Authors: ůůĞƌƚ:ſŶũƂƌŶƐƐŽŶ;džĂŵ/͗ϮϴϯϵϯϵͿ ŽŶŐƉƌŽũĞĐƚůĞĂĚƟŵĞ Ͳ,ŝŐŚĐŽŵƉĞƟƟǀĞƌŝǀĂůƌLJ /0)12307E1?045==17=6 Ͳ'ĞŽŐƌĂƉŚŝĐĂůůLJĐŽŶƐƚƌĂŝŶĞĚ

!"#$ ͲŶǀŝƌŽŶŵĞŶƚĂůƌĞŐƵůĂƟŽŶƐ Ͳ'ƌĞĞŶƐƟŵƵůƵƐƉĂĐŬĂŐĞƐ /0'7C0?7F3;9892@ ͲhŶĞdžƉůŽŝƚĞĚŐĞŽƚŚĞƌŵĂů 000=769D=F76 /0.;F=7561;207;7=2@0:7>5;:

Ͳ>ŽǁŶĂƚƵƌĂůŐĂƐƉƌŝĐĞ ͲůƚĞƌŶĂƟǀĞƐŽƵƌĐĞƐŽĨ 000=7;7C548707;7=2@ Ͳ>ŽǁŽŝůƉƌŝĐĞ

#,,#%$-'.$.&!

$)%&*$!

Source: Own creation & Lynch (2006).

31

3.5 Summary The geothermal industry faces many challenges, as there are high upfront capital requirements, long project lead-times, geothermal heat is geographically constrained and there are a number of other competing alternative energy producers. On the other hand, the renewable energy industry as a whole is growing at a fast pace, and with the high base load and continuous flow of energy, geothermal heat is an exciting energy alternative. There are many opportunities in the sector, for example involving incentive schemes and the public awareness of global warming. Although the geothermal sector has been experiencing negative publicity this year and the stocks have fallen drastically, it could soon turn the corner as several geothermal projects are nearing significant milestones, which could give the sector some boost.

32

4 Analyzing Factors that Influence Stock Prices Throughout the history there have been numerous stock market crashes that have awoken interest in research on why stock returns and volatility are propagated across world markets. Not all factors can be measured statistically, but can in fact influence indirectly prices in stock markets. Among factors that can influence stock prices are for example, the market sentiment; if the market is declining, so will most of the stocks and vice versa. Company announcements can also have their impact, when it is revealed that the company does not perform like the traders predicted. Interest rate change can also have its effect, as well as an evolution of a specific industry sector, which could affect most of the companies in a specific sector (Meta4forexBroker 2010). Furthermore, one possible interpretation is an informational link across markets, which means news in one country is perceived as informative to fundamentals of stock prices in another country. Another possible interpretation for this issue is market contagion. Stock prices in one country are affected by changes in another country beyond what is conceivable by connections through economic fundamentals. According to this view, overreaction, speculation, and/or noise trading are transmissible across borders (Frankel 2008). The following chapter will look into some variables that can affect the stock market and will be used in the analysis of change in stock prices in the renewables sector.

4.1 Theoretical Background - Arbitrage Pricing Theory A. Craig MacKinlay (1997) said that economic models could be cast as restrictions on statistical models to provide more constrained normal return models. Two commonly used economic models, which provide those restrictions, are the Capital Asset Pricing Model (CAPM) and the Arbitrage Pricing Theory (APT). Whereas the CAPM is an equilibrium theory where expected return of a given asset is determined by its covariance with the market portfolio, the APT is an asset pricing theory where the expected return of a given asset is a linear combination of multiple risk factors, and is based on the principle of no arbitrage. MacKinlay continued by stating that general findings suggest that with the APT, the most important factor behaves like a market factor and additional factors add relatively little explanatory power. Furthermore, Roll & Ross (1980) say that the APT is an appropriate alternative to CAPM because it agrees

33

perfectly with what appears to be the intuition behind the CAPM. The APT is built on a linear return generating process as a first principle and it is not restricted to a single period, as it will hold in both multi-period and single period cases. According to Hillier et al. (2008), the APT requires only four assumptions; 1. A factor model can describe returns. 2. There are no arbitrage opportunities. 3. There are a large number of securities, so that it is impossible to form portfolios that diversify the firms-specific risk of individual stocks. 4. The financial markets are frictionless. If we consider investment i with expected returns generated by the K-factor model represented by: !! ! !! ! !!! !!! ! ! ! !!" !!! ! !! The return on asset i is a function of asset i expected return and the sensitivity (beta) of the asset of each common factor (F). Because systematic factors are not identified and the existence of the linear relation between the factors and securities is only an assumption of the APT model, there are no suggestions of which variables should be used on the right hand side of the equation above (MacKinlay 1997; Hillier et al. 2008; Michailidis 2009). Hillier et al. (2008) say that there are fundamental differences between the APT and the CAPM models and often a multifactor APT is more favorable to use than the CAPM; as APT is more successful when it comes to explaining historical returns. They also conclude that APT can be controversial, because it is difficult to determine what factors to use and which factors explain expected returns. Although the CAPM and APT do not hold in reality, they can still be useful and have become increasingly important analytical tools when it comes to evaluating investment projects. However, though these models are important, it needs to be used with caution, while understanding their limitations.

34

4.1.1 Arbitrage Pricing Theory and Factor Analysis Michailidis (2009) says that there has been an ongoing debate among academics when it comes to the link between macroeconomic variables and financial market volatility. He continues, and states that stocks respond to common factors, i.e. the existence of industrial, utility and transportation indices which supports the premise of APT, where security prices respond to systematic factors. Studies have been made on the effect of macroeconomic variables on share returns in U.S. market, and Fifield et al. (2002) say that there is abundance of empirical evidence that have confirmed that macroeconomic variables can be used to explain share returns in developed market. Fama (1981; 1990) found out that, often more than 50 percent annual stock-return variances can be traced to forecasts of variables that include industrial production, real GNP and investments that are relevant to determine cash flows to firms. Chen et al. (1986) used among other factors, changes in monthly industrial production, inflation and risk premium in their analysis. In their paper they came down to the conclusion that stock returns are exposed to systematic economic news and the stock is priced in accordance with their exposures. There have been series of tests proposed when determining the number of factors to use to test the APT using factor analysis, and there has been more emphasis on testing the number of statistical factors in the returns equation, rather than the number of priced factors in the expected returns equation. Factor analysis and principal component analysis have been used to reduce the number of variables and to detect relationship between them. Furthermore, Brown & Weinstein (1983) found out that number of factors used when testing asset pricing models was limited to five as their analysis showed that a seven factor model did not explain the data better than using three or five factor model. Berry et al. (1988) also say that when using the APT model with five risk factors, it is superior to both the market model and the CAPM when it comes to explaining stock returns. They continue to state that these results are not restricted to the five factors they chose, but another set of five factors could also do well. This is supported by prior testing which have shown that the CAPM is inferior to an APT model, which incorporates unanticipated changes in five macroeconomic variables. Michailidis (2009) declares that after the publication of the tests done by Berry et al. a consensus began to develop in the literature.

35

When dealing with a large set of variables, principal component analysis and factor analysis can be used to reduce the number of variables and extract factors that include variables that are related to each other. This method can reduce noise from each independent variable as it is contributing to a factor along with other variables. Roll & Ross (1980) say that when dealing with time series data, they appeal to the statistical technique of factor analysis, and continue by stating that although the exact weights of the factors are altered by orthogonal transform, the linear hypothesis remains true. This is because that orthogonal transforms leave that space unchanged, altering only the directions of the defining vector; the column vectors of the loadings. As a result from rotating the factors and restricting them to be orthogonal, they become independent and have unit variance.

4.2 Sample Description and Data Collection In our analysis, monthly data of stock prices of renewable energy companies and monthly data of geothermal companies are analyzed. The examination period ranges from January 2004 to June 2011. To analyze the stock prices, 20 different macro- and energy variables were included. The reasoning for choosing this time period is that in 2008 there were major changes in the world economy, which has had dramatic effect on stock prices and investment choices. Therefore, we chose to split up the sample into two periods, the first labeled before crisis and ranges from January 2004 throughout December 2007. The second period is labeled after crisis and ranges from January 2008 throughout June 2011. For the purpose of analyzing stock prices of renewable energy companies, WilderHill Clean Energy Index (ECO) was chosen. This index is meant to define and track the Clean Energy sector and, the stock and sector weights in the ECO index are based on their importance for clean energy, technological influence and relevance to prevent pollution (WilderShares, LLC 2009). For the purpose of the investigation, the returns of the index were calculated using natural logarithm. The ECO index can be accessed through Bloomberg. For analyzing geothermal stock prices, the six largest geothermal companies in North America were chosen for the analysis. Íslandsbanki’s Geothermal Energy Research has compiled an index of six publicly traded geothermal energy companies in North America and they provided us with data from Bloomberg 36

containing historical stock prices for the six companies. In the beginning of 2004 only two of the six companies were listed, in 2005 the number of companies were up to four and in 2009 the last two geothermal energy companies were listed. This caused some problems, as we wanted to analyze historical stock prices and most geothermal energy companies are particularly new on the public market. To analyze the stock prices, we chose therefore to calculate monthly weighted average return, using natural logarithm, for the companies listed at each the time. The calculation of the monthly weighted average return can be seen in Appendix 1. The next step was to get monthly data of the 20 variables we wanted to include in the analysis. Most of the macroeconomic variables were extracted from the DataStream database, but monthly U.S. GDP was found at YCharts.com (2011), the Volatility Index (VIX) was extracted from Yahoo! Finance (2011a) and the consumer price index was found at the U.S. Department of Labor (2011) website. The energy variables were all found at the U.S. Energy Information Administration (2011) website.

4.3 Methodology In order to examine the performance of publicly listed geothermal companies the analysis was separated into two stages for simplification. The first stage includes principal component analysis and the second stage is a multiple regression analysis. The first part involves decreasing the number of variables by extracting several principal components that are composed by the original variables. The second part involves performing multiple regressions using stock returns against the extracted components. This section will however start by explaining the initial process of the analysis by employing a unit root tests to test if the data series is stationary or not. Then, both stages of the analysis will be briefly discussed to illustrate the process of the analysis.

4.3.1 Unit Root Test Before a PCA is performed on time series data, one crucial assumption has to be tested; the stationarity of the series. Michailidis (2009) says that this is a critical assumption, and if time series are stationary, its mean and variance is constant over time and the

37

value of covariance between two periods depends on the distance or lag between the two time periods. Stationarity can often be reached by taking logarithm of the data, which would eliminate correlations and multicollinearity in time series that show a clear trend of, for example, a rise. One way of checking for stationarity is to perform an augmented Dickey-Fuller test (ADF). ADF tests for unit root in time series sample and gives a negative number. The lower the negative number is, the stronger the hypothesis of unit root can be rejected (Economics.About.com 2011; Michailidis 2009).

4.3.2 Principal Component Analysis PCA is a method mainly used to reduce a large number of variables to a much smaller number of PCs whilst retaining as much as possible of the variation in the original variables. Therefore, using PCA we can group the variables into principal components and thereby classify them. In more specific terms, PCA combines the correlated variables into a single principal component and the extraction of the component would amount to a variance maximizing rotation (varimax rotation) of the original variable space (Statsoft 2011a). As the proposed task in this thesis is to test the possible relationship between returns of geothermal companies and diverse factors, it is crucial to distinguish the relevant factors for the analysis. Fifield et al. (2002) say that PCA is a method, which is used to reduce the number of variables to a smaller set of derived orthogonal variables that retain most of the information of the original variables. These derived variables are called principal components. When PCA is performed, the dominant principal components are extracted and used as input in regression analysis, in our case stock performance. Fifield et al. continue by stating that PCA is attractive for two reasons, first, it allows for theoretically important factors to be considered. Second, it can be used in a multiple regression and because the derived variables are orthogonal, they eliminate the possibility of multicollinearity.

4.3.3 A Multiple Linear Regression The term multiple regression was first used by Pearson in 1908, and it is meant to analyze the relationship between a number of independent variables (or predictor variables) and a dependent variable. This is a widely accepted and commonly used 38

method and is a good tool for researchers to find good predictors of the dependent variable (Statsoft 2011b). In the analysis of stock returns, we will use a multiple linear regression to analyze the relationship between the extracted principal components and the performance of listed geothermal stocks.

4.4 Selection of State Variables This sub-section describes the state variables that are included in the principal component analysis. The selected variables were pinpointed out as factors that were most likely to influence stock returns in both the U.S. and the Canadian geothermal industry. The variables that are analyzed are of some economic interest and many of them have been widely used in the financial literature. To get a deeper understanding of the topic under discussion, personnel in Alterra Power Corp. gave their view on factors that they believe could influence stock returns. Their contribution is considered as an important input into this analysis. Appendix 2 summarizes the variables included in the analysis. Next, these factors will be briefly discussed and their existence in the analysis will be explained.

4.4.1 Selection of State Macro Variables Stock market prices fluctuations are certainly linked with economic activities, this fact was confirmed by the present financial and economic crisis of October 2008. In theory, stock price expectations are based on economic fundamentals. In a macroeconomic level, these anticipations depend largely on market expectation of future economic growth level. Nowadays, financial markets are even considered as a leading indicator of world economies. Theoretical stock valuation models such as Dividend Discount Model (DDM), Free Cash Flow Valuation, and Residual Income Valuation illustrates well the relationship between stock prices and macroeconomic variables. According to these models, “The current prices of an equity share is approximately equal to the present value of all future cash flows; thus any variable affecting cash flows and required rate of return in turn influences the share value as well” (Ozbay 2009: p.6). The macroeconomic factors

39

included in this analysis are all from the U.S. economy, gross domestic product, industrial production, U.S. Federal Reserve, Treasury bill, Ted Spread, inflation, unemployment rate, export, import and volatility. Gross Domestic Product (GDP): The possible relation between stock market returns and real economic activity growth is captured by the GDP. The GDP indicator can be related to the discussion from the PESTEL analysis where it was stressed that GDP is the single most important factor as it affects other important economic indicators. For this reason a further discussion of economic indicators can often be related to GDP. Naoui et al. (2010) supports the aforementioned view that market expectation depends on future economic growth level. He explains that according to the future cash flow (dividend) actualization model, stock market prices must reflect investor´s anticipations about future real economic activities. Based on this, the fundamental value of stock prices will be equal to the actual value of future cash flows or dividends, which furthermore, is generated by the firm´s real economic activities. Therefore, future cash flows must reflect real economic activity explained by GDP or industrial production. Industrial Production: Industrial Production index is an economic indicator that is released monthly by the Federal Reserve Board. A useful proxy for the level of real economic activity is the industrial production index. The indicator consists of the manufacturing, mining, and electric and gas utilities industries (Board of Governors of the Federal Reserve System 2011). Theoretically, industrial production increases during economic expansion and decreases during a recession. A reason for this is that these sectors included in the industrial production are highly sensitive to interest rates and consumer demand. Thus, a change in industrial production would signal a change in an economy and therefore it is an important tool for forecasting future GDP and economic performance. During economic growth, the productive capacity of an economy rises, which in turn contributes to the ability of firms to generate cash flows. Ozbay (2009) illustrates this as a reason why industrial production can act beneficially on expected future cash flows and therefore explains a positive relationship between real economy and stock prices. The U.S. Federal Reserve: The Federal Reserve, the central bank of the United States, provides the nation with a safe, flexible, and stable monetary and financial system. From an economic perspective a monetary policy influences the general economy 40

through a transmission mechanism, for example, short-term interest rates is one of the main policy instrument to control the economy. Restrictive and expansionary monetary policy might both have bilateral effects. These policies have the purpose of stabilizing the economy which is important as was discussed in the case of Iceland. In case of expansionary policy, a central bank generates excess liquidity by engaging in open market operation which results in lower interest rates and an increase in bond price. Lower interest rates means in a simple term that banks have more liquidity to lend since they have access to cheaper capital. Lower interest rate offered on loan will stimulate investments, which is a typical reaction if an economy is in recession. This reaction is intended to boost the economy so economic activities grow. In case of restrictive monetary policy, a reduction in the money supply would result in a decrease in the supply of funds for working capital and less expansion for all business. For investors, this would mean market interest rate increases which hence raise firm´s cost of capital. Consequently, investment would diminish. This discussion could be related to the Fisher separation. In Fisher separation, firms use the return available in financial markets to determine how much they should invest. The return from financial markets is the discount rate which reflects both the time value of money and the riskiness of the stock. The risk free rate represents the time value of money. A risk premium represents compensation for risk i.e. measured relative to the risk free rate. The discount rate is then perceived by an investor as a required rate of return8. In other words, discount rate is used in investors’ NPV calculations to determine whether to invest or not (Ozbay 2009). Treasury Bill (T-Bill): T-Bill is a short-term debt obligation, which is backed by the United States government and has a maturity of less than one year. Most common maturities are one month, three months or six months. A T-Bill does not pay coupons, instead the value of it, comes from the difference between the discounted value, which is originally paid, and the amount which is received after the maturity (Investopedia 2011b). Whereas T-Bills are also traded on a secondary market, their prices tend to rise when interest rates go down and vice versa. T-Bills are also often called “safe” investments because of the short-term maturity and they are backed by the U.S. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 8

The CAPM is a method to determine the required rate of return, see e.g. Berk &DeMarzo 2011: p.399-

401.

41

Treasury and would only become worthless if the U.S. Treasury itself would go bankrupt. Often investors see T-Bills as the single best place to put their money, when they do not want to be exposed by any risk or need the money back after a short period of time (About.com 2011). Because of the financial crisis and the high volatility in the stock, the three month T-Bill is therefore included, for the purpose of our analysis. TED Spread: TED spread is the price difference between three-month future of U.S. TBills and three months contract for Eurodollars (LIBOR). Often the spread is used as an indicator of credit risk as the T-Bill is considered a safe investment and LIBOR rates thought to reflect the credit ratings of corporate borrowers. The spread is expressed in terms of basis points, which reflects the risk premium that the market has assigned to corporate lending. When the spread increases, default risk is considered to be increasing and vice versa. The historical average of the spread is around 0.5 percent (50 basis points) but in the times of credit crisis it rises, as was the case in 2008 when it reached almost 450 basis points (Investopedia 2011c & Mysmp 2011). Inflation9: In theory, the Fisher Effect attempts to explain the relation between asset returns and inflation. According to the theory, the Fisher hypothesis states that nominal rates should move one-to-one with the expected inflation rate. Over the years the Fisher Effect has also been extended to the stock market, i.e. expected nominal rate of return on stock is equal to expected inflation plus the real rate of return, where the expected real rate of return is independent of expected inflation. Therefore, the Fisher hypothesis predicts there is a positive relationship between stock returns and inflation. In other words, the Fisher hypothesis implies that stocks offer a hedge against inflation. As inflation is growing, nominal stock return grows in parallel, which might be explained through a rise in energy prices. This would leave the real stock returns unaffected (Kumari 2011). The positive relation of stock return and inflation is however the opposite from many empirical findings. Gultekin (1983) for example did not support the Fisher hypothesis when he tested the hypothesis for 26 countries for the period of 1947-1979. Moreover, the analysis revealed that the regression coefficients are predominantly negative. Bahr (2010) argues that high inflation leads to high inflation uncertainty, and high inflation uncertainty dampens economic activity resulting in lower stock return. Choudhry & !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 9

The annual percentage change in a consumer price index is used as a measure of inflation.

42

Pimentel (2010) support this and explain from their empirical investigation of the Fisher Effect, for the stock market, that the market most commonly finds that stock returns and the inflation rate have a negative relationship. That would mean that stock return diminishes with rising inflation, which is surprising since the Fisher hypothesis implies that stocks, as a claim against real assets, should compensate for movement in inflation. Unemployment Rate: The nature of investors is to figure out what the market expects. As discussed, economic indicators influence investors in a way that it can influence their behavior. Unemployment is, for example, an indicator which can be used to predict future economic performances. Birz & Lott Jr. (2011) investigated the effect of macroeconomic news on stock returns and their findings indicate that news about unemployment does affect stock returns. Boyd et al. (2005) find that news about rising unemployment leads to lower expected earnings and therefore, results in lower stock prices. However, in expansion, the same news about rising unemployment leads to lower expected interest rates on government bonds, causing stock prices to rise. U.S. Exports/Imports: Economic growth is partly driven by the balance between export and import (net export). This is the balance of international payments, commonly known as the balance of payments, which is the overall accounting of nation´s international economic activities. Net export is partly controlled by the exchange rate; if a currency is relatively strong it could reflect in an increase of import and less money is received for the exporting products. The difference between export and import implies whether a country runs a trade surplus or a trade deficit which affect GDP. As previously discussed, future cash flows reflect real economic activity and therefore the impact of export and import on GDP is an important indicator to examine when analyzing factors which could influence stock return (Blanchard 2006). Volatility S&P 500 (VIX): The volatility index is an index that measures expectation of volatility in the stock market. Put differently, it measures fluctuations in stock price of the S&P 500 index. The index is also known as the “fear index” because a high VIX represents uncertainty about future prices, which means that investors expect the value of the S&P 500 to fluctuate wildly in the next 30 days. The more quickly a price changes up or down, the more volatile it is. The difference between volatility and implied volatility is that volatility is observed by looking at past data, but implied volatility represents expectations about future fluctuations (Wikinvest 2011).

43

4.4.2 Selection of State Energy Price Variables Coal, natural gas, and oil constitute more than 88 percent of total U.S. energy consumption, of which 23.8 percent is imported. If it is assumed that the three fuels are to serve the same markets, they can be considered as close substitutes and changes in one price will reflect movements in the other prices. This is probably what suppliers of these energy sources have realized and can be supported by the Porter analysis where substitutes limit potential return. Therefore, a ceiling is placed on the price that all suppliers can profitably charge and therefore limit their profits. Also described in the Porter analysis, price is set according to supply and demand. And since future demand is predicted to rise by 49 percent from 2007 to 2035, the suppliers of energy are likely to increase price steadily with the demand according to an estimate found in the World Energy Outlook 2010 (U.S. Energy Information Administration 2010). Oil Prices: Industrialized countries continue to be heavily dependent on oil, which is being continuously imported. Consequently, political events, such as oil embargoes, and collaborative price setting on the part of the oil exporting nations can have drastic effects on the world economy. Changes in the price of crude oil are often considered an important factor for understanding fluctuations in stock prices. These fluctuations can be explained both in a positive and in a negative manner on stock prices. The Financial Times August 21 (2006), for example, attributes the decline of the U.S. stock market to an increase in crude oil prices, which is caused by concerns about the political stability in the Middle East (including terrorist attacks by Islamic militants, the Iranian nuclear program, and the fragility of the ceasefire in Lebanon). Thus, a sudden increase in oil prices can considerably affect the state of the global economy as they can trigger inflationary trends, cause economic slowdowns that could lead to downturns in the world stock markets. In the PESTEL analysis this kind of a situation was described to hinder geothermal developments and investments. Oil price increase is indeed what the world economy has experienced in recent years. For example, oil price have been increasing for the past years and is forecasted by OPEC´s world oil outlook to increase by about 60 percent in 2035 (Organization of the Petroleum Exporting Countries 2011). On the other side, Fridleifsson (1996) points out the opposite effects on stock price. He explains in his work that the growth rate of geothermal development in the past has 44

been highly affected by the prices of competing fuels, especially oil and natural gas on the world market. He supports this by taking an example of the oil crisis in 1978 to 1985, and states that in this period there were very high annual growth rates for geothermal generated electricity. Whether oil price changes have positive or negative effects on stock price, it surely can affect financial performance through economic conditions or competing fuels. Natural Gas Prices: Natural gas and refined petroleum products have been seen as close substitutes in the U.S. industry and electric power generation. Guo & Yucel (1994) for example found that crude oil prices were shaped by world oil market conditions and, U.S. natural gas prices adjusted to oil prices. This statement could be related to Nesbitt’s (2009) findings. He illustrated from his article what determines gas prices and in the following, he explains the relation between oil and gas prices. By plotting out the prices over the past 50 years for crude oil and natural gas, he concluded that it seems they have a similar pattern. The reason however is not necessarily that oil and gas are substitute products. The reason is that both oil and gas depend in the same way on the same variables i.e. interest rates. Coal Prices: In the article U.S. Geothermal Development from Íslandsbanki (2011) it is stated that prices of coal and natural gas followed oil prices to new highs as strong demand raised concerns over future supply. These same fears were reflected in the share performance of public renewable energy organizations, which in many cases outperformed other energy sectors. This is because investors have become more environmentally conscious and therefore looking to alternative means of energy. According to Oberndorfer (2009), coal price developments affect stock returns of European utilities in his sample period from January 2002 to August 2007. Oberndorfer, however, concludes that coal price impacts stock return much less than oil price, which he explains, is the main energy indicator affecting stock return.

4.4.3 Selection of State Energy Variables Several energy variables are included in the analysis, which is intended to explain whether there is a link between geothermal energy stock prices. The main energy

45

variables are energy production10, energy consumption11, CO2 emissions12 and electricity prices13. Since the world energy economy has a significant influence on the decisions that people and governments make and the current global consumption rates are diminishing the planets ability to sustain their way of life; increased energy demand would mean increasing prices in most sectors of the world economy. This is where alternative energy companies are creating new economies, in order to facilitate a more sustainable market (Alternative Energy 2011). Jennifer Kho (2009) said that although there is little real-world connection between most renewables and oil, there is an interesting correlation between them; when oil prices rise, clean-energy shares tend to rise as well. Although oil is mainly used to make fuel, and for example solar, wind and geothermal power produce electricity, investors still make the assumption that increasing oil prices will benefit all renewables. One of the reasons for this relation could be that political support, which has been widely discussed in this paper, and increased public awareness, were inducing serious investment in the renewable sector. Although renewables stock prices have followed oil prices in the past, investors are getting more and more familiar with the renewables sector and could soon realize that the price of oil does not change the outlook for renewables stocks. This could imply that there could be a change in prospects and signs of maturity in the renewable-energy sector that might eliminate the fact that oil prices are triggers for investing in, or divesting from it (Kho 2009). Because the analysis of the thesis is focused on listed geothermal energy companies, we found it important to include these elements for the purpose of getting a broader overview of variables that could influence stock prices.

4.5 Summary Twenty different macroeconomic and energy variables were chosen to analyze the changes in stock prices in the geothermal industry and the Wilder (ECO) index. The examination period ranges from January 2004 to June 2011. The examination period was then split in two, so we had two periods, which are called before crisis and after crisis. The background of APT and the methodology used in the analysis of geothermal !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 10

i.e. Total Renewable Energy Proudction & Primary Energy Production. i.e. Total Primary Energy Consumption & Renewable Energy Consumption. 12 i.e. Total Energy CO2 Emissions. 13 i.e. Average Retail Price of Electricity & Electricity End Use. 11

46

stock returns were briefly discussed. The decision to use PCA, was mainly to reduce the number of variables in the analysis and to get a smaller set of derived variables. By performing PCA with varimax rotation, the new set of variables are forced to be orthogonal, which in turn eliminates the possibility of multicollinearity and can be used in a multiple regression.

47

5 Principal Component Analysis The descriptive statistics of the time series used in the analysis can be found in Appendix 3. There, high values of standard deviation can be observed in a couple of variables especially in; Natural gas price, oil price, US imports, US exports and the Volatility index. This can be related to fluctuating prices of oil and natural gas over the past years and the change in imports, exports and volatility can be related to the economic crisis. Figure 5 below illustrates the change in exports, imports and volatility from January 2002 to June 2011. Figure 5: Change in Exports, Imports (left axis) & Volatility index (right axis) $'!!!!"

*!" )!"

$!!!!!" (!" '!"

#'!!!!"

#!!!!!"

&!"

-."/012345"

%!"

-."6712345"

$!"

869"

'!!!!"

!"

$!!$+!#" $!!$+!'" $!!$+!," $!!%+!#" $!!%+!'" $!!%+!," $!!&+!#" $!!&+!'" $!!&+!," $!!'+!#" $!!'+!'" $!!'+!," $!!(+!#" $!!(+!'" $!!(+!," $!!)+!#" $!!)+!'" $!!)+!," $!!*+!#" $!!*+!'" $!!*+!," $!!,+!#" $!!,+!'" $!!,+!," $!#!+!#" $!#!+!'" $!#!+!," $!##+!#" $!##+!'"

#!" !"

Source: Own Creation; DataStream; Yahoo! Finance (2011a). As we can see, in 2008 there is a massive slowdown in the U.S. economy and the volatility index rises from under 20 points in 2007 to about 70 points in 2008. This is why the analysis in this thesis is split up into two periods. In Figure 6 we can see that oil prices had been increasing until 2008 where there is a sharp decrease in oil prices and in gas prices as well. As mentioned earlier, there has been a correlation between renewable energy stocks and oil prices and the increasing oil prices in the past can be the cause for increasing renewable energy stock prices. But in 2009 the oil prices starts to increase again, while the renewable stock price seems to continue to fall. This can be because of the decoupling of the price between oil and gas in 2009. Íslandsbanki (2011) supports this view and say that the recent increase in oil

48

prices have turned the focus of investors to alternative sources of energy, that being said, geothermal energy stocks have fallen 75 percent since the start of 2010. They confirm that this could be due to the decoupling of gas and oil prices as they consider natural gas as a clean fuel source and therefore a direct competitor to renewable energy. Figure 6: Correlation between Oil and Natural Gas Prices *!!" )!!" (!!" '!!" &!!"

-./"01.23" %!!"

456715/"859"01.23"

$!!"

!"

$!!$+!#" $!!$+!'" $!!$+!," $!!%+!#" $!!%+!'" $!!%+!," $!!&+!#" $!!&+!'" $!!&+!," $!!'+!#" $!!'+!'" $!!'+!," $!!(+!#" $!!(+!'" $!!(+!," $!!)+!#" $!!)+!'" $!!)+!," $!!*+!#" $!!*+!'" $!!*+!," $!!,+!#" $!!,+!'" $!!,+!," $!#!+!#" $!#!+!'" $!#!+!," $!##+!#" $!##+!'"

#!!"

Source: Own Creation; DataStream.

5.1 Univariate Test for Unit Roots In the starting phase of the analysis, a crucial assumption needed to be checked before the analysis went any further; to check for stationarity in the time series we were working with. To do this, we performed an Augmented Dickey-Fuller test in Eviews on all time series variables. The results showed that all the variables in the analysis were in fact non-stationary. To reach stationarity, the method of taking logarithm of the data and calculate the returns is often used. This is because many time series tend to rise, which can lead to time series to encounter spurious correlations and even multicollinearity (Michailidis 2009). Next we used logarithm to get derived series of our initial time series. The derived series can be seen in Appendix 4. Using these derived series, the assumption that all the series were in fact stationary could be made as Augmented Dickey-Fuller test statistics, for all the derived variables, were lower than the critical values at the one percent level. As well, the Durbin-Watson statistics were just below or

49

over 2 in most instances which indicates that there was no autocorrelation. One variable however had a Durbin-Watson statistic of 1.4533 but it is acceptable as it was significant at the one percent level in the Augmented Dickey-Fuller test.

5.2 Carrying Out the Principal Component Analysis This section will analyze the data with a method named Principal Component Analysis (PCA). As discussed before, the original purpose of PCA is to reduce a large number of variables to a much smaller number of components whilst retaining as much as possible of the variation in the original variables. Furthermore, it is used to detect the structure in the relationship between variables in each component and possible correlation. Therefore, PCA analysis is applied as a data reduction or a structure detection method and for the purpose of this study the tests were conducted using the SPSS data analysis program. The PCA involves the subsequent steps which will be explained in more detail in the following sub-sections. 1. Get correlation matrix, communalities and check the test reliability. 2. Get eigenvectors, which are the direction of principal components. 3. Choice of the rotation method and doing the rotation. 4. Naming the factors and calculation of the combined variables.

5.2.1 Correlation Matrix, Communality and Test Reliability Initially, there were 20 variables that were used as a starting point for the analysis. These variables have been introduced earlier in the thesis. A critical test was made to see if all variables were independent, with no requirement of equal variance. The next step was to start the PCA analysis but before the analysis started, criteria for the subsequent tests needed to be analyzed. We found out that the criteria for the principal component analysis differs in many researches and it seems the interpretation of the principal component depends mostly on the researcher. We therefore tried to follow the most common criteria in each subsequent step, while making assumptions on our own to get the principal components we found a good fit for the purpose of the analysis.

50

Therefore, a correlation matrix between the variables needed to be examined was created. Variables that had too high correlation coefficient (above 0.9) with other variables were excluded to prevent maintaining more than one variable that seem to be measuring the same thing (Michailidis 2009). Thus, to reduce the variance between the related variables they were excluded from the analysis. From Appendix 5, it turned out that two variables needed to be excluded, which left 18 variables for the PCA. The excluded variables were the CO2 emissions and renewable energy consumption (REC). Appendix 6 shows the communalities extraction. The initial extraction is 1.0 for all variables and the final communality shows the proportion of the variance of that variable that can be explained by the common factors. In other words, some variables are not well suited to measure what they are meant to measure. Those variables do not correlate highly with the others and get low loadings. From Appendix 6 we can see that the variables TED Spread and coal prices have extracted value below 0.514 and are therefore excluded from the analysis. A fundamental process is to examine additional criteria for the appropriateness of the analysis and the sampling adequacy. The Kaiser-Meyer-Olkin (KMO) and Bartlett´s test for sphericity explains whether we should actually be doing a principal component analysis to begin with. The minimum requirement of KMO value is 0.5 (Michailidis 2009; Ahmadi et al. 2010) and from Appendix 7, the overall KMO for the examined series has acceptable value at 0.6129 which indicates that the test is reliable and can be used. PCA also requires that the probability associated with Bartlett´s test of sphericity be less than the level of significant (5 percent). The probability associated with the Bartlett test is zero which satisfies this requirement. These two tests imply that there is at least one statistically significant correlation in the correlation matrix.

5.2.2 Choice of the Number of Factors Now that we have calculated the eigenvalues, we can see how much variance each of the principal component extracts from the total. Because we are using PCA to reduce the number of variables, the next step would be to find how many factors to extract. As !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 14

Michailidis (2009) proposed that communality should be more than 0.5, while Ahmadi et al. (2010) proposed that it should be more than 0.6.

51

we can see in Appendix 8, the first factor has the highest eigenvalue and as we extract consecutive factors, they account for less and less variability and the decision to stop extracting factors, does in fact depend on when there is little “random” variability left (Statsoft 2011a). There are two methods often used to find how many factors to retain, the first is to retain all principal components that have an eigenvalue greater than 1. This means that unless a principal component extracts at least as much as the equivalent of one original variable, we have to drop it. This is a widely used method and was first proposed by Kaiser in 1960 (Michailidis 2009). Another method of determining how many principal components to retain is called the scree test, which is a method that was proposed by Cattell in 1966. When using the scree test, where eigenvalues are plotted in a simple line plot, Cattell suggested that you had to find the place where the smooth decrease of eigenvalues seem to level off to the right of the plot (Statsoft 2011a). Taking a look at the eigenvalues in our analysis, we can see in Appendix 8 that five out of sixteen factors have values greater than one. This would imply that according to the Kaiser criterion we would extract five principal components and their cumulative percentage of total variance is 68.37 percent. When using the second proposal, introduced by Cattell, we can see from the scree plot in Figure 7, that there seems to be a break in the plot after two factors and after five factors. As the two first factors only represent roughly 30 percent of total variance we can with the assistance of Kaiser Criterion, conclude that we should extract five factors from the sixteen variables.

52

Figure 7: Scree Plot

!

"

#

$

* $

#

"

!

%

&

'

(

)

$*

$$

$#

$"

$!



%$$&

+,-.,/0/1234-506

Source: Own Creation.

5.2.3 Choice of the Rotation Method and Doing the Rotation Henry F. Kaiser (1958) said that the analytic criterion for rotation in factor analysis is defined as one that imposes mathematical conditions beyond the fundamental factor theorem, which means that a factor matrix is uniquely determined. Hervé Abdi (2003) discussed when and why factor rotation is used and said that rotation is performed mainly to simplify the factor structure and therefore makes the interpretation easier and more reliable. The orthogonal rotation method, varimax, was developed by Kaiser in 1958 and is by far the most popular rotation method. When using varimax rotation, each factor has a small number of large loadings and large number of small loadings, which simplifies the interpretations of the factors. In addition to this, some factors can also be interpreted from the opposition of a few variables, for example when there are few variables with positive loadings and few variables with negative loadings. Michailidis

53

(2009) adds that although this method maximizes the variation of the factor loadings, the factors are still independent. The rotated component matrix can be seen in Table 2 below. There it can be seen how the variables are classified into each component. We can see that four of the components include three variables and one component has four significant variables. Furthermore, all the variables have loadings higher than 0.3 on at least one component, which according to Michailidis (2009) is often a criterion used to see if the variables explain substantive variance. Table 2: Rotated Component Matrix

!PEP !TPEC !TREP !FED !TBILL !VIX !EX !UR !IM !GDP !INF !NGP !OP !ARPE !EEU !IP

1 0.8977 0.8667 0.8322 -0.0901 -0.1606 -0.0591 0.0781 -0.0097 0.1145 -0.1183 -0.1918 0.1267 0.0522 -0.0311 0.5665 0.0803

Principal Components 2 3 4 -0.0412 0.1192 0.0200 -0.0384 -0.0597 -0.1427 -0.0965 0.0105 0.1763 0.8604 0.2292 0.0900 0.7766 0.2426 -0.0183 -0.6467 0.2134 -0.1589 0.2617 0.7028 0.2051 -0.1664 -0.6923 0.2122 0.0402 0.6871 0.4414 -0.1172 0.5790 0.2598 0.3083 0.1993 0.7708 -0.1257 0.0632 0.6835 0.4183 0.1420 0.6617 0.0944 -0.1190 0.1895 0.0291 -0.1348 -0.0066 -0.0377 0.3342 -0.0421

5 0.0721 0.2733 -0.1153 0.0741 0.2000 0.2546 0.0550 -0.0592 0.0740 -0.1001 0.1636 -0.0699 0.1657 0.8527 0.7239 0.6181

Source: Own Creation.

5.2.4 Naming the Factors and Calculation of the Combined Variables Earlier in the analysis we have discussed how we got the five factors and the factor loading matrix shows how each variable fits into the components. The variables should only show high loadings in one factor each. We have also discussed the communalities between the variables, where we have chosen to retain only variable that were under 0.5, as it loaded well in the factor loading matrix. To get this result, we first used the un-

54

rotated solution to see how many factors we should extract and when we came to the conclusion to retain five factors, we rotated the factor solution to get orthogonal factors and make them easier to interpret. As discussed earlier we chose to retain five factors instead of two as it is often better to extract more factors than less. Michailidis (2009) said that although extracting the correct number of factors is always the best solution, it could be a good strategy to lean towards over-extraction rather than under-extraction to avoid the greater error found with under-extraction. He continues by stating that general basis of large loadings is when values are above 0.3 and a well-defined factor should have at least three variables loading highly on it. So when a factor has for example only one high loading, it would indicate that the numbers of factor that have been extracted are too high. The next step would be naming the five factors we have extracted, and the first component includes three energy variables that have high loadings; primary energy production, total primary energy consumption and total renewable energy production. This component can therefore be labeled as the “Energy production factor”. The second factor also has three high loadings; the variables that load high on the factor are U.S. Federal Reserve short-term interest rate, 3-month U.S. Treasury bill and the implied volatility index. This is a very interesting combination, as those variables tend to change in economic crisis, and the factor can be seen on Figure 8. Figure 8: The Crisis Factor $# %# &# !&#

./0101#23456/#

!%#

!"#

'((')(&# '((')($# '((')(*# '((%)(&# '((%)($# '((%)(*# '((+)(&# '((+)($# '((+)(*# '(($)(&# '(($)($# '(($)(*# '((,)(&# '((,)($# '((,)(*# '((")(&# '((")($# '((")(*# '((-)(&# '((-)($# '((-)(*# '((*)(&# '((*)($# '((*)(*# '(&()(&# '(&()($# '(&()(*# '(&&)(&# '(&&)($#

!$#

Source: Own Creation.

55

As we can see from Figure 8, there is a huge drop in the crisis factor in 2008. This is a result of a massive drop in interest rates and subsequently a very high implied volatility for the next year. This would lead to the conclusion that we name the second component the “Crisis factor”. We can see from the crisis factor that from the start of 2007, the U.S. economy appears to be experiencing a small crisis or heading towards crisis as the volatility index is rising and the interest rates start to drop. After the crisis, the crisis factor seems to be very unstable, which would imply that the economy is still recovering from the crisis and the volatility on the market is still above the historical average. The third component is classified as the “Macroeconomic factor” since all the variables in this category were discussed in the macro variable section (4.4.1). This component consists of the following U.S. economic variables; export, import, unemployment and GDP, which are all highly correlated as discussed in the PESTEL analysis earlier. There it was stressed that GDP is the single most important factor because GDP affects other important economic indicators. These indicators are important for investors because they can be used to predict future economic performances. The fourth factor includes three price related variables, namely; inflation, natural gas price and oil price. Therefore, we can name this component the “Inflation factor”. We chose to use the consumer price index to calculate monthly changes in inflation instead of using the core inflation index. The reason for using the CPI is that energy prices in total only accounted for 6.6 percent in 2001 and rose to 8.6 percent in 2009 in the United States (OECD 2011a). We assume that this is such a small amount of the total CPI that energy prices will not be too correlated with the CPI in the analysis. The fifth component can be named the “Electricity factor” due to the fact that the variables’ average retail price of electricity (ARPE) and electricity end usage (EEU) has the highest loadings. Another variable in this component is industrial production (IP) which could relate to the others through demand and supply for electricity. Since the industry depends on electricity in its production, it is safe to assume that if there is an increase in the production level, the demand for electricity rises.

56

The correlation between a greater demand for electricity and price leads to a conclusion that supply for electricity increases. Based on this simple economic reason, price for electricity should rise with an increasing demand from the industry.

5.3 Summary Before analyzing the stock returns, principal components were derived from the twenty original variables. Before the PCA was carried out, a unit root test was performed where we checked for stationarity in the time series we were working with. To reach stationarity, we used the method of taking logarithm of the data and calculated the returns. The next step of the analysis was to remove variables that were too correlated with each other and exclude the variables that showed too low communalities with other variables. The KMO test and Bartlett´s test for sphericity were performed and they implied that there is at least one statistically significant correlation in the correlation matrix and this confirmed that PCA can be performed on the data. Next the eigenvalues were calculated and plotted up in a scree plot. We decided to extract five principal components, and used the orthogonal rotation method, varimax, to extract the rotated component matrix. This method maximized the variation of the factor loadings while still keeping the factors independent. When the five factors had been extracted, the next step was to look into the factors and name them. The names of the factors are: Energy production factor, Crisis factor, Macroeconomic factor, Inflation factor and Electricity factor.

57

6 Multiple Linear Regression This section will reveal and describe the results from the regression analysis. The examined period covers eight years and six months from January 2004 to June 2011. This choice is motivated by the fact that the examined period incorporates the characteristics of a changing economic environment. The tests have been made firstly for the period from January 2004 to December 2007 that will be labeled as the period before crisis. By contrast, the second period ranges from January 2008 to June 2011 and represents the period after crisis. As stated before, the main reason for separating the time period is to investigate the major changes in the world economy, which has had dramatic effects on stock prices and investment choices. Events on the world’s financial markets before crisis led investors to have easier access to liquidity, based on bank offerings. The oil prices were rising rapidly in the period before the crisis, which could have led investors to look at alternative sources of energy and, supporting green energy was getting more popular. After the crisis, there is a visible decoupling of natural gas and oil prices, which could affect the renewables sectors as well. Fekedulegn et al. (2002) say that principal component regression (PCR) is a technique used to handle the problem of multicollinearity and produce meaningful estimates for regression coefficients. Furthermore, because macroeconomic variables can sometimes be highly intercorrelated, the use of ordinary least squares (OLS) to estimate the parameters of the response function can result in instability and variability of the regression coefficients. So, when PCR is used successfully it could result in better estimation and prediction than OLS. In the last chapter the initial variables were transformed to orthogonal principal components and the five components that contributed the most to the total variance were chosen while the rest was eliminated. The next step is to perform a multiple regression analysis by using the stock returns from the Wilder index and from the North America Geothermal index against the reduced set of variables (the principal components).

6.1 WilderHill Clean Energy Index (ECO) The Wilder ECO index consists of listed companies on a major U.S. stock exchange in the Clean Energy sector. Market capitalization for a majority of Clean Energy index 58

stocks is typically $200 million and above. The index uses modified equal dollar weighting which means that no single stock may exceed 4 percent of the total Clean Energy index weight at the start of quarterly rebalancing (WilderShares, LLC 2009). Using monthly data, two regression equations have been examined. Consider first the following multiple linear regression model: !!"#$%& ! !! ! !!! !"#$ ! !!! !"#$#$ ! !!! !"#$% ! !!! !"# ! !!! !"!#$ where, RWILDER is the Wilderhill Clean Energy index (ECO). The Energy Production factor is the first principal component and is called ENPR, CRISIS is the Crisis Factor, MACRO is the Macroeconomic factor, INF is the Inflation factor and ELECT is the Electricity factor. In the regression model presented above, the size of the coefficient for each independent variable gives the size of the effect that each variable is having on the dependent variable and the sign of the coefficient (positive or negative) gives the direction of the effect. The coefficient expresses how much the dependent variable is expected to increase when the independent variable increases by one, holding all other variables constant. From the equation these effects are represented by the parameter beta (!). When estimating the regression model, a natural question arises of how well the estimated regression line fit the observations. A measure of goodness of fit uses the proportion of the variance of the dependent variable that is explained by the model, the rest is in the error term. The variable is called the R2 and is only appropriate to use if the model is estimated by OLS. Sometimes the R2 is interpreted as a measure of quality of the statistical model, while in fact it measures nothing more than the quality of the linear approximation. Verbeek (2008) explains that when there is more than one independent variable included in the model it would be better to use adjusted R2. He explains that too many variables will not be able to explain the model´s coefficients, as they may be estimated rather inaccurately. One reason for this is that the R2 will never decrease if the number of independent variables is increased, even if the additional variables have no real explanatory power. The adjusted R2 allows for degrees of freedom associated with the sums of the squares. Therefore, even though the residual sum of square decreases or remains the same, as new independent variable are added, the residual variance does not. For this reason, adjusted R2 is generally considered to be

59

a more accurate goodness-of-fit measure than R2. The F-test then measures the overall joint significance of the model. More formally, it is a test of whether all of the coefficients are jointly equal to zero. If they are, the model is not really explaining anything. The critical value, when using a multiple linear regression, is the t-value which explains how confident we can be that each individual variable has some correlation with the dependent variable. The results of the regression analysis crisis of the Wilder Eco index can be seen from Table 3: Table 3: Regression Analysis of the Wilder (ECO) Index Energy Production

Crisis Factor

Macro Factor

Inflation Electricity Factor Factor

Wilder Index Total coefficent std.error t-test p-value

0.0033 0.0089 0.3766 0.7074

0.0547 0.0080 6.7967 0.0000

0.0048 0.0083 0.5809 0.5629

0.0304 0.0085 3.5797 0.0006

Wilder Index Before Crisis coefficent std.error t-test p-value

-0.0062 0.0113 -0.5512 0.5844

0.0141 0.0303 0.4638 0.6452

0.0014 0.0173 0.0801 0.9366

Wilder Index After Crisis coefficent std.error t-test p-value

0.0138 0.0149 0.9262 0.3605

0.0567 0.0099 5.7283 0.0000

-0.0014 0.0115 -0.1192 0.9058

Adj.R 2

F-test

P-value

-0.0037 0.0088 -0.4195 0.6759

0.3860

12.1922

0.0000

0.0114 0.0124 0.9124 0.3668

0.0005 0.0108 0.0493 0.9609

-0.0825

0.2836

0.9195

0.0384 0.0135 2.8462 0.0073

-0.0130 0.0146 -0.8923 0.3782

0.5090

9.4996

0.0000

Source: Own creation. The results of the regression analysis of the Wilder index before crisis indicates that the model does not explain adequately the variation of stock returns for the period 2004 to 2008. All of the calculated variables are not statistically significant with t-value less than 2. Additionally, the model provides poor adjusted R2 results with a negative value of 0.0825. The adjusted R2 should be carefully interpreted as it can take on any value less than or equal to 1, where a value closer to 1 indicates a better fit. Negative value can mean that the model contains terms that do not help to predict the response (The University of New South Wales Sydney-Canberra-Australia 2010). A more interesting result of the regression analysis after crisis (2008-2011) indicates that both the Crisis factor and the Inflation factor appear to be statistically significant, where both of the factors show t-values greater than 2. Also the adjusted R2 improves 60

significantly as it measures at 0.5090, and the F-value significant as a goodness of fit. The relationship is positive, which means the Crisis factor and the Inflation factor tend move in the same direction as the Wilder index. This means that if the Crisis factor goes down, indicating a deeper crisis, the Wilder index seems to lose value as well. If the Inflation factor goes up, the stock prices in the Wilder index seem to react positively, which does not come as a surprise as two of the three variables making up for the Inflation factor are natural gas prices and oil prices. In a statistical term, the low standard errors and high t-tests resulting in low p-values indicate that the results can be confirmed with a 95 percent certainty (" = 5%). This can be seen from the highlighted figures in Table 3. Other factors in the analysis do not seem to have any significant impact on the returns in the Wilder index. It is surprising to see that both the Macroeconomic and the Energy Production factor showed an insignificant result in relation to the Wilder index. Throughout this thesis there has been a discussion that the macroeconomic factors are interconnected i.e. they affect each other, which eventually could slow down the economy. This has mainly been explained with a GDP variable. Furthermore, there are strong arguments from several researchers, e.g. Birz & Lott Jr. (2011), who comment on the relation between stock returns and macro variables. They say that, such disconnections lead us to conclude that the U.S. clean energy stock markets are not efficient and that stock prices do not depend on economic fundamentals. Instead, they are the consequence of speculative investors behavior. Fifield et al (2002) say that the influence of GDP in the analysis can mainly be expected in the emerging markets, which have been characterized as high-growth economies, and have demonstrated superior economic growth rates over a long period of time relative to developed economies. Initially, the Energy Production factor was also likely to correlate with the Wilder index because an increase in the production level is likely to result in a higher price level. In fact, the price level of energy over the past few years has risen significantly. A forecast made by the World Energy Outlook in 2010 predicts that future energy demand will rise by 49 percent from 2007 to 2035, which means that suppliers of energy are likely to increase price steadily with the demand (U.S. Energy Information Administration 2010). Perhaps, this factor will have a stronger influence more in the near future and could therefore be useful for further researches in this field.

61

6.2 North America Geothermal Index The Geothermal index consists of six of the largest geothermal companies in North America. Since most of the companies are relatively newly listed on the stock market, a weighted average returns for the companies, listed at each time in each month, was calculated for the purpose of this analysis. In 2004 only two companies were listed and had historical monthly stock prices available, but in 2009 all six companies used in the analysis were listed. The calculations of the monthly weighted average return of the six largest geothermal companies in North America can be seen in Appendix 1. The second regression equation is as follows: !!"# ! !! ! !!! !"#$ ! !!! !"#$#$ ! !!! !"#$% ! !!! !"# ! !!! !"!#$ where, RGEO is the North America Geothermal index. The Energy Production factor is the first principal component and is called ENPR, CRISIS is the Crisis Factor, MACRO is the Macroeconomic factor, INF is the Inflation factor and ELECT is the Electricity factor. The results of the regression analysis of the North America Geothermal index can be seen from Table 4: Table 4: Regression Analysis of the North America Geothermal Index Energy Production

Crisis Factor

Macro Factor

Inflation Electricity Factor Factor

Geo Index Total coefficent std.error t-test p-value

0.0075 0.0135 0.5551 0.5803

0.0600 0.0123 4.8961 0.0000

-0.0053 0.0123 -0.4164 0.6782

0.0335 0.0129 2.5859 0.0114

Geo Index Before Crisis coefficent std.error t-test p-value

0.0075 0.0172 0.4338 0.6667

-0.0036 0.0464 -0.0780 0.9382

-0.0185 0.0265 -0.6970 0.4897

Geo Index After Crisis coefficent std.error t-test p-value

0.0095 0.0202 0.4690 0.6419

0.0568 0.0134 4.2383 0.0002

-0.0236 0.0156 -1.5118 0.1393

Adj.R 2

F-test

P-value

0.0099 0.0134 0.7371 0.4631

0.2354

6.4803

0.0000

0.0048 0.0191 0.2531 0.8014

0.0205 0.0165 1.2382 0.2225

-0.0496

0.5557

0.7332

0.0532 0.0183 2.9091 0.0062

-0.0103 0.0197 -0.5245 0.6032

0.3974

6.4071

0.0002

Source: Own creation.

62

The results from the regression analysis of the North America Geothermal index before crisis does not explain the variation in stock returns for the period from 2004 to 2008. This result is similar to the Wilder index in the previous sub-section where all variables were not statistically significant with t-value less than 2. This model produces poor adjusted R2 results with a negative value of 0.0496 and the overall test significance through the F-test does not fulfill the requirements at the five percent significance level. The outcome from the analysis of the North America Geothermal index after crisis (2008-2011) give similar results as using the Wilder index. This supports our findings that the Crisis factor and the Inflation factor do correlate positively with stock returns of the six largest geothermal companies in the U.S. and the Wilder Clean Energy index, which is based upon renewable energy companies. Although the results are similar, the adjusted R2 is significantly lower (0.3974) when the factors are regressed on the geothermal index, which indicates that the predictive power is not as strong when using the model on geothermal companies as when using it on the Wilder index. This can be due to the limitations of the geothermal index, as it contains only six companies and some of the companies were not listed until 2009. This would mean that bad publicity or performance in one company has a dramatic effect on the index as a whole. Despite this limitation, the adjusted R2 is acceptable and it shows a significant F-value as a goodness of fit. We can thereby confirm with 95 percent certainty that the Crisis factor and Inflation factor are statistically significant when explaining deviations in geothermal stock prices. Just like the Wilder index, other factors in the analysis do not seem to have any significant impact on the North America Geothermal index. The statistics can be seen in Table 4.

6.3 Summary This chapter introduced the multiple linear regression models used to analyze the stock returns for the Wilder index and the North America Geothermal index. The results showed that the Crisis factor and the Inflation factor are both statistically significant when it comes to explaining the stock returns both in the Wilder index and the Geothermal index after the crisis. On the other hand, no factors were statistically significant before

the crisis. This can be a consequence of speculative investors

behavior. These results did however differ when the predictive power of the models was 63

observed. The adjusted R2 was 0.5090 after the crisis when regressing on the Wilder index, while the adjusted R2 for the Geothermal index was 0.3974 during the same period. This indicates that the predictive power is not as strong when using the Geothermal index, and the reason could be that the Geothermal index does have it limitations when it comes to number of companies listed and how much effect each company has on the index as a whole. Other extracted principal components did not show any statistical significance in either of the regressions.

64

7 Analyzes of the Significant Factors from the PCA The previous chapter showed that there is statistical significance between two of the factors produced, namely the Crisis and Inflation factor, and the geothermal index and the Wilder (ECO) index. Now these factors and the variables that make up for those factors will be analyzed in more detail. To see the relationship between the factors and the initial variables, the time series are plotted to see how they behave, and if they share some common attributes. By doing this we are trying to make sense of the statistically significant principal components.

7.1 The Crisis Factor The first factor that showed statistical significance was the Crisis factor, which included the following variables; US-Federal short term interest rate, 3-month T-bill, and the VIX volatility index. Figure 9 shows the relationship between interest rates and the Wilder index. Figure 9: Relationship between Interest Rates (right axis) and the Wilder Index (left axis) #)!"

("

#(!" '" #&!" #$!"

&"

#!!" %" )!" (!"

$"

-./012"34015" 6789:;"