.4385 -0.2604 -0.1156 0.1072 0.1033 0.1848 -0.0825 -0.0677 0.0050 0.6427 1.0000 (12) Energy mix * LongInv -0.1904 0.5396 0.0293 0.0544 0.0650 0.0588 0.0009 0.1063 0.0315 0.5505 -0.0468 1.0000 (13) Emission score * LongInv -0.0771 -0.0234 0.4199 -0.0017 0.2788 0.1824 0.0005 -0.1259 0.0761 0.6714 0.5734 0.3774 1.0000

GHG Intensity is the ratio of a firm’s emitted greenhouse gases to sales, Energy mix is a matrix reflect the sustainability of the energy mix, Emission score

reflects current pollution levels and policies set to reduce them, Volatility is measured using past five years of stock returns, Size is the logarithm of assets,

Leverage is the ratio of debt to assets, Market-to-book is the ratio of market value to book value, Inflation is the realized inflation level, Bias is the error in

analysts’ forecast relative to the share price, LongInv reflects the share of long-term investors. LongInv * GHG Intensity, LongInv * Energy mix, and LongInv

4. Results

4.1 Preliminary analysis

Before continuing with testing the hypotheses some preliminary analyses are discussed. All results are reported in Appendix B. Table B.4 shows the lack of homoskedasticity of the residuals, this addresses the urgency of using robust errors. In Fig. B.1 it is made visual that the ex-ante cost of equity, the dependent variable, is not normally distributed. Table B.5 shows a formal test to confirm this. In Table B.5 it is also reported that the residuals are not normally distributed. The Q-Q plot is presented in Fig. B.2 and shows similar results. These issues are addressed in the section of the robustness checks, where it is shown that they do not affect the results. Table B.6 shows the Woolridge test for first-order serial correlation. Each of the panels shows the results with Eq. (1) and a different variable of interest. When using GHG intensity to capture environmental sustainability first-order serial correlation is detected. To deal with this issue, clustered errors are employed at the firm level.

4.2 Univariate analysis

The univariate tests compare the mean value of utility firms with a high and low profile of environmental sustainability. As discussed, this study uses three estimates for environmental sustainability. All of these estimates are used in the univariate analyses and shown in Table 4. Panel A shows the analyses based on the GHG intensity and panel B based on the sustainability of the energy mix. Both variables reflect the sustainability of a firm’s current operations. In panel A, utility firms with a lower GHG intensity have an average ex-ante cost of equity of 10.81%. This value is higher than the average ex-ante cost of equity for the utility firms with a higher GHG intensity (10.07%). This finding suggests that utility firms with a cleaner production mix have higher costs to attract equity. The difference is statistically significant at the 5% level. Similar results arise in panel B. Cleaner utility firms have a higher ex-ante cost of equity. The utility firms with a dirtier energy mix have a 0.80% lower ex-ante cost of equity than the utility firms with a cleaner energy mix. This difference is statistically significant on the 1% level.

4.3 Multivariate analysis

Table 5 shows the results when testing hypotheses 1 and 2a. Model 1 shows the ex-ante cost of equity regressed on all control variables. Riskier firms, proxied with the volatility, experience a higher cost of equity. In this sample, no evidence is found that the size of a firm affects the cost of equity. Higher market value to book value decreases the cost of equity. Utility firms with more leverage experience a higher cost of equity. The inflation rate increases the cost of equity as expected. There is no evidence found that a positive analysts’ bias increases the cost of equity.

In model 2 the GHG Intensity is included as the independent variable of interest. The parameter of the GHG Intensity is negative. Ceteris paribus, a utility firm with 1 more tCO2 equivalent emissions per 1 million U.S. dollar has a 0.15% lower cost of equity. This finding is significantly different from 0 at the 5% level. This finding is different from what is predicted in hypothesis 1. It suggests that utility firms who pollute more greenhouse gases relative to their sales have lower costs of equity financing. Continuing with model 3, the Energy mix is used to capture environmental sustainability. In hypothesis 1, a higher score decreases the cost of equity. Similar to model 1 the sign is the opposite of what is predicted. Yet, the coefficient is insignificant at all levels. If the Emission score is used as the independent variable of interest different results arise. ‘Greener’ utility firms, based on current pollution intensities and a mix of policies to reduce pollution levels, have a lower cost of equity. The coefficient is significantly different from 0 at the 5% level. The economic significance is only modest. Ceteris paribus, a utility firm with the maximum score possible has a 1% lower cost of equity than a firm with the lowest possible score. When comparing models 1-4, it is visible that most variation in the

ex-Table 4. Univariate test of means between different subsamples. The cost of equity is the ex-ante cost of equity. All samples are divided using the median value. Low GHG intensity, High Energy mix, and High Emission score all indicate a ‘greener’ utility firm.

Observations Cost of equity

Panel A. Low GHG intensity 675 0.1073 High GHG intensity 676 0.1016 Difference 0.0057 t-Statistic 1.7655** Panel B.

High Energy mix 627 0.1132

Low Energy mix 690 0.1052

Difference 0.0080

t-Statistic 2.4424***

Panel C.

High Emission score 843 0.0992

Low Emission score 844 0.1091

Difference -0.0099

t-Statistic 3.7780***

ante cost of equity is explained by model 2. This model shows an adjusted R-squared of 43.11% compared to 30.22% when only the control variables are used. Model 3 and 4 explain less variation of the ex-ante cost of equity and are barely an improvement compared to model 1. Overall, the models 2-4 suggest that hypothesis 1 does not hold. When considering the current environmental sustainability of a utility firm, a higher level of sustainability is linked with a higher cost of equity.

Table 5. All models use the ex-ante cost of equity as the dependent variable. GHG Intensity and Energy

mix reflect the environmental sustainability of the current production processes. Emission score reflect

both the current emission intensities and the policies set to reduce them. All models hold country, sub-industry, and year fixed-effects. Robust standard errors are employed at the firm level.

(1) (2) (3) (4) (5) (6)

Control variables

GHG Intensity

Energy mix Emission score GHG Intensity & Emission score Energy mix & Emission score GHG Intensity -0.0015** (0.0007) -0.0015** (0.0007) Energy mix 0.0001 0.0001* (0.0001) (0.0001) Emission score -0.0001** (0.0001) -0.0002** (0.0001) -0.0002*** (0.0001) Volatility 0.0763*** 0.0805** 0.0852*** 0.0712*** 0.0526 0.0769*** (0.0210) (0.0345) (0.0259) (0.0214) (0.0334) (0.0255) Size -0.0000 0.0018 -0.0012 0.0011 0.0034 0.0031 (0.0019) (0.0022) (0.0032) (0.0019) (0.0023) (0.0035) Leverage 0.0469*** 0.0640*** 0.0775*** 0.0458*** 0.0470*** 0.0587*** (0.0128) (0.0154) (0.0163) (0.0126) (0.0141) (0.0178) Market-to- book -0.0067*** (0.0024) -0.0083*** (0.0025) -0.0087*** (0.0032) -0.0066*** (0.0024) -0.0061** (0.0026) -0.0080** (0.0034) Inflation 0.4797*** 0.2741** 0.3616** 0.4775*** 0.4550*** 0.4382** (0.1274) (0.1261) (0.1585) (0.1280) (0.1589) (0.1801) Bias 0.0100 -0.0154 0.0665 0.0124 -0.0350 -0.0086 (0.0506) (0.0490) (0.0581) (0.0502) (0.0601) (0.0645) Constant 0.0935*** -0.0064 0.0965* 0.0862** 0.0638 0.0429 (0.0350) (0.0776) (0.0548) (0.0344) (0.0456) (0.0611)

Year effects Yes Yes Yes Yes Yes Yes

Sub-industry effects

Yes Yes Yes Yes Yes Yes

Country effects

Yes Yes Yes Yes Yes Yes

Clustered errors

Firm Firm Firm Firm Firm Firm

Observations 1,580 1,305 1,232 1,580 1,073 982

Adjusted R-squared

0.3022 0.4311 0.3444 0.3040 0.3669 0.3004

Model 5 and 6 allow testing hypothesis 2a. As explained in Table 3 there is no issue of multicollinearity between the Emission score and either the GHG Intensity or Energy mix. Therefore, these variables can be used together in one model so that GHG Intensity and Energy mix capture the current environmental sustainability. The Emission score reflects mostly policies related to future environmental sustainability since the emission intensity is already captured by the other variables. The results in model 5 suggest that similar to model 2, a higher GHG Intensity is linked with a lower cost of equity. When a utility firm shows the ambition to reduce GHG emissions in time, by using a set of policies, the cost of equity decreases. The latter finding has more economic significance than in model 4. Both coefficients are significantly different from 0 at the 5% level. From a general point of view, model 6 holds similar findings to model 5. A cleaner energy mix is linked with a higher cost of equity. Ceteris paribus, a utility firm with dirtiest energy mix has a 1% lower cost of equity than a firm with the cleanest energy mix. This coefficient is significantly different from 0 at the 10% level. The Emission score has a similar coefficient as in model 5. This result is significantly different from 0 at the 5% level and states that, ceteris paribus, a utility firm with the highest score possible has a 2% lower cost of equity than a firm with the lowest possible score. Hypothesis 2a holds, future environmental sustainability is linked with a lower cost of equity, current environmental sustainability is linked with a higher cost of equity. Although, for current environmental sustainability this positive link is unexpected.

4.4 Analysis over time

To test hypothesis 2c, the dependent variable is replaced by the ex-post cost of equity. Here the cost of equity is based on ex-post returns. It is predicted that, for this reason, the effect of environmental sustainability is not fully incorporated into the ex-post cost of equity, yet.

Table 6. The dependent variable is the ex-ante cost of equity. GHG Intensity and Energy mix form indicators for a utility firm’s current level of environmental sustainability. Emission score reflect both current emission intensities and a mix of policies set to reduce them. The variables of interest are shown in two different periods. All models use country, sub-industry, and year fixed-effects. The standard errors are robust and clustered on the firm level.

(1) (2) (3) (4) (5) (6) GHG Intensity 2009-2014 GHG Intensity 2015-2019 Energy mix 2009-2014 Energy mix 2015-2019 Emission score 2009-2014 Emission score 2015-2019 GHG Intensity -0.0017** (0.0007) -0.0016* (0.0009) Energy mix 0.0001 0.0000 (0.0001) (0.0001) Emission score -0.0002** (0.0001) 0.0001 (0.0001) Volatility 0.0580* 0.1399*** 0.0952*** 0.1682*** 0.0759*** 0.1092*** (0.0328) (0.0476) (0.0247) (0.0489) (0.0214) (0.0386) Size 0.0025 0.0010 0.0001 -0.0019 0.0028 -0.0012 (0.0029) (0.0025) (0.0036) (0.0037) (0.0025) (0.0020) Leverage 0.0587*** 0.0556*** 0.0970*** 0.0572*** 0.0604*** 0.0282 (0.0219) (0.0177) (0.0222) (0.0193) (0.0157) (0.0184) Market-to- book -0.0069** (0.0033) -0.0072** (0.0028) -0.0091** (0.0040) -0.0061 (0.0039) -0.0058** (0.0026) -0.0067** (0.0032) Inflation 0.2513 0.1291 0.1040 0.3332** 0.2723 0.3557** (0.2747) (0.1165) (0.3169) (0.1345) (0.1971) (0.1372) Bias -0.0795 0.0376 0.0572 0.1427*** -0.0064 0.0711 (0.0748) (0.0420) (0.0801) (0.0475) (0.0650) (0.0568) Constant 0.0785 0.0013 0.0791 0.0525 0.0768* 0.0731 (0.0529) (0.0799) (0.0615) (0.0678) (0.0434) (0.0456)

Year effects Yes Yes Yes Yes Yes Yes

Sub-industry effects

Yes Yes Yes Yes Yes Yes

Country effects

Yes Yes Yes Yes Yes Yes

Clustered errors

Firm Firm Firm Firm Firm Firm

Observations 665 640 653 579 913 667

Adjusted R-squared

0.3660 0.5533 0.2921 0.4933 0.2817 0.3880

Models 1-3 of Table 7 use the cost of equity based on ex-post returns explained by a firm’s beta. The control variables Inflation and Bias are dropped since they are irrelevant for the ex-post cost of equity.

Table 7. All models are regressed on the ex-post cost of equity and use ex-post returns to estimate the cost of equity. In models 1-3, the last 10 years of monthly returns are explained by a firm’s beta. For model 4-6, the last 10 years of monthly returns are explained by a firm’s beta, size, market-to-book ratio, profitability, and investment patterns. GHG Intensity and Energy mix reflect the current environmental sustainability of a utility firm. The Emission score includes current emission intensities and policies set to reduce them. All models hold country, sub-industry, and year fixed-effects. Robust errors are used clustered at the firm level.

(1) (2) (3) (4) (5) (6) GHG Intensity Energy mix Emission score GHG Intensity

Energy mix Emission score GHG Intensity 0.0001 (0.0003) -0.0002 (0.0004) Energy mix -0.0001** -0.0000 (0.0000) (0.0000) Emission score -0.0001* (0.0000) -0.0000 (0.0000) Volatility 0.0879*** 0.0930*** 0.0950*** 0.0603*** 0.0853*** 0.0936*** (0.0223) (0.0167) (0.0159) (0.0174) (0.0185) (0.0199) Size -0.0000 -0.0009 0.0001 (0.0013) (0.0013) (0.0014) Leverage 0.0152 0.0301*** 0.0194** 0.0043 0.0210** 0.0258** (0.0094) (0.0105) (0.0097) (0.0111) (0.0096) (0.0113) Market-to-book -0.0009 (0.0011) -0.0019 (0.0012) -0.0014 (0.0012) Constant 0.0170 0.0415* 0.0157 0.1004*** 0.0270*** 0.0693*** (0.0251) (0.0251) (0.0267) (0.0141) (0.0089) (0.0171)

Year effects Yes Yes Yes Yes Yes Yes

Sub-industry effects

Yes Yes Yes Yes Yes Yes

Country effects

Yes Yes Yes Yes Yes Yes

Clustered errors

Firm Firm Firm Firm Firm Firm

Observation s 1,181 1,073 1,375 1,191 1,089 1,393 Adjusted R-squared 0.5944 0.6210 0.6324 0.5603 0.5301 0.5039

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

similar negative coefficient is found, in model 3, for the Emission score. This result is significantly different from 0 at the 10% level. This points in the direction that utility firms with lower emission intensities and more policies set to reduce these emissions hold a lower ex-post cost of equity. The economic significance of both the Energy mix and Emission score is similar, ceteris paribus, a utility firm with the highest possible score yields a 1% lower cost of equity than a firm with the lowest possible score.

To continue with models 4-6 of Table 7. Here, the ex-post returns are explained by the five factors of Fama and French (2015). Next to a firm’s beta, its size, market-to-book value, profitability, and investment patterns explain the last 10 years of monthly stock returns. The control variables Size and Market-to-book are dropped since they are included in the dependent variable. In contrast to models 2 and 3, no evidence is found that environmental sustainability affects the ex-post cost of equity. All coefficients of the variables that capture environmental sustainability are insignificant on all levels. Overall, the findings of the models 1-6 suggest that some evidence exists that a higher level of environmental sustainability of utility firms leads to a lower ex-post cost of equity. However, this effect is incorporated in both or either the size of a firm and its market-to-book value. Also, with the results of Table 7 it is clear that hypothesis 2c does not hold.

4.5 Risk mitigation

In hypothesis 3 it is predicted that the effect of environmental sustainability on the cost of equity is stronger for the dirtier utility firms than for the cleaner firms. Table 8 shows the results where this hypothesis is tested. The sample is divided, in each model, based on the median value of GHG Intensity, Energy mix, and Emission score. Models 1, 3, and 5 reflect the dirtiest 50% of the sample, models 2, 4, and 6 the cleanest 50% based on the corresponding estimate for environmental sustainability.

4.6 Long-term investors

With hypothesis 4 it is tested whether the effect of environmental sustainability increases when more long-term orientated shareholders are attached to a firm. This sub-section tests this hypothesis and the results are reported in Table 9. GHG Intensity is the proxy for environmental sustainability in model 1, for model 2 this is Energy mix, and for model 3 this is Emission score. All these models include a proxy for the share of long-term investors, LongInv, and the

Table 8. All models use the ex-ante cost of equity as the dependent variable. GHG Intensity and Energy

mix reflect the environmental sustainability of a firm’s current operations. Emission score include both

current emission intensities and a mix of policies set to reduce them. For each of the three variables of interest the median value is taken two create two subsamples. Hence, models 1, 3, and 5 represent the more dirty utility firms. Models 2, 4, and 6 samples the cleaner utility firms. All models use country, sub-industry, and year fixed-effects. Robust errors are employed clustered on the firm level.

(1) (2) (3) (4) (5) (6) GHG Intensity dirty firms GHG Intensity clean firms Energy mix dirty firms Energy mix clean firms Emission score dirty firms Emission score clean firms GHG Intensity -0.0009 (0.0007) 0.0067 (0.0097) Energy mix 0.0003 (0.0004) 0.0002** (0.0001) Emission score -0.0002 (0.0001) -0.0002 (0.0001) Volatility -0.0026 0.2118*** 0.0411 0.1351*** 0.0517** 0.1252** (0.0247) (0.0633) (0.0372) (0.0395) (0.0227) (0.0615) Size -0.0006 0.0035 -0.0041 0.0044 -0.0022 0.0059* (0.0021) (0.0031) (0.0029) (0.0053) (0.0027) (0.0032) Leverage 0.0650*** 0.0890*** 0.0824*** 0.0868*** 0.0522*** 0.0546*** (0.0241) (0.0273) (0.0211) (0.0230) (0.0185) (0.0149) Market-to- book -0.0056 (0.0044) -0.0116*** (0.0034) -0.0069* (0.0040) -0.0088** (0.0040) -0.0081** (0.0036) -0.0071** (0.0031) Inflation 0.0927 0.3953** 0.1302 0.3480* 0.3809** 0.6344*** (0.1776) (0.1545) (0.2569) (0.1997) (0.1502) (0.2303) Bias -0.0743 -0.0105 0.0652 0.0410 0.0035 0.0523 (0.0667) (0.0755) (0.0726) (0.0824) (0.0614) (0.0747) Constant 0.1295** -0.0133 0.1484*** -0.0249 0.1758*** -0.0083 (0.0627) (0.0728) (0.0473) (0.0983) (0.0417) (0.0605)

Year effects Yes Yes Yes Yes Yes Yes

Sub-industry effects

Yes Yes Yes Yes Yes Yes

Country effects Yes Yes Yes Yes Yes Yes

Clustered errors

Firm Firm Firm Firm Firm Firm

Observations 653 653 653 579 790 790

Adjusted R-squared

0.4157 0.5022 0.2865 0.4288 0.2726 0.3850

interaction term of this variable with the corresponding environmental sustainability proxy. In the models 1, 2, and 3 all these variables show insignificant results at all levels.

Table 9. All models use the ex-ante cost of equity as the dependent variable. GHG Intensity and Energy

mix represent the current environmental sustainability of the production process. Emission score reflect

both current emission intensities and a mix of policies set to reduce them. LongInv is a self-created proxy for the share of long-term investors. An interaction term is made between each variable that captures environmental sustainability and the share of long-term investors. All models have country, sub-industry, and year fixed-effects. Robust errors are used clustered on the firm level.

(1) (2) (3)

GHG intensity & Long-term investors

Energy Mix & Long-term investors

Emission score & Long-term investors GHG Intensity -0.0013 (0.0011) Energy mix -0.0000 (0.0001) Emission score -0.0001 (0.0001) LongInv 0.9533 0.7611 0.0806 (1.6858) (1.2510) (1.3186) GHG Intensity * LongInv -0.1874 (0.2647)

Energy mix * LongInv 0.0108

(0.0177)

Emission score * LongInv -0.0092

(0.0262) Volatility 0.0715** 0.0772*** 0.0701*** (0.0357) (0.0291) (0.0231) Size 0.0008 -0.0029 0.0009 (0.0022) (0.0027) (0.0018) Leverage 0.0615*** 0.0727*** 0.0413*** (0.0164) (0.0167) (0.0133) Market-to-book -0.0063** -0.0071** -0.0060** (0.0025) (0.0030) (0.0024) Inflation 0.2201* 0.3230** 0.4434*** (0.1306) (0.1560) (0.1211) Bias -0.0296 0.0549 -0.0109 (0.0506) (0.0592) (0.0518) Constant 0.0323 0.1293*** 0.0889*** (0.0723) (0.0471) (0.0341)

Year effects Yes Yes Yes

Sub-industry effects Yes Yes Yes

Country effects Yes Yes Yes

Clustered errors Firm Firm Firm

Observations 1,046 994 1,244

Adjusted R-squared 0.4442 0.3575 0.3125

For space issues, similar results are shown in Table B.8.A and Table B.8.B in Appendix B. In these tables, different self-created proxies for the share of long-term investors are used, LongInv2 and LongInv3 respectively. Also, the interaction terms are shown similarly to Table 9. No new results arise. With all variables in this study used for both environmental sustainability and the share of term investors, no evidence is found that the share of long-term investors affects the link between environmental sustainability and the cost of equity for utility firms. Therefore, hypothesis 4 does not hold.

4.7 Robustness

So far the ex-ante cost of equity based on the average of four models (Gebhardt et al., 2001; Claus and Thomas, 2001; Easton, 2004; Ohlson and Juettner-Nauroth, 2005). In Table C.1A., Table C.1.B, and Table C.1.C. of Appendix C, the hypotheses 1 and 2a are tested when the ex-ante cost of equity is based on each of the models separately. It Table C.1.A, models 1-4 test the first hypothesis with the GHG Intensity as the variable of interest. Only when the cost of equity is estimated with the model of Gebhardt et al. (2001) a significant, on the 10% level, a negative coefficient is reported. Similarly, in models 5-8 there is no plain proof that higher GHG intensities and policies set to reduce emission intensities reduce the ex-ante cost of equity. In Table C.1.B, the Energy mix is used to proxy current environmental sustainability. None of the models support hypothesis 1. Only model 8 finds evidence for hypothesis 2a. At last, the Emission score is used as the variable of interest in Table C.1.C. Here only hypothesis 1 is tested. Two out of four models fail to report a significant result. In model 2, where the ex-ante cost of equity is based on Ohlson and Juettner-Nauroth (2005) a negative relationship is found between the cost of equity financing and environmental sustainability. In model 3, where the model of Claus and Thomas (2001) is adapted, a positive relationship is found. Overall, these findings fail to make the results more robust. However, they also stress the importance of using the average of the four models as results can be driven by one model.

In the preliminary results, it is reported that both the ex-ante cost of equity and the residuals of model 4 in Table 5 are not normally distributed. For this reason, the top 1% (5%) of the ex-ante cost of equity is stripped from the sample. The hypotheses 1 and 2a tested again and reported in Table C.6.A (C.6.B). In Table C.6.A, the results keep indicating that a higher level of current environmental sustainability is linked with a higher ex-ante cost of equity. A mix of policies that intend to reduce the emission intensities in the future is linked with a lower ex-ante cost of equity. In Table C.6.B., three out of four models find evidence that cleaner utility firms experience higher costs to attract equity. In both tables, the test for normality of the dependent variable and residuals is reported in panels A and B. By stripping the top values of the ex-ante cost of equity most of the non-normality issues are dealt with. Also, the non-normality in the residuals were driven by the highest values of the ex-ante cost of equity. The results in both tables relax the concerns related to both the non-normality in the dependent variable and the residuals.

There exist some endogeneity issues in this study. One of these issues could be that the results are driven by simultaneity or reversed causality. Hence, lower or higher costs to attract equity could affect the level of environmental sustainability of a firm. To address these issues, a simple technique is used and reported in Table C.7. Hypothesis 1 is re-tested using the one-year lagged and one-year lead variables of GHG Intensity, Energy mix, and Emission score. If the effects found in this study flow from environmental sustainability to the cost of equity, the one-year lagged variable should report much more convincing evidence than the one-year lead variables. For the variable GHG Intensity both the one-year lagged and lead variable report significant results. For the Energy mix no significant results are found similar to the results in Table 5. For the variable Emission score a significant result is found for the lagged variable and not for the one-year lead variable. Thus, only when the Emission score is used as a proxy for environmental sustainability it is found that the results are not driven by simultaneity or reversed causality.

5. Discussion

cost of equity. This is contradicting the model of Heinkel et al. (2001). Additionally, no evidence is found that long-term investors affect the link between environmental sustainability and the cost of equity. Either long-term investors are a poor substitute for ‘green’ investors or, again, the effect of the investor base on the cost of equity is minimal.

This leaves risk to be the main driver to explain the positive link between current environmental sustainability and the negative link between future environmental sustainability with the cost of equity. Hence, holding a clean energy mix is perceived as risk increasing while reducing emissions in time is perceived as risk reducing. It suggests that, in the period of this study, producing energy from renewable sources is riskier than using conventional sources. The fact that future environmental sustainability is perceived as risk reducing implies that in time risks related to the energy transition become more severe such that conventional sources hold more risk than renewables. This reasoning implies that investors perceive most energy transition relates risks to occur in the future and delayed the transition of utilities in the timeframe of this study. The decreasing link between GHG intensity and the cost of equity is also contrasting other literature. Trinks et al. (2017) find a positive link between GHG intensity and the cost of equity based on ex-post returns using an international dataset. Using an ex-ante cost of equity, Kim et al. (2015) find a similar positive link between GHG intensity and the cost of equity for South Korean firms. The findings of this study suggest that, for the utility sector, this positive link does not hold. Further, environmental responsibility is linked with a lower ex-ante cost of equity (Chava, 2014; El Ghoul et al., 2018; Gupta, 2018). This is again in contrast with the negative link between GHG intensity and the ex-ante cost of equity in this study. However, when focusing on future environmental sustainability the results are in line with previous literature.

The contrasting findings between current and future environmental sustainability open an interesting topic for future research. Since a current clean energy mix is perceived as risk increasing while setting policies to enhance future environmental sustainability is thought of as risk-reducing, there should be a date in the future where a cleaner energy mix is perceived to reduce risk as well. Furthermore, collecting and reporting GHG emissions of firms is still a young practice. Future research can, therefore, rely on more data. The techniques to measure the emission of greenhouse gases are likely to advance as well. This improves future research as the reliability of reported GHG emissions increases.

Following this analysis another explanation appears for the contrasting results in this study. As shown in Table C.5. of Appendix C, when the GHG intensity is measured using the market value, utility firms with a lower GHG intensity experience a small reduction in their costs of attracting equity. This is the opposite effect than in the scenario where either the total amount of sales or common equity is used to capture GHG intensity but is in line with hypothesis 1 of this study. Then, it is possible that investors do take the environmental sustainability of utility firms into account to mitigate risk or for individual preferences. However, these investors capture the GHG intensity using the market value which results in this finding. It is obvious that the questions remain why the contrasting findings appear when using the different methods to capture GHG intensity. And, what eventually would be the best method to measure GHG intensity. This leaves an interesting topic for future research and debate.

This study fails to show the main evidence on all robustness checks. For example, the technique of taking an average value to estimate the ex-ante cost of equity drives the results. On an individual level, the four different techniques to estimate the ex-ante cost of equity fails to show results leading to one direction. Previous studies show more robust evidence on this matter but can rely on 12,195 (El. Ghoul et al., 2011) and 7,122 (El Ghoul et al., 2018) firm-year observations. The finding that a current cleaner energy mix leads to a higher cost of equity is driven by the smaller firms of the used sample. This may also suggest that, for the bigger utilities, having a sustainable energy mix is not perceived as riskier. Also, when the Energy mix is used to capture the current sustainability of the energy mix the evidence is less convincingly. Again, with more data available, future research can define whether the results in this paper continue to hold.