Master Thesis MSc. Finance
Systematic risk factors in the returns of
energy related stocks
Influence of the environmental focus across countries and over time
Jochum YskaUniversity of Groningen
Abstract
This study investigates the impact of companies’ environmental focus on the influence of systematic risk factors and on their returns. By using the rolling regression method the time variability of the influence of risk factors is studied in six different regions. The findings show that energy related stock returns, in general, are significantly influenced by technology stock returns, oil price changes and the general market return. The sensitivity per risk factor depends on a company’s choice for renewable or non-renewable energy, the region under consideration and the time period. When financial markets are in distress, the influence per risk factor is likely to change. Furthermore, there is significant evidence that European electricity producers’ returns are positively influenced by using more than the country’s average of renewable inputs. Of the European companies indirectly related to the production of electricity, only the solar sector’s returns are significantly negatively affected by an increase in the share of solar power in a country’s total electricity generation.
Student number: S3274012
Supervisor: Prof. dr. W. Bessler
Program: MSc. Finance
Date: 07-06-2018
Word count: 17,145
2
1. Introduction
Sparked by worldwide awareness about the environmental effects of the use of fossil fuels, the demand for renewable energy has seen a rapid increase in recent years. Bloomberg New Energy Finance (2017)1 states that the global investments in renewables in 2016 totaled 241.6 billion dollars. The total annual amount of money invested has decreased compared to 2015 as the result of lower costs per megawatt. The amount of annually added capacity, however, has increased to 138.5 gigawatts (GW). On a global scale, 138.5 GW is 55% of all generating capacity added in 2016. Investments have increased the relative share of renewables at total global power generation over the past years. In 2007 renewables accounted for 5.2% of the worldwide power generation and this has increased to 11.3% in 2016.
The growing share of renewables in global energy generation displays the increasing importance of this sector. The increasing importance has caused an inflow of investments in previous years. Eyraud et al. (2013) investigate the factors influencing direct investments in renewable energy. They find that economic growth, interest rates and fuel prices positively affect investments in renewables. From the different policies countries can adopt, feed-in-tariffs and carbon pricing schemes have a positive impact on investments. The increase in popularity of renewable energy and the increase in investments resulted in an increase in the number of companies operating in the renewable energy sector.
Not only are there are there many of new entrants, the large traditional energy companies have started to split their operations into a traditional energy branch and a renewable energy branch. Examples are the German energy suppliers RWE and E.ON2. The German split indicates that the energy industry is at a crossroads, where it can choose between focusing on renewable energy3 or non-renewable energy4. Choosing for renewable energy is considered highly environmentally focused while in contrast the choice for non-renewable energy is considered a sign of a low environmental focus. Renewable energy stocks (hereafter RES) are therefore assigned to the former category and traditional energy stocks (hereafter TES) are assigned to the latter.
Different studies analyze the specifics of energy stocks, both TES and RES. Since the awareness of the need to increase renewable energy has become apparent only more recently, the literature investigating RES is scarce compared to TES. The number of studies regarding RES has increased rapidly in recent years, however. The studies find report results. Ortas and Moneva (2013) find outperformance of renewable equity indices compared to other general indices. Anderloni and Tanda (2017) find no long term performance differences between RES and TES. Bohl et al. (2013) find outperformance of RES between 2004 and 2007, however, between 2008 and 2011 RES show signs of underperformance. Rezec and Scholtens (2017) compare the returns of fourteen renewable energy indices, consisting of stocks from different continents, with the returns of their benchmarks. The majority of the renewable energy indices underperform their benchmarks, while some of them outperform. Their general conclusion is that the poor performance makes the renewable energy indices an unattractive investment.
The existing literature has also looked into the systematic risk factors influencing the returns of RES and TES. Early studies have investigated the influence of energy price changes on the returns of energy companies in general. Different studies find that energy stocks show
1 New Energy outlook (2017). Retrieved on March 9, 2018 from Bloomberg: https://about.bnef.com/new-energy-outlook/
2 Eon and RWE pursue radical restructurings (2016). Retrieved March 20, 2018 from FT.com: https://www.ft.com/content/316ce884-1cdc-11e6-a7bc-ee846770ec15.
3
Renewable energy is used interchangeably with alternative energy.
3
a more pronounced reaction to oil price changes than the general stock market does. Gormus et al. (2015) and Sadorsky (2012) find that both TES and RES react to changes in the oil price. The effect of these changes is more pronounced for the TES than it is for RES. Kumar et al. (2012) explain that RES react positively to changes in oil prices because of the higher costs it imposes to TES, leading investors to search for alternatives such as RES. Bohl et al. (2013) use different multifactor models to investigate what influences the returns of German RES. They find that increasing global competition and overcapacity in the solar sector weigh heavily on the profitability of RES. The returns of German RES react positively to the price momentum factor in times of positive returns and negatively to the same factor in times of negative returns. With a market beta of two, the shares are considered riskier than the general stock market. Bohl et al. (2015) investigate whether the German results hold for other countries as well. They find that US, European and other international RES are highly sensitive to changes in market fluctuations and tilt towards small-cap and growth stocks. European and international indices show negative returns in the period between 2008 and 2013 which can be explained by price momentum, which has been substantially positive until 2007. The returns of US stocks are positively correlated with fossil fuel prices, something which does not seem to be the case for European stocks. As with the German stocks, the collapse of RES returns after 2008 is likely the result of decreasing margins. Gupta (2017) expands existing multifactor models with factors from an economical and societal point of view. The findings show that RES benefit from country-level technology and innovations. Social factors that positively impact RES returns are uncertainty avoidance and long-term orientation. Countries that rank high in indulgence show lower RES returns.
The existing literature shows that most energy industry studies focus either on RES or TES, while simultaneous studies are scarce. The scope of this paper is to investigate the impact of the environmental focus of an energy company on the systematic risk factors influencing their returns. In addition, the influence of the region a company is situated in and the time variability of the systematic risk factors is studied. Finally, the effects of the energy transition on electricity generating companies and companies indirectly related to the generation of electricity are investigated. In contrast to other studies, this study investigates RES and TES simultaneously; it therefore gives an all-encompassing insight in the effects of the energy transition on the energy industry. The objective of this study is to answer the following research question:
How does the environmental focus of energy companies affect the systematic risk factors and returns of energy companies across different regions and over time?
The research question is addressed by studying the returns of five different energy sectors across six regions between 2000 and 2016 using individual stock data with monthly and annual intervals instead of frequently used index data. The energy sectors consist of oil and gas producers, oil and gas equipment suppliers, suppliers to the alternative energy industry and electricity producing companies which are either conventional or alternative. The regions included are Australia, Canada, China, Europe, New Zealand and the US. The regions are chosen since they are geographically different from each other and because their environmental focus varies. The rolling regression method as employed by Bessler and Opfer (2004) is used to study the influence of the systematic risk factors over time.
4
general, RES are considered most risky with high market betas on average. TES are influenced by the returns of technology stocks, however, less than RES. When controlling for technology stock returns, oil price changes and the general market returns, other company specific, macroeconomic and monetary factors almost all become insignificant, indicating that both TES and RES returns are significantly influenced by only a small number of risk factors. The influence each risk factor has is likely to change around periods of distress in financial markets.
5
2. Theory
This study investigates the risk factors influencing energy stock returns using multifactor asset pricing models. This literature review describes the existing literature that is relevant for the study. First, the general literature on multifactor asset pricing models is discussed. The review then focuses on literature regarding risk factors influencing general stock returns, from which the focus narrows down to energy stocks. Eventually the stability of findings over time is discussed.
2.1. Multifactor asset pricing models
One of the most important debates in recent market research is analyzing the behavior of stock returns. Over the years, many researchers have proposed and tested different models with the goal of providing an explanation for the return behavior of stocks and other assets. The foundations of modern multifactor asset pricing models (hereafter multifactor models) are set by Sharpe (1964) and Lintner (1965), who introduced the Capital Asset Pricing Model (CAPM). The CAPM measures the sensitivity of an asset’s expected returns to the expected returns of a market portfolio. This measure is referred to as the beta of an asset and indicates how risky an asset is relative to the market portfolio.
Alternative to the CAPM is the Arbitrage Pricing Theory (APT) introduced by Ross (1976). The APT is a model which is based on the idea that the expected returns of an asset can be predicted by the relationship between the asset and multiple risk factors. The risk factors generally consist of different macroeconomic factors or proxies for these factors. A market factor, contrary to the CAPM, is not required to be one of the factors.
Since the introduction of the CAPM and the APT, many researchers have proposed variations on these models by using additional factors. In the current literature there are a number of variations which serve as a benchmark. These models are known as the Fama and French (1993) three factor model, the Carhart (1995) four factor model and the Fama and French (2016) five factor model. These models all consist of factors which are based on the difference in returns between a set of decile portfolios. The portfolios are based on different criteria. The five factor model uses the excess returns of a local market over the risk free rate, the difference in returns between small and big stocks, high and low value stocks, profitable and unprofitable stocks and conservative and aggressive stocks as factors. The former three are used in the three factor model and together with a momentum factor in the four factor model.
Discussion on what the Carhart and Fama and French factors actually measure and thus their economic meaning has increased in recent years. Lakonishok et al. (1994) suggest the return difference between small and big stocks is a proxy for an investor bias in earnings-growth expectation. Ferguson and Shockley (2003) and Rolph (2003) note that the Fama and French factors measure leverage effects. Kothari et al. (1995), Berk (1995) and Ferson et al. (1999) argue that the factors’ explanatory power is spurious. Other research by Aretz et al. (2010) and Petkova (2006) focuses on explaining the factors using macroeconomic variables. Bessler and Conlon (2018) examine the factors’ sensitivity to certain macroeconomic state variables and find that a small number of variables account for a large proportion of the explained variation.
2.2. Macroeconomic and monetary variables
6
The underlying assumption is that stock prices respond to external economic forces, the variables influencing the economy therefore influence stock returns. They present evidence of significant and insignificant factors. The future one year growth rate of industrial production is significant. In addition, changes in the risk premium and in the slope of the yield curve are found to be significant. The default risk premium is measured as the return of a BBB-rated bond portfolio minus the return of a portfolio of long term government bonds. The yield curve slope is defined as the difference between the yield of a long term and a short term government bond. Changes in the unanticipated and expected inflation are weakly significant. The level of consumption and oil price changes are deemed insignificant. Another finding is that a stock market index has an insignificant influence on pricing when compared to economic state variables.
Following the paper of Chen et al. (1986), many researchers have focused on explaining stock returns using macroeconomic variables. Research on US stock returns is in abundance, the results however are mixed. Chan et al. (1998) and Maio and Philip (2015) state there is no significant relationship between stock returns and macroeconomic variables. Fama (1981) finds a strong connection between stock returns and the growth rate of production. Fischer and Merton (1984) add that stock returns predict future production. Production growth is not the only factor influencing returns, Ang and Bekaert (2007) note that the short term interest rate is the most predictive variable. Stock prices are positively related to industrial production and negatively to long-term interest rates according to Humpe and Macmillan (2009).
Less research has been performed in Europe; the results on the other hand appear to be more unambiguous. According to Wasserfallen (1989) a weak relationship exists between European stock markets and economic variables. Peiró (1996) states that, similar to the US, prices react positively to production and negatively to interest rates. Nasseh and Straus (2000) confirm this finding. Rapach et al. (2005) conclude that the most reliable predictor of European stock returns is the interest rate. Peiró (2016) compares the influence of industrial production and interest rates in France, Germany, the UK and the US. Across the European countries, both factors appear to be equally important. The results for the US, on the other hand, show that production is the only significant variable influencing stock returns in the same period. Asprem (1989) investigates the relation between macroeconomic variables and the stock returns of ten European countries and finds that the significance per variable varies between countries.
The mentioned literature shows that there is significant evidence of macroeconomic variables influencing stock returns. An import aspect, however, appears to be the influence of the country in which the research is conducted. In similar time periods, the results vary per country. This finding gives rise to the first hypothesis:
: The importance of systematic risk factors varies between regions.
2.3. Industry and company specific factors
7
In addition to the macroeconomic and monetary variables that are assumed to influence the stock market as a whole, there are variables which are related to one sector, but have no evident relationship with another sector. This study focuses on stocks from the energy industry and therefore on the literature for energy related stocks. Energy related entails both TES and RES.
The literature investigating TES and RES simultaneously is scarce. Gormus et al. (2014) conduct research on the volatility transmission between energy-related asset classes such as petroleum, coal, solar, biofuels and others. They find that risk spills over from energy companies to commodities in a unidirectional fashion. Gormus et al. (2015) investigate the responses of the same energy-related asset classes to outside price shocks of oil and gold. Of the TES, coal and natural gas stocks show the strongest response to oil price shocks, while solar gives the strongest response for RES. In the short run there are no responses to gold price shocks for any sector, in the long run there is opposite evidence for almost all sectors. The majority of the literature regarding energy stocks investigates TES and RES separately. TES studies focus mainly on the impact of oil price changes along with other factors. Bianconi and Yoshino (2014) study 64 oil and gas companies across 24 countries. They find that general factors such as the general market return, the volatility index, the price of crude oil and exchange rates are important in explaining TES returns. Sadorsky (2001) states that the Canadian oil and gas industry stock prices are significantly impacted by exchange rates, oil prices and interest rates. Stock prices are positively related to oil price changes and negatively to exchange rates and interest rates. Giovannini et al. (2005) indicate a low to extreme dependence of oil companies’ returns on oil prices and local market returns.
8
Bessler and Opfer (2004) suggest that macroeconomic and monetary factors have a varying impact on stock returns, depending on the industry the stock is situated in. Narrowing down to energy industry specific factors, few comparative studies between RES and TES are conducted. Gormus et al. (2015), however, provide signs of diverse reactions to risk factors between energy sectors such as gas, coal, solar and biofuels. Although other factors, such as particular macroeconomic factors and technology stock returns have not been tested simultaneously on RES, TES and across sectors, existing studies provide evidence that they impact stock returns differently depending on the industry or sector a company operates in. This assumption is tested with the following hypothesis:
: The importance of systematic risk factors varies between energy sectors.
2.4. Consistency of findings over time
The existing literature, as mentioned before, gives reason to believe that systematic risk factors influence returns differently depending on the country or industry the stock belongs to. Studies investigating the influence of risk factors in a certain country or industry sometimes come to different conclusions when different sample periods are used. The time period in which the research is conducted therefore seems to influence the results.
Bessler and Opfer (2004) show the relationship between macroeconomic variables and a stock’s return is time variant for different industries. Over time the explanatory power and beta coefficients per factor change, which is likely to be the result of a company’s success in mitigating certain risks. Chen et al. (1986) conclude oil does not have a significant impact on US stock returns. A more recent study by Sadorsky (1999) shows that after 1986 oil prices have become more important in explaining stock returns. Mollick and Assefa (2013) show a significant relation between US stock prices and oil price changes. They note that the relationship changes after the crisis of 2008. Degiannakis et al. (2011) state the correlation between oil prices and stock market prices is time-varying and changes the most in times when financial markets are in distress. Evidence on time varying risk factors in European RES is given by Bohl et al. (2015). Although all of the tested factors have a different impact over time, the momentum factor changes drastically at the beginning of the 2008 financial crisis. Bianconi and Yoshino (2014), Kumar et al. (2012), Henriques and Sadorsky (2008), among others, provide evidence that the risk factors influencing RES returns are not constant over time.
The evidence stated above suggests that the influence of a systematic risk factor is time dependent, which results in hypothesis three:
9
3. Methodology
The main focus of this study is to investigate the effect that the environmental focus has on the systematic risk factors of energy related stock returns over time and in different regions. To study the influence of multiple factors, different methods are employed. This chapter provides an overview of the general model and the methods that are used in this research.
3.1. General model
The general model used is a multifactor model which uses a number of factors to explain the returns of an asset. Return explaining models are based on the Capital Asset Pricing Model (CAPM) introduced by Sharpe (1964) and Lintner (1965):
( ),
where is the expected return on asset , the risk free rate, the expected return of the market and the sensitivity of the stock’s returns to the excess market returns. The formula can be rewritten in such a way that the left hand side of the equation shows the excess returns of a stock over the risk free rate:
.
An alternative to the CAPM is the Arbitrage Pricing Theory (APT) suggested by Ross (1976) which has more than one factor:
, with being the risk premium associated with factor and the sensitivity of the asset’s returns to factor .
A more general econometric representation of Equation (3), presented by Fama and French (1996) can be expressed for any risky asset:
,
in which is a constant return factor not explained by the other factors, is a systematic risk factor and is a parameter indicating the sensitivity of asset ’s excess returns to risk
factor . The asset’s idiosyncratic risk is represented by
10
3.2. Dependent variable
The research focuses on the excess returns of existing indices and individual assets. The individual assets are combined into value weighted portfolios based on the sector they operate in and their geographical location. Existing indices’ returns are already value weighted. The portfolio’s excess returns are calculated as:
∑
∑
where is the market value of asset in month . The asset’s returns are calculated as:
in which is the end of the month price for asset in month . By constructing value
weighted portfolios based on sectors and geographical location it is possible to compare the impact of the environmental focus across regions and over time.
3.3. Time variability
Bessler and Opfer (2004) state that the underlying assumption in most multifactor models is that the beta coefficients are constant over time. They, however, present evidence of coefficients being time variant for different industries. Bohl et al. (2015) support this evidence by finding time varying betas of RES.
To test the assumption of time invariant beta coefficients in this research’s sample period, a test for breakpoints by Bai and Perron (2003) is used. This test is programmed in Eviews to detect structural breaks. The Bai-Perron test is run on two existing global indices and indicates at least 5 structural breaks for both indices at the 0.05 significance level.
The indicated dates are tested using the Chow (1960) test. The Chow test splits the data from each index into two groups using the following equation for group one and for group two. Note that this is a general equation different from the actual equation used, which has more independent variables. Under the null hypothesis the assumption is , and . The F statistic from the chow test is:
where is the residual sum of squares from the full sample period, from the sample before the break and from the sample after the break. and are the number of observations before and after the break, respectively. The number of parameters is . The results of the Chow test on the two indices show significant support at the 1% level of significance, of time varying beta coefficients.
Based on the results of existing literature and the Chow test, the rolling regression technique employed by Bessler and Opfer (2004) is used to study how the beta coefficients of the indices vary over time.
11
+1. The total number of estimated sub-samples with length in the time series with a total length of is . In this study the length of the estimation period is set to 36 months. This short estimation period is chosen to increase the amount of subsamples over the relatively short time series length. Bessler and Opfer (2004), mention that the impact of shocks is reduced in a short estimation period, which is considered an advantage. Furthermore, a disadvantage of a longer estimation period could be that coefficients are smoothed out over different business cycles.
To test the significance of each beta coefficient estimate ̂, the t-value of a two sided t-test is calculated:
̂
̂
The significance level is 0.05. The t-value therefore has to be over 1.96 or under -1.96 in order to be significant.
3.4. Variance decomposition
In order to study the relative impact of different factors on the risk regarding energy stocks, the variance of excess returns is decomposed. Variance is used as a measure of an asset’s total risk (Campbell and Ammer, 1993). The total variance of the excess returns is decomposed into the variances of the different factors.
Bessler and Opfer (2004) note that the variances of the factors from Equation (4) multiplied by their squared beta coefficients, together with the security specific variance are a measure of the total variance of the excess returns . Mathematically this is:
.
It can be proven that this equation indeed is an accurate representation of variance decomposition by using a two-factor model as illustrated:
( ) .
The relative impact per factor is obtained by dividing by the total variance. The relative impact of the idiosyncratic variance is calculated by dividing by the total variance.
The equations above show an important aspect of variance decomposition, namely that the two factors are assumed to be independent:
12
Since this study uses different macroeconomic factors, there is reason to believe the factors are not independent over time. Appendix A shows the correlation coefficients between different factors in Europe. The definitions of these factors are explained in the following chapter. It is clear, with correlations up to 0.912 there is evidence of correlation among the independent variables, which can cause problems for the variance decomposition. Similar results are obtained for others regions. Bessler and Opfer (2004) orthogonalize the factors to overcome the problem of multicollinearity. This study uses the same approach.
3.5.1. Orthogonalization
Orthogonalization is used in this study to remove the correlation between different factors. The process can be used for models with two or more factors. As an illustration of the process, a two-factor model is considered:
where factors and are correlated, meaning . To remove the effect of from , is regressed on . This results in the following regression:
here and are the coefficients from the regression and is the random error term. Using the standard regression method causes the random error term to be uncorrelated with , thus . The random error term
represents without the effect of . If it is now defined that
,
the two factor model from Equation (12) becomes
where , and . The two factors are constructed to
be orthogonal. This process can be extended to models with more than two factors by regressing an additional factor on the original two factors and using the random error term as a new independent variable.
3.5.2. Order of orthogonalization
13 Figure I
Order of orthogonalization
Note: Hierarchical variance decomposition over time in different orders. i) industry-company-macroeconomic/monetary ii) industry-company-macroeconomic/monetary-industry-company iii) used in this study. The dates indicate the end date of the 36 month estimation period.
14
company specific factors decreases. This study uses a mixed order where the environmental policy stringency and share of renewable energy sources in total electricity production are put in first. The input factors oil, gas and coal are put in next, after which the macroeconomic and monetary factors are put in. Finally, the momentum factor, the technology stock returns factor and the general market factor are put in. The reason that the technology and market factors are put in last is that they are too dominant when put in at or near the beginning. Other factors are therefore put in first to achieve a better insight in the variance explained by these factors.
3.6. Impact of renewable energy on electricity generating companies
The recent changes in the German energy market are a striking example of how the energy market is evolving. To study which factors may drive the change in the energy industry, the returns of electricity generating companies are studied. The recent German changes imply that using either traditional or renewable energy sources for the generation of electricity impacts a firm’s returns. An obvious method is to regress the returns of a company on the percentage of renewable or traditional energy used to produce electricity. The downside of this method is that the share of renewables has increased for every company in the recent years, while returns have decreased, implying that using renewable energy impacts returns negatively. While this can be a correct conclusion, it disregards the fact that consumers may choose to only consume electricity that is mainly produced by renewable energy sources. Producers can therefore be penalized for using too much non-renewable energy sources compared to renewable energy sources. To account for this fact, the country’s average energy input mix used for the generation of electricity is subtracted from the company’s energy input mix. The energy input mix is divided into three categories being coal and gas, nuclear and renewable. Coal and gas are not presented separately since companies do no report their input mix consistently detailed to make this distinction. When a company uses more than the country’s average of a certain input, the result is a positive number and vice versa. The country’s average is chosen since it is assumed to reflect the demands of all customers. The other factor used is the local market return. The dependent variable is the excess return of an electricity producing company.
The beta coefficients per factor are obtained using the OLS regression method in combination with the panel data approach. Multicollinearity between the dependent variables is likely because the increase (decrease) in the use of one input is reflected in the decrease (increase) of another input. By using the Ridge regression method, the multicollinearity is accounted for. This method, however, may give biased results since it shrinks the beta estimates towards zero in order to reduce the variance of the estimate. A tradeoff between bias and variance therefore has to be made. The main advantage of the Ridge regression method is that it can be applied to panel data.
3.7. Subdividing renewable energy stocks
Renewable energy is a collective name for different categories of nontraditional energy sources. How energy is generated differs widely between the renewable categories. To study whether the impact of systematic risk factors varies between the categories, the excess returns of companies per category are analyzed. OLS regression with the panel data approach is used to study the impact of different risk factors. The companies under consideration range from suppliers to maintenance companies specialized in a renewable category and are thus different from the electricity producing companies discussed in the previous section.
15
16
4. Data
This chapter contains an overview of the data used for this study. It presents a general description of the variables, together with descriptive statistics per variable. Data is retrieved from Thomson Reuters DataStream unless stated otherwise.
4.1. General data description
This study analyzes the returns of four existing global equity indices and 29 equity indices constructed from five different categories of stocks based on six regions. The analysis focuses on the impact of different factors on the excess returns of the indices. As explained in the previous chapter, the excess returns are calculated by subtracting the risk free rate from the returns of an asset. The risk free rate is the minimum expected rate of return for an investor, any deviation from the risk free rate is therefore considered excess return. This study uses the yield to maturity (hereafter YTM) of a 3-month, zero coupon government bond as the risk free rate. Since the bonds are backed by a government, the probability of default is marginal. The yield on the bonds however is based on annualized returns. To transform the annualized returns into monthly returns, the following adjustment is made:
( ( )) .
For the global indices, the 3-month US T-Bill is used, as most of the weight is placed on assets from the USA in these indices. The other region specific indices use a 3-month, zero coupon bond denominated in their local currency which is backed by the local government as risk free rate.
4.1.1. Global indices
The global indices used are the MSCI World Index (MSCI), the MSCI World Energy Index (MSCIE), the Ardour Global Alternative Energy Index (AGIGL) and the Winderhill New Energy Global Innovation Index (NEX). The MSCI index consists of large- and midcap stocks from 23 developed countries and has 1,648 constituents from 11 sectors in 20175. The weights per stock are based on their relative market capitalization. Because of the diversified composition, the index can be used as a global market proxy. A variation on the MSCI index is the MSCIE index, which holds 83 constituents from 23 developed countries6. Every constituent is classified in the energy sector according to the Global Industry Classification Standard (GICS). The list is dominated by TES.
Opposite to the MSCIE, which mainly holds TES, are the AGIGL and NEX indices. These two indices only contain RES. The AGIGL index contains 90 stocks from 13 countries. The countries are both developed and emerging. The stocks can be divided into seven sectors of which information technology, industrials and utilities cover over 85% of the total market capitalization7. The NEX index contains 105 securities from 28 countries. The weightings of the securities are based on market capitalization and are adjusted for several factors related to exposure to renewable energy8.
5 MSCI World Index (2018). Retrieved March 25, 2018 from MSCI: https://www.msci.com/world
6
MSCI World Energy Index (2018). Retrieved March 25, 2018 from MSCI:
https://www.msci.com/documents/10199/de6dfd90-3fcd-42f0-aaf9-4b3565462b5a 7 Ardour Global Alternative Energy Index (2018). Retrieved March 25, 2018 from S-Network:
http://ardour.snetglobalindexes.com/sites/snardour/constituent-data?index=AGIGL 8 Winderhill New Energy Global Innovation Index (2018). Retrieved March 25, 2018 from NEX:
17
4.1.2. Sector and region specific indices
Using existing indices has a number of disadvantages. Gupta (2017) states returns from indices can give biased results due to their composition. Indices place too much weight on a certain region or sector. In addition, returns are overstated and less volatile due to the survivorship bias which is the result of using larger, well known stocks instead of smaller stocks with a higher probability of default. To avoid this problem, this study generates monthly value weighted indices containing both live stocks and stocks that have gone bankrupt or have been delisted during the sample period from six different regions. The regions are Australia, Canada, China, Europe, New Zealand and the USA. For every region there are five indices, except for New Zealand which has four, each from a different sector. The selection of assets to include is based on their sector classification given by the Industrial Classification Benchmark (ICB)9. The sectors included are:
530 Oil & Gas Producers
570 Oil Equipment, Services and Distribution 580 Alternative Energy
7535 Conventional Electricity 7537 Alternative Electricity
The numbers in front of the sectors, instead of the sectors’ names are used throughout this study to improve readability.
The composition of sectors is chosen to reflect both TES and RES. Sector 530 represents companies producing traditional oil and gas. Sector 570 is added to include companies which are related to traditional energy in an indirect fashion. Sector 580’s constituents are indirectly related to the alternative energy industry and are thus considered RES. Companies in sector 580 vary from suppliers of equipment and services to distributors. By using the mentioned three sectors, an all-encompassing insight in non-electricity generating companies is achieved.
Another sector in which the energy transition is becoming more evident is the utilities sector focusing on the production of electricity. Sector 7535 represents the companies using traditional power plants running on fossil fuels. The plants using renewable energy as inputs are situated in sector 7537. Including utilities in the study broadens the scope of the research, which improves explaining the overall impact of using either traditional or renewable energy sources.
4.1.3. Factors
In order to investigate what drives the excess returns of energy stocks, the influence of different factors is evaluated. Previous studies find multiple factors influencing stock returns. This study uses several of these factors. The factors used are considered either company specific, industry specific and macroeconomic or monetary.
4.1.3.1. Company specific
Variables that are assumed to be significant for one company or a group of similar companies, but not for another group of different companies, are used as company specific factors. The assumption of whether a factor is significant or insignificant is based on previous studies.
9
18
4.1.3.1.1. Environmental policy stringency
The environmental policy stringency factor is used to study the influence of the environmental focus of a country. The factor is created by the Organization for Economic Co-operation and Development (OECD)10. The OECD defines stringency as the degree to which environmental policies put a price on polluting and other environmental harmful behavior. It is measured on a scale of 0 to 6 where 0 is not stringent and 6 is highly stringent. This factor is not available for New Zealand. Since it is a specific factor, no substitute can be used and this variable is therefore omitted from the analysis of New Zealand.
4.1.3.1.2. Renewable energy as percentage of total electricity production
To analyze the influence of the share of renewable energy in the total energy production of a country, a proxy is used. Instead of measuring the total amount of energy produced, the electricity a country produces is measured. The reason for choosing this proxy is because of the data availability per country. The World Bank11 has available data for the sample period of every region in this study. The data contains the percentage of electricity produced from renewable energy sources with annual intervals.
4.1.3.1.3. Technology stocks
Only little research has been conducted on factors other than oil regarding TES. For RES, most studies focus on both oil and technology stocks returns. Kumar et al. (2012) find that RES returns are impacted significantly by technology stock returns, since investors see alternative energy stocks as being similar to technology stocks. This study analyzes the impact of technology stocks returns on both TES and RES. It is measured by the monthly growth rate of an existing index consisting of technology stocks from the region under consideration:
, where is the value of the index at the end of month .
4.1.3.1.4. Momentum factor
The momentum factor, introduced by Carhart (1995), measures the difference in returns between a portfolio containing high performing stocks and a portfolio containing low performing stocks. The historic data for this factor is available online for multiple regions and is updated monthly. Canada and the USA use the momentum factor of North-America. Australia, China and New Zealand use the data of the Asia Pacific region excluding Japan. Europe uses the European momentum factor. Bohl et al. (2013) find the momentum factor to be significant for German renewable energy stocks over a period of multiple years. This implies that the momentum factor does not capture seasonal momentum; the momentum factor should then only be significant for short periods of time every year.
10
Environmental policy stringency (2018). Retrieved March 26, 2018 from OECD: https://stats.oecd.org/Index.aspx?DataSetCode=EPS
11 Electricity production from renewable sources (% of total) (2018). Retrieved March 26, 2018 from The World Bank:
19
4.1.3.2. Industry specific
Industry specific factors are designed to capture the influence of variables which mainly impact energy related companies and are assumed to have little or no impact on companies operating in other sectors.
4.1.3.2.1. Oil
Oil is considered industry specific in this study, since it has been proven to impact both TES and RES, but has no significant influence on the general stock market. The monthly percentage of price change in the 1-month West Texas Intermediate future is used as a measure for monthly oil price changes in month :
The 1-month future is used instead of the spot price since it is less sensitive to short run price fluctuations which are the result of temporary shortages or surpluses (Sadorsky, 2001).
4.1.3.2.2. Coal
Coal is industry specific since it is an important input source in the energy industry. The change in coal price in month is calculated as:
where is the end of the month price of coal in Australia12
. Australia is chosen since it the biggest exporter of coal and the price data is available on a monthly basis. Comparable data from other regions is unavailable.
4.1.3.2.2. Gas
Similar to coal, gas as well is an important input source. The monthly US natural gas price is used and is obtained from the US Energy Information Administration13. The US price data is chosen since it available on a monthly basis. Again, comparable data from other regions is unavailable. The monthly price change is calculated as:
4.1.3.3. Macroeconomic and monetary
The macroeconomic and monetary factors are based on the work of Chen et al. (1986), who find that certain economic variables systematically influence the stock market. The variables which are found to be significant are used as a factor in this study.
4.1.3.3.1. Industrial production
The macroeconomic variable is industrial production, which is a measure of the rate of change in industrial production over a year. It is different to other factors since it measures the change over a year instead of a month. Chen et al. (1986) argue that stock prices are based on the valuation of long term future cash flows, monthly changes in production may therefore not be
12
Global price of coal (2018). Retrieved May 26, 2018 from FRED: https://fred.stlouisfed.org/series/PCOALAUUSDM
20
related to changes in stock prices in the same month. The factor industrial production is calculated as:
, where is the industrial production in month . The regional industrial production data is based on production indices measuring the total production in a region.
4.1.3.3.2. Interest rates
The first monetary variable is the interest rate level. The YTM of a 3-month, government backed, zero coupon bond is used as a measure of the interest rate level.
4.1.3.3.3. Term spread
The second monetary variable is the influence of unanticipated changes in the returns on long term bonds which captures the influence of the shape of the term structure. It is measured as:
in which is the yield on a 3-month, zero coupon, government backed bond in month . The long term risk free rate is based on the yield of a 10 year, zero coupon, government backed bond .
4.1.3.3.4. Risk premium
The third monetary variable measures the unanticipated change in risk premium required by investors. It is measured by subtracting the long term risk free rate from the returns of a rated bond portfolio. Since not all countries have sufficient data available to construct low-rated bond portfolios, a proxy is used. The ICE Benchmark Administration14 constructs portfolios from below investment grade bonds and reports the annual yield for Europe and the USA. The European yield is used for the European region and the USA yield is used for every other region in this study. The difference between the below investment grade yield and the long term risk free rate is used as a factor:
4.1.3.3.5. Excess market returns
Although Chen et al. (1986) argue that using the above mentioned macroeconomic and monetary factors can make the market factor’s explanatory power insignificant, the market factor is still used to capture effects not explained by any of the other factors. The market factor is the monthly return of a local market index minus the risk free rate :
. 4.1.4. Electricity generating companies
In recent years the effects of the energy transition have become evident in the electricity generating industry in Germany. Companies split up their generating activities in parts focusing either on renewable energy sources or traditional energy sources. To study how the
21
energy input mix affects the returns of an electricity generating company, the largest German and UK companies in this sector are analyzed. The German companies are chosen because of the recent events in their industry. To test whether the results are the same in a different region, the UK sector is studied. Although the UK is situated in Europe, it is not connected to the European Union in the same fashion as Germany. Next to that, there are geographical differences. The UK is surrounded by water while Germany is only partially connected to open water, also the countries’ climates are different.
The reason for focusing on the largest electricity generators is that almost all of these companies are publically listed. The energy input mix is therefore publically available as these are stated in the annual reports. The data used for the input analysis comes from the companies’ annual reports between 2007 and 2016. The starting point of 2007 is chosen since the largest electricity generators all report their input mix from that year on. The transition from non-renewable to renewable inputs also becomes more pronounced from 2007 onwards. Not all inputs are reported equally detailed for all companies, mainly because companies not always report coal and gas separately, therefore three input groups are chosen:
Coal and gas (Includes all types of coal and gas) Nuclear
Renewables (Includes all types of renewables)
The input mix of a company is compared to the average input mix a country uses for the generation of electricity. Data of the country’s usage is retrieved from Eurostat15
. The database is updated until 2016, which explains the end date of the sample period. The input mix is reported annually, the analysis is therefore based on annual intervals. The remaining factor, the market factor is the annual excess return of the local market.
4.1.5. Categories of energy stocks
After analyzing the electricity producing companies, the companies indirectly related to the production of electricity are analyzed. These companies vary from suppliers to maintenance firms. To expand on the research conducted on electricity producing firms, again companies from Europe are used and the sample period remains 2007 to 2016.
All companies in the energy industry are assigned to a sector based on the ICB classification. The renewable energy sector 580 consists of stocks which are focused on a wide variety of renewable energy sources. The ICB standard does not subdivide renewable energy stocks based on their energy source, therefore another standard is used. The North American Industry Classification System (NAICS)16 gives a more detailed insight in RES. From this classification system, 5 renewable energy categories arise. For two of these categories, hydro and geothermal, insufficient data is available. The remaining categories are solar, wind and biomass.
The factors used in the analysis are the local market return, the share of electricity produced from the renewable input source under consideration and the returns of a Chinese peer group. The input share is retrieved from Eurostat. The Chinese peer group returns are the weighted returns of Chinese companies operating in the same renewable energy category.
15 Energy from renewable sources (2018). Retrieved March 28, 2018 from Eurostat: http://ec.europa.eu/eurostat/web/energy/data/shares
22
4.2. Descriptive statistics
The descriptive statistics are divided based on their geographical origin. Global indices are presented as a group and region specific indices are presented separately. The statistics are based on data between January 2000 and December 2015, unless stated otherwise. Table I shows the descriptive statistics for the global indices. Tables for the descriptive statistics per region are placed in appendix B. All stated returns are excess returns over the risk free rate.
4.2.1. Global
Not every global index exists as long as MSCI and MSCIE indices. The RES indices, AGIGL and NEX have been introduced later. The estimation period is therefore between February 2001 and December 2015. The RES indices have a higher standard deviation of returns than the other indices. Between the RES indices is a notable difference in mean and median return. The MSCI energy index has the highest mean return.
TABLE I
Descriptive statistics (Indices)
Table I shows the descriptive statistics of the indices for the period between 02-2001 and 12-2015.
MSCI MSCIE AGIGL NEX
Mean 0.19% 0.34% 0.02% 0.27% Median 0.79% 0.14% 0.11% 1.13% Maxium 14.51% 16.8% 24.49% 20.73% Minimum -18.85% -17.29% -35.24% -32.28% Std. Deviation 4.95% 6.26% 9.79% 8.23% Skewness -0.58 -0.33 -0.47 -0.57 Kurtosis 4.55 3.53 3.88 4.53 Jarque-Bera 28.07 5.39 12.27 27.04 Observations 179 179 179 179 4.2.2. Regions
In Australia, both RES sectors 580 (20.25%) and 7537 (12.26%) have the highest mean return standard deviation. 580 has the highest mean return of 3.42% and the lowest median return of -1.09%. The oil and gas sector 530 has a relatively high mean return of 1.32% with a standard deviation of 6.43% which is relatively low compared to other sectors.
The Canadian sectors 530 and 570 have the highest mean return with 2.22% and 2.10%, respectively. Although these sectors have a relatively high standard deviation (12.65% and 9.96%) compared to the local market (4.63%), sector 580 has the highest standard deviation (19.11%). The mean return of 1.64% is higher than the local market mean return of 0.23%.
23
The highest mean return in Europe comes from sector 580 (3.16%), which exceeds the second best performing sector, 570, (1.80%) evidently. The standard deviation of the former is more than double as high as the latter, with 15.28% and 7.44% respectively. The alternative electricity sector (1.50%) outperforms the traditional electricity sector (0.98%), with standard deviations of 7.51% and 5.47% respectively.
New Zealand is the only region with negative mean returns in a sector other than the local market. Sector 580 has a negative mean return of -2.38% and 570 -0.41%. Remarkable is that sector 580 has a standard deviation of 12.51% and 570 0.17%. The best performing sectors are 530 and 7537 with mean returns of 1.56% and 0.89% respectively.
24
5. Results
This section presents and discusses the results of the study. First, the global indices are considered. Next, the different value weighted sector indices are reviewed on country level. Then the electricity generating firms are discussed after which the renewable energy sector is analyzed in more detail.
5.1. Global indices
The four global indices are studied in the time period between February 2001 and December 2015. The starting point of this period equals the starting point of the NEX index. The other factors have data from before this period. For reasons of comparability this period is chosen. The MSCI index is used as general market proxy for the other three indices and therefore lacks the market factor in the results.
5.1.1. Variance decomposition
Table II presents the maximum, average and minimum factor specific explanations of the different factors for the global indices. The most dominant factor is the price change in oil, which on average explains between 26% and 47% of the total variance. Oil impacts the MSCI-E index, consisting of TES, relatively high with 47% on average. The average impact of oil on the RES indices and the MSCI index is nearly identical. The different impact of oil price changes on RES and TES indices confirms the findings of Sadorsky (2001). Technology stock returns explain a large share of the variance of the RES indices and the MSCI index, with averages between 26% and 33%. The MSCI-E index, with an average of 10%, is least affected by technology stock returns. Macroeconomic and monetary factors do not appear to account for a significant share of the total variance for any of the indices, neither do gas and coal.
5.1.2. Time variability of variance decomposition
The maximum, average and minimum factor specific explanations differ widely for individual factors, implying the variances are not stable over time. Figure II shows the percentage of factor specific explanation over time for the MSCI-E index and the AGIGL index, Appendix C shows the same for the MSCI index and the NEX index. Oil price changes are most dominant in explaining the MSCI-E’s variance over the entire sample period. Technology stock returns explain most of AGIGL’s variance until 2009, where after oil becomes most dominant. The share of oil subsequently decreases and technology returns become more important towards the end of the sample period. The remaining factors only have a marginal influence on both indices throughout the sample period.
5.1.3. Beta coefficients
25
RES indices based on the beta coefficients and the significance percentages. The RES indices are most sensitive to technology stock returns, with betas of 1.06 and 0.83.
Similar to the results of the variance decomposition, literature finds there is evidence of time varying beta coefficients (Bessler and Opfer, 2004), this study confirms this finding. Figure III shows how the beta coefficient estimates of technology stock returns and oil price changes behave over time for all sectors. The RES indices have the highest technology beta coefficient over the entire period; this difference becomes smaller near the end of the estimation period, indicating that other sectors are influenced more significantly by technology returns over time. The MSCI-E index is positively influenced by oil price changes across the entire sample period, while the other sectors are negatively influenced until 2003. The estimates for all indices increase simultaneously until 2012, where after they decrease. The MSCI is least sensitive to oil price changes, which is in line with the findings of Chen et al. (1986), who state that oil price changes do not affect general stock market returns significantly.
Figure II
Variance decomposition over time (MSCI-E and AGIGL)
MSCI-E AGIGL
26 TABLE II
Variance explanation per factor (Indices)
Table II shows the maximum, average and minimum variance explanation per factor for the indices between 02-2001 and 12-2015.
Index Oil Gas Coal Production Interest rate Term spread Risk premium Momentum Technology Market Specific MSCI Max 64% 11% 14% 22% 11% 11% 29% 6% 62% - 34% Average 26% 2% 4% 5% 1% 2% 8% 1% 33% - 18% Min 0% 0% 0% 0% 0% 0% 0% 0% 11% - 7% MSCI-E Max 74% 10% 8% 9% 7% 5% 13% 11% 24% 37% 38% Average 47% 3% 2% 2% 1% 1% 3% 1% 10% 14% 16% Min 10% 0% 0% 0% 0% 0% 0% 0% 1% 4% 6% AGIGL Max 59% 7% 21% 15% 14% 12% 24% 10% 81% 14% 36% Average 27% 2% 3% 2% 2% 3% 6% 1% 32% 6% 17% Min 0% 0% 0% 0% 0% 0% 0% 0% 8% 0% 7% NEX Max 66% 6% 25% 13% 22% 18% 23% 10% 64% 17% 37% Average 28% 2% 3% 3% 2% 4% 7% 2% 26% 7% 17% Min 0% 0% 0% 0% 0% 0% 0% 0% 6% 1% 5% TABLE III Beta coefficients (Indices) Table III shows the beta coefficients per factor for the indices between 02-2001 and 12-2015.
Index Alpha Oil Gas Coal Production Interest rate Term spread Risk premium Momentum Technology Market
MSCI 0.002 0.215 0.028 0.097 0.316 3.175 -1.353 -1.290 0.030 0.562 -
MSCI-E 0.001 0.466 0.053 0.096 0.243 3.479 -0.870 -1.293 -0.009 0.440 1.233
AGIGL 0.000 0.412 0.067 0.147 0.454 3.482 -3.772 -2.348 -0.140 1.063 1.147
27 Figure III
Betas estimates of oil and technology between 02-2001 and 12-2015 (Indices)
Note: The left graph shows the oil beta for all indices. The right graph shows the technology beta for all indices. The dates indicate the end point of the 36 month estimation period.
5.2. Region and sector indices
To study the influence of systematic risk factors across countries and sectors, value weighted indices are created from individual stocks, both dead and alive. The reason to include dead stocks is to eliminate the effect of the survivorship bias as proposed by Gupta (2017). Since there are six countries and five sectors, except for New Zealand which has four, the results of 29 individual sectors are discussed. To keep the presentation of the results concise and clear, not all tables are presented in this section; instead they are attached as Appendices E and F. Sectors 530 and 580 for all six countries are presented and discussed below to indicate differences across countries and sectors over time.
5.2.1. Variance decomposition across regions, sectors and over time
Appendix E shows the maximum, average and minimum factor specific variance explanation for all regions and sectors. Appendix F shows the variance decomposition over time for sectors 570, 7535 and 7537. The variance decomposition in Figure IV gives an insight in how the factors influence the returns of different sectors in different regions over time and thus provides information needed to answer Hypotheses I, II and III.
28
market returns. Although this factor has a relatively high impact on the return variance in different regions, there are no other regions in which this factor explains a similarly high share of total variance. The impact of technology stock returns is most noticeably present in Europe and the US, where is explains between 5% and 23% on average. The variance not explained by the common risk factors, the idiosyncratic risk, is highest for Australia and New Zealand, which indicates their returns are affected most by factors other than those used in this study. The findings suggest that the importance of risk factors varies between regions, which is in favor of Hypothesis I.
Hypothesis II states that the importance of systematic risk factors varies between energy sectors. Figure IV gives a comparison of the oil sector (530) and the alternative energy sector (580) for all six regions. From this comparison it immediately becomes apparent that oil explains a smaller share of total variance for sector 580 than it does for sector 530. Looking at the remaining sectors in Appendix F, only sector 570 is similarly influenced by oil as sector 530. The electricity generating sectors 7535 and 7537 are influenced less by oil price changes. As a result of the lower impact of oil, other factors have a relatively greater impact on total variance for sectors 580, 7535 and 7537. The environmental policy and the total percentage of renewable energy sources in total electricity production have the greatest impact on sectors other than 530. Although the average variance explanation of gas and coal is small, sectors 7535 and 7537 are more sensitive to changes in coal prices than the other sectors. An explanation for this fact is that coal is an important input source used in the generation of electricity. The macroeconomic and monetary factors only have a marginal contribution to the total variance for all sectors, with percentages around 5% or less. Of all sectors, alternative energy sector 580 is impacted most by technology stock returns, whereas oil sector 530 is impacted most by the general market return. The idiosyncratic risk is highest for sectors 580, 7535 and 7537. The results show that the impact of factors varies between TES and RES. In addition, there are differences between electricity generating companies and non-electricity generating companies. The findings are consistent with the findings of Bessler and Opfer (2004) and Gormus et al. (2015), who indicate that factors have a different impact on returns depending on the industry the stock is situated in. The evidence therefore is in conjunction with Hypothesis II.
29 Figure IV
Variance decomposition over time (Left: Oil & Gas producers and Right: Alternative Energy)
30 Europe New Zealand USA
31
5.2.3. Beta coefficients and significance
To study the impact of a factor on the returns of a sector, the beta coefficients are analyzed. The majority of the tables containing the average beta coefficient estimates per sector and region, together with the percentage of times the estimate is significant, can be found in appendix G. Since the remaining parts of the results section focus on Europe, the European beta coefficients and average significance are presented in table IV and V, respectively. Similar to the variance decomposition section, the results are analyzed per hypothesis.
The variance decomposition has already indicated that the impact per factor on total variance varies between regions. Similar results are found at the beta coefficient estimates and their significances. When first considering the average percentage of times a factor is significant, several similarities are evident. Across most regions there are three factors that are predominantly significant. These factors are oil price changes, technology stock returns and the general market return. The importance of these factors is also apparent in the variance decomposition. The beta coefficients of these factors vary between countries, however. The impact of oil price changes is positive in every region and the betas range from 0 to 0.549. Canada and the US are most sensitive to oil price changes, whereas New Zealand appears to be least influenced. The technology stock returns show a more mixed result. Canada and New Zealand are least affected by technology returns, while China, Europe and the US are most affected. Of all sectors in every region, only Canada’s sector 570 and New Zealand’s sector 7535 are negatively impacted by technology returns. The market factor indicates the riskiness of the energy sectors relative to the market portfolio. In general, the energy sectors are most risky in Australia and the US with ranges between 0.440 and 1.933 and 0.535 and 1.419, respectively. Of the monetary factors, the interest rate influences Chinese returns negatively and the US returns positively, while the results within the remaining regions vary. The remaining factors are all insignificant a relatively large proportion of the time. The differences in average beta coefficient estimates are in favor of Hypothesis I and agree with the variance decomposition findings.
The impact each systematic risk factor has varies between sectors according to
