1
University of Groningen (UG)
European electricity prices and alternative
energy companies’ stock prices
Master Thesis
MSc Finance
Robin Klaus
Student-No. 3034119
2
Table of contents
Introduction ... 3
Literature Review ... 5
Methodology ... 9
Data & Variables ... 13
Vector autoregression & discussion ... 19
Conclusion ... 30
References ... 33
3
Introduction
In recent years alternative energy as means of energy production became more widespread. Electricity producers having to participate in the European emissions trading scheme, subsidies as well as shifts in taxation, increased the incentives associated with renewable measures of energy production (Mulder and Scholtens, 2013). As renewable approaches of energy production such as electricity generation from windmills, solar cells or biomass are characterized by very low costs of production in comparison with conventional measures, an effect from a relative increase in these measures of production, on electricity price levels is to be expected in the short-run. In the long-run prices are anticipated to average long-run costs.
Beside the hypothetical impact, this strengthening of alternative energy companies had in recent years, there are also other factors hypothetically affecting electricity prices on a great scale.
Most noticeably, conventional fuels such as gas or oil should have a great impact on the production costs of electricity, as generation from these fuels does constitute the
greatest share in the generation mix (Eurostat, 2016). Additionally, influencing the costs of electricity generation from these measures are, since its introduction in 2005
(European Commission, 2017), European CO2 Emission Allowances as these directly affect the cost factor in generating electricity from combustible fuels. Therefore,
European CO2 Emission Allowance prices (EUAs) should have an impact on overall electricity prices, just as prices of the fossil fuels used in electricity generation
themselves.
Support for the assumption that electricity prices should decrease at least in the short term as reaction to an increase of alternative energies in the generation mix is
4
However, all of the following factors generally determine electricity prices at production level: Fuel costs (which have a great impact on marginal costs), operating expenses of the plants and the transmission/distribution systems, weather conditions and regulations. Here, European CO2 Emission Allowances constitute the investigated variable
representing interfering regulation. An influence on the stock performance of electricity firms caused by variations in EUA prices has already been attested (Oberndorfer, 2009). Thus, an empirical relationship between electricity prices and prices of European CO2 Emission Allowances was anticipated.
Further, a positive relationship between gas prices and electricity prices was expected in the subsequent analysis, which focused on investigating the empirical relationship between the different variables of interest (electricity prices, EUA prices, gas prices, stock prices of renewable energy companies) by estimating a vector autoregressive model.
Beside the absence of fuel costs in renewables, the steady Europe-wide policy shift towards incentivizing alternative energies does provide an argument for the anticipation of a Europe-wide decrease in electricity prices at production level, even though fossil fuels should remain the main determinant of the costs of production.
An increase of alternative energies in the generation mix, also comes with an increase in susceptibility towards weather conditions and therefore with a risen volatility in electricity production output (available capacities) (Mulder and Scholtens, 2013). While these measures of energy production could have a transnational impact on electricity prices in Europe, a scenario in which electricity prices affect alternative energy companies’ performance is also conceivable.
5
As renewable power plants, remain engaged, independent from current levels in demand, they present intangibility regarding electricity prices. Given this, it is safe to conclude that an increase in electricity price should at least have a short term, positive impact on the financial performance of companies affiliated with electricity generation from renewable measures.
Thus, certain electricity price levels might have the potential to benefit electricity
production from renewable resources in contrast to conventional resources, especially in an altering environment such as the European Union. Under these conditions, it is
unidentified, how renewable energies and electricity prices affect each other and to what extent. This thesis aims to examine the empiric relationship between both of these variables under inclusion of European CO2 Emission Allowances and European gas prices as additional factors holding some anticipated predictive power over the variables included.
In order to investigate the sought empirical relationship, this thesis answers the following research question: What is the empirical relationship between electricity prices and alternative energy companies’ stock prices in Europe under consideration of European CO2 Emission Allowances and European gas prices? By estimating a vector
autoregressive model and consecutively applying impulse response tests it could be determined that electricity prices are indeed affected by stock prices of alternative
energy companies. Further, an empirical relationship between gas prices and alternative energy companies’ stock prices has been exposed.
Literature Review
The influence of macroeconomic variables such as energy prices on stock markets has been investigated in multiple scenarios. Here, closest in answering the research
6
While both papers conducted their research on different periods (2001-2006 & 2008-2012) the methodologies used differ as well. Sensfuß et al. used an agent-based simulation platform while Cludius et al. conducted a regression analysis, utilizing hourly spot market prices for electricity prices and official feed-in figures of renewables and other energy sources. Nonetheless, both come to the overlapping conclusion that
renewable electricity generation has a considerable impact on electricity market prices in Germany (Sensfuß, Ragwitz and Genoese, 2007).
There is a positive relationship between the increase of renewables in Germany’s generation mix and the observed decrease in electricity prices. The two papers prove these findings by arguing with the occurrence of the merit order effect. The merit order effect is the displacement of power plants with high marginal costs by the market entrance of generating plants, able to produce electricity at comparatively lower costs. As electricity prices are still solely determined by supply and demand in the market, the electricity price decreases accordingly, if the marginal power plant is now a generating plant, which produces electricity at a lower cost than before the emergence of alternative energy.
However, it is unidentified if growth in alternative energy companies caused a Europe-wide decrease in power prices, what makes it worthwhile to expand the research conducted on the empiric relationship between both of these variables. Additionally, it has to be examined if power prices do as well have the potential to influence alternative energy companies’ performance. A focus on variables having a Europe-wide impact is put in the works of Oberndorfer (2009) and Da Silva, Moreno & Figueiredo (2016) as they investigated the influence of EU Emission Allowances (EUA) on stock market performance. Oberndorfer (2009) conducted the first econometric analysis on stock market effects of the EU Emission Trading scheme (ETS) and excluded herein corporations with main activities in renewables, based on the assumption of these
7
Further, Da Silva et al. (2016) show that asymmetry and EUA effects are power firm-specific. These findings let us assume that profitability in renewables also increases with rising prices in electricity generation, as European emission allowances increase the marginal costs of production in electricity generation from conventional sources. Additionally, Oberndorfer determined that EUA price increases positive affected stock returns of electricity corporations due to grandfathering, which implies a profit increase for the infra-marginal unit. Stock returns of alternative energy companies should react positively in response to emission price increases independent from grandfathering. These findings reinforce our assumption that electricity prices also do have an effect on renewables instead of only the contrary being the case.
Mulder and Scholtens (2013) conducted research on the direct relationship between the increase of renewable energy in the generation-mix and electricity prices in the
Netherlands. Especially relevant herein is the finding of a modest impact of an increase in renewables in the generation portfolio, on electricity prices (Mulder and Scholtens, 2013).
Nonetheless, regarding the interrelation between electricity prices and renewable energies, Mulder and Scholtens (2013) found Dutch electricity prices remaining closely related to the marginal costs of conventional gas-fired power plants. In accordance with their expectation, gas prices appear to be a key factor behind electricity prices.
Another article putting emphasis on the relationship between gas- and electricity prices tested the influence of shortages in gas supply on electricity prices in the EU-28 (Deane, Ciaráin & Gallachóir, 2017). The main finding herein is that a 28% increase in gas prices would cause a 12% rise in electricity prices, considering these findings, a direct
8
Lastly, the work of Henriques & Sadorsky “Oil prices and the stock prices of alternative energy companies”, from 2008 focused on testing the influence of oil prices on stock prices of alternative energy companies. This article can be treated as model for the subsequent research to be conducted, as the methodology applied allows examining granger causality between all variables of interest. The approach utilized by Henriques & Sadorsky (2008) enables quantifying the interrelation between time series of variables and will be used in testing the following hypotheses. Their dealings with matters such as the tracking of the performance of alternative energy companies’ stocks will also be considered in the subsequent data gathering process.
While Henriques & Sadorsky (2008) aimed to investigate the sensitivity of renewable energy companies to fluctuations in oil prices, they developed a vector autoregression model utilizing four variables. Beside the variables of which the hypothetical interrelation was supposed to be tested, also other variables have been included in order to increase the model’s precision. The approach of this paper will also include an additional variable beside the two variables of interest. As beside electricity prices and alternative energy companies’ performance also European CO2 Emission Allowances and European gas prices presumptive have explanatory power regarding the interrelationship of the variables included. As determined by Da Silva et al. (2016) an increase in EUA prices should be positively related to alternative energy companies’ financial performance and could therefore have more explanatory power regarding index performance than the to be tested electricity prices.
The inclusion of EUA prices and Egix gas prices as third and fourth variables should allow us to examine in a more precise manner which interrelations in the form of granger causality are present and should additionally allow to discuss possible backgrounds of relations found.
9
Therefore the hypotheses to be tested are:
H1: The price of electricity Granger causes alternative energy companies’ stock prices. H2: Alternative energy companies’ stock prices Granger cause the price of electricity. H3: The price of electricity Granger causes the prices of European CO2 Emission Allowances.
H4: Alternative energy companies’ stock prices Granger cause the prices of European CO2 Emission Allowances.
H5: The prices of European CO2 Emission Allowances Granger cause the price of electricity.
H6: The prices of European CO2 Emission Allowances Granger cause alternative energy companies’ stock prices.
The concept of granger causality and the established vector autoregression model will be explained in the following methodology section.
Methodology
Granger causality is present when one time series is useful in forecasting another (Granger, 1969). This condition utilized as hypothesis can be tested by performing a Granger causality test.
For our analysis this translates into applying a statistical tool that determines whether the lagged values of one time series of data included in our analysis (e.g. price of electricity), can be used to predict prospective values of a second time series of data (e.g. alternative energy companies’ stock prices).
10
A vector autoregressive model is a natural generalization of a univariate autoregressive model, as such a VAR can capture multiple variables with individual trends (multiple dependent variables) (Brooks, 2008).
In order to test the elaborated hypotheses, a vector autoregressive model is defined and consecutively estimated. This vector autoregression (VAR) allows investigating the statistical relationship between electricity prices, alternative energy companies’ stock prices, prices of European CO2 Emission Allowances and gas prices. Since the to be tested hypotheses assume that each included variable can be utilized in predicting future values of the other variables included, a four variable vector autoregression model fits the use of testing the formulated hypotheses best.
Using a VAR allows for testing all variables’ dependence on the lagged values of all other variables; in our analysis: testing for the interdependency of electricity prices, alternative energy companies’ stock prices, prices of European CO2 Emission Allowances and gas prices (Henriques & Sadorsky, 2007).
The general equation explaining each of the variables included in the four variable vector autoregression looks like follows:
=
+
+
+
+
+
(1)
Here, lagged values of all variables (including the values of the variable explained by the equation itself) and an error term explain the variable of interest and its evolution. The set of equations used in the four variable vector autoregression consists of four
11
A more generalized form can be written as:
= ∑
+
(2)Where, is a vector of variables to be explained by the other variables’ previous (lagged) values . is the matrix of regression coefficients to be estimated and the error term (Henriques & Sadorsky, 2007). Here, the indices of the regression coefficient matrices
= 1 …
number the matrices from the first to the-
lag.Beside the advantages the utilization of a vector autoregression model entails there are also drawbacks and limitations that need to be considered. One of the most important choices in estimating a VAR is the one of how many lags should be included in the model (Brooks, 2008). If the set lag length is too large relative to the used sample size, the degrees of freedom will be used up, which consequently leads to large standard errors on the estimated coefficients (Henriques & Sadorsky, 2007). A too low lag length on the other side might prevent the model from fully capturing the dynamic properties of the data used. Further, if the variables used in the estimated vector autoregression model are integrated or cointegrated, conventional asymptotic theory is not applicable to hypothesis testing (Toda & Yamamoto, 1995).
12
Thus, we can confirm whether granger causation is present. The Wald test statistic is:
( )
( )
(3)
Where is the maximum likelihood estimate of the parameters, which is compared with proposed values of the parameters .
However, before conducting the aforementioned model estimation and consecutive tests our variables need to be examined for potential nonstationary behavior. Testing for nonstationarity is done by applying pretests on our variables, which indicate whether unit roots of cointegration are present. Consecutively this information can be incorporated in estimating the VAR model; this approach would reduce estimation uncertainty.
Nevertheless, such pretests exhibit a lack in robustness (Gospodinov, Herrera & Pesavento, 2013), which is why a robust testing procedure, will be applied instead. Thereby, existing integration and cointegration in the data utilized, can be ignored without repercussions on significance.
The lag augmented VAR introduced by Toda and Yamamoto (1995) is robust to integration and cointegration properties of the data and avoids pretest bias. Lag augmentation refers to estimating a vector autoregression model of the (
+
)
th-order, where is the lag length determined by using a usual lag selection procedure and is the maximal order of integration of the variables included. This approach in estimating the VAR helps to overcome potential problems in hypothesis testing.13
Impulse response functions allow us to examine the reaction of our dynamic system to an impulse. Here, a one standard deviation shock of one variable, constitutes the applied impulse, the impact of this impulse on the other variables in the system will be examined (Henriques and Sadorsky, 2007). As before generating results robust to the ordering of our variables (integration/cointegration) is favorably, therefore we apply impulse responses, which are robust to these properties. Generalized impulse responses render imposing an orthogonality condition redundant (Pesaran & Shin, 1998).
Thus, the responsiveness of each variable implemented, to shocks in each variable can be investigated. Therefore, applying this approach gives us additional insight into the empiric relationship between the chosen variables in the model.
Data & Variables
In order to examine the empiric relationship between electricity prices, alternative energy companies’ stock prices, prices of European CO2 Emission Allowances and European gas prices, we need data on all of these variables from the same geographic region. The financial performance of alternative energy companies in terms of stock prices for the region of Europe will be measured by using the Ardour Global Alternative Energy Index – Europe (AGIEM), which is part of the index family Ardour Global Index (S-Network Global Indexes, 2017). This index is published and developed by S-(S-Network Global Indexes, Inc. and is quarterly reconstituted if needed. The Ardour Global
14
Constituents are pure-play (no secondary field of activity) companies selected by a positive screening process considering capitalization, float, exchange listing, share price and turnover. In detail, these screening criteria look like follows:
Each constituent has to be resided in the technology industry, has to be listed on an approved stock exchange, needs to have a minimum of $50 million free-float capital and a daily trade volume of at least $1 million. The current constituents of AGIEM are 18 alternative energy companies residing in 10 different European countries. Beside the difference in location, these companies are also operating in varying industry sectors: Energy, Industrials, Information Technology, Materials and Utilities.
These fields of operation are specified further by S-Network Global Indexes, Inc., which states more precise characteristics that constituents feature in their industry breakdown. These are, Alternative Energy Resources, Distributed Generation, Environmental
Technologies, Energy Efficiency and Enabling Technologies (S-Network Global Indexes, 2017). As all included companies are either directly engaged in the generation of power from renewable energy sources and/or suppliers/supporters for related technologies the AGIEM index should move in accordance with the performance of the alternative energy industry sector.
Hence, the chosen index serves as a proxy for the alternative energy industry in Europe in the upcoming analysis. Data regarding the Ardour Global Alternative Energy Index – Europe will be utilized as time series consisting of daily prices, comprising a period ranging from 14 May 2012 (as this is the earliest date for which data is available) to 3 April 2017.
15
Regarding electricity prices applicable to the region of Europe, the findings of Oberndorfer come in useful to our analysis. As Oberndorfer (2009) noted in his
publication, German/Austrian electricity prices can serve as proxy for overall European electricity price developments. This approach is eligible due to a price convergence within this geographic area.
In order to obtain the most precise data, prices at the European Energy Exchange (EEX) will serve as source for the electricity price component in the upcoming analysis. The EEX is one of the most liquid European power exchanges and the leading energy exchange in Central Europe. Its Physical Electricity Index is published daily and serves as reference price for the sought German/Austrian market area. Data on the Physical Electricity Index (Phelix) is available in various forms, while there are spot prices listed (day-ahead, auction), prices of various futures contracts with different maturities (yearly, quarterly, monthly, weekly, for weekends and daily) are available as well. Beside these differentiations, each price is available for different kinds of system loads and
timeframes (base, peak, offpeak and for each individual hour of a day).
When deciding on which contract’s price to be used in the upcoming analysis (spot prices, futures prices), it has to be considered that electricity prices exhibit high volatility and strong mean-reversion (due to the inability of storing electricity efficiently). These properties cause futures prices to be less volatile than spot prices. In addition, the merit-order effect of renewables is supposed to be greater on the spot market than on the futures market.
Consequently, a time series of electricity futures prices will be analyzed concerning its interrelation with the AGIEM index performance, European CO2 Emission Allowance prices and European gas prices. This procedure should render a result independent from the short-term volatility of spot prices and allows to examine whether a hypothetical long-term interrelation between those variables exists.
16
The same criteria is used to decide on which system load prices are most suitable. Based on trade volumes and previously mentioned criteria, Phelix Futures Year Base prices will be used. Prices of this time series are stated in the unit of EUR/MWh.
To match the time series of electricity prices to the series of prices for the Ardour Global Alternative Energy Index – Europe, the available data is narrowed down to daily prices for a timeframe encompassing as well a period ranging from 14 May 2012 to 3 April 2017.
Figure 1. AGIEM index and Phelix electricity prices (the left y-axis displays the AGIEM index values, while the right y-axis displays Phelix Futures prices in EUR)
The time series plot of the AGIEM (Ardour Global Alternative Energy Index – Europe) and Phelix electricity prices is shown in Figure 1.
On first observation, it can be assumed that the AGIEM index performance and the Phelix Baseload Year futures prices are negatively correlated. This finding gives first support for the initially made assumption that renewable approaches of energy generation might indeed have a decreasing-effect on European electricity prices.
0 10 20 30 40 50 60 0 200 400 600 800 1000 1200 1400 1600 05 /2 01 2 07 /2 01 2 08 /2 01 2 10 /2 01 2 11 /2 01 2 01 /2 01 3 03 /2 01 3 04 /2 01 3 06 /2 01 3 07 /2 01 3 09 /2 01 3 11 /2 01 3 12 /2 01 3 02 /2 01 4 04 /2 01 4 05 /2 01 4 07 /2 01 4 08 /2 01 4 10 /2 01 4 12 /2 01 4 01 /2 01 5 03 /2 01 5 05 /2 01 5 06 /2 01 5 08 /2 01 5 09 /2 01 5 11 /2 01 5 01 /2 01 6 02 /2 01 6 04 /2 01 6 05 /2 01 6 07 /2 01 6 09 /2 01 6 10 /2 01 6 12 /2 01 6 01 /2 01 7 03 /2 01 7
AGIEM & Phelix
17
In fact, the correlation coefficient between both time series of data is -0.904, which can indicate strong dependency. Nonetheless, other factors are also likely to have impacted the process of how these variables developed over time. In which direction, if not both, the relationship between both variables exerts influence, will be determined in the upcoming analysis.
Our third variable to be included in the subsequent analysis are prices of European CO2 Emission Allowances. These tradable CO2 emission rights are also known as European Union Allowances (EUAs) and entitle the holder of it to emit one metric ton of carbon dioxide (CO2). The most important exchange for trading Emission Allowances is again the European Energy Exchange (EEX). Here again, daily closing prices will be utilized. The concept of EUA prices being positively related to the performance of alternative energy companies seems natural, as rising EUA prices should only increase production costs of competitors of green energy electricity producers. Further, European CO2 Emission Allowances have been proven to influence the stock performance of electricity firms, thus some kind of impact is to be expected.
Figure 2. AGIEM index and EUA prices (the left axis displays the AGIEM index values, while the right y-axis displays EUA prices in EUR)
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 0 200 400 600 800 1000 1200 1400 1600 05 /2 01 2 07 /2 01 2 09 /2 01 2 11 /2 01 2 01 /2 01 3 03 /2 01 3 05 /2 01 3 07 /2 01 3 09 /2 01 3 11 /2 01 3 01 /2 01 4 03 /2 01 4 05 /2 01 4 07 /2 01 4 09 /2 01 4 11 /2 01 4 01 /2 01 5 03 /2 01 5 05 /2 01 5 07 /2 01 5 09 /2 01 5 11 /2 01 5 01 /2 01 6 03 /2 01 6 05 /2 01 6 07 /2 01 6 09 /2 01 6 11 /2 01 6 01 /2 01 7 03 /2 01 7
AGIEM & EUA
18
Figure 2 shows how EUA prices relate to the performance of the alternative energy index chosen. The corresponding correlation coefficient between both time series of data is 0.019, which constitutes a weak empirical relationship. Nonetheless, also here, the consequent analysis will clarify the relationship between both variables in terms of granger causality. In addition, based on the findings elaborated in other works covered earlier, the implementation of an additional variable reflecting European gas prices in our vector autoregression model appears to be promising, regarding the validity of our
model. Here, the European Gas Index – Egix will serve as representative for overall European gas prices. Just like the Physical Electricity Index (Phelix), the European Energy Exchange (EEX) introduced the Egix. It is calculated based on trades executed on the EEX for the market areas NCG and GASPOOL (Germany) (EEX, 2017). Utilized in its calculation are front month contracts (closest expiration date to current date), front month contracts are generally the type futures contracts with the highest occurrence of trade. Due to unavailability of daily prices in Egix Futures, monthly prices will be
adjusted to fit the daily time series of the other variables deployed. Egix gas prices are stated in the unit of EUR/MWh. Expected is a strong relationship between Egix prices and prices of the Phelix, as gas prices are a determining factor of electricity prices.
Figure 3. Egix gas prices and Phelix electricity prices (the left y-axis displays the Egix Futures prices in EUR, while the right y-axis displays Phelix Futures prices in EUR)
0 10 20 30 40 50 60 0 5 10 15 20 25 30 05 /2 01 2 06 /2 01 2 08 /2 01 2 10 /2 01 2 11 /2 01 2 01 /2 01 3 02 /2 01 3 04 /2 01 3 05 /2 01 3 07 /2 01 3 08 /2 01 3 10 /2 01 3 12 /2 01 3 01 /2 01 4 03 /2 01 4 04 /2 01 4 06 /2 01 4 08 /2 01 4 09 /2 01 4 11 /2 01 4 12 /2 01 4 02 /2 01 5 03 /2 01 5 05 /2 01 5 07 /2 01 5 08 /2 01 5 10 /2 01 5 11 /2 01 5 01 /2 01 6 03 /2 01 6 04 /2 01 6 06 /2 01 6 07 /2 01 6 09 /2 01 6 10 /2 01 6 12 /2 01 6 01 /2 01 7 03 /2 01 7
Egix & Phelix
19
The time series of Egix gas prices and Phelix electricity prices is shown in Figure 3. A first visual analysis already suggests a relatively strong positive empirical relationship between both of these variables and is confirmed by a corresponding correlation coefficient of 0.773. Despite this significant figure, a vector autoregression model including the variables stated remains to be estimated in order to ascertain granger causality.
Missing data for dates without the occurrence of trade and thus without the
determination of a price, will be supplemented with data from the most recent trade. All time series utilized are openly available online.
Vector autoregression & discussion
In order to utilize the previously described methodology including the usage of a vector autoregression (VAR) after Toda and Yamamoto (1995), different statistical tests need to be conducted prior. Applying this lag-augmented vector regression allows us to determine granger causality between the four variables of interest.
Before estimating our four variable vector autoregressive model, the time series of the Ardour Global Alternative Energy Index – Europe (AGIEM), the physical electricity index Phelix, the European CO2 Emission Allowances and of Egix gas prices are converted into the series of their natural logarithms, hereafter: LAGIEM, LPHELIX, LEUA and LEGIX. This measure constitutes one of the potential fixes to prevent heteroscedasticity within the set of variables.
In order to test the elaborated hypotheses, which would allow for drawing conclusions regarding the effects we are interested to shine a light on, we apply the lag
20
As previously stated, represents the lag length determined by usual lag length
procedures, for example the Schwarz information criterion (SIC), the Akaike information criterion (AIC), the Final Prediction Error criterion (FPE), etc. which indicate the relative quality of vector autoregression models of different lag lengths. Further, is the maximal order of integration of the time series utilized in the estimated vector
autoregression model.
To investigate the integration properties of the time series involved, unit root tests, such as the Augmented Dickey-Fuller test (ADF), Phillips-Perron test (PP) and the
Kwiatkowski-Phillips-Schmidt-Shin test (KPSS) will be executed. In contrast to the Augmented Dickey-Fuller and Phillips-Perron tests, the null hypothesis of the
Kwiatkowski-Phillips-Schmidt-Shin unit root tests state that the series of data examined is stationary. Investigating the integration properties of our data, allows us identifying the sought order of integration of each series.
Table 1 shows the results of the unit root tests applied.
The lag length in the ADF regression uses the Schwarz information criterion as determinant, while the KPSS and PP regressions utilize the Newey-West bandwidth. ***, **, *, designate 1%, 5%, and 10% levels of significance.
As it can be retrieved from the results displayed in Table 1 the order of integration for each series of data is one at most, as the first differences of all time series of variables included are stationary.
Table 1
Unit root tests Levels First differences
ADF PP KPSS ADF PP KPSS
LAGIEM -1.280 -1.286 3.961*** -32.702*** -32.652*** 0.126
LPHELIX -1.828 -1.831 3.787*** -33.653*** -33.653*** 0.190
LEUA -2.769* -3.382** 0.411* -19.487*** -39.053*** 0.061
21
Despite this, the data series of CO2 Emission prices (LEUA) even shows some evidence at lower levels of significance in the Augmented Dickey-Fuller and Phillips-Perron test for an order of integration of zero. Consequently, a of 1 is chosen and used for the upcoming estimation of the lag augmented vector autoregression model.Now that we have determined the maximal order of integration we use likelihood ratio statistics to find the appropriate VAR lag length . Here, just as previously described,
conventional model selection criteria are used to determine the sought variable.
Table 2
VAR Lag Order Selection Criteria
Lags LR FPE AIC SIC HQ
0 NA 1.32E-06 -2.183 -2.166 -2.176
1 22275.03 1.65E-14 -20.385 -20.302* -20.354*
7 132.913* 1.60E-14* -20.417* -19.934 -20.235
LR is the sequential modified LR test statistic and HQ is the Hannan-Quinn information criterion, all other lag length selection criteria as previously described. * indicates the lag order selected by the respective
criterion.
As Table 2 illustrates, a VAR lag length of 7 is selected based on the results of the utilized lag length selection criteria on a range of 12 included lags.
Thus,
( +
) = (7 + 1) = 8
(3)Decisive here for making the decision are the model selection criteria LR, FPE and AIC. Subsequently, a VAR of the 8th order is estimated. This approach is in line with the lag
22
The estimated lag augmented VAR can now be exposed to different model fit tests. These different parameters indicate how well the estimated model fits the used data. Further, these can indicate the tightness of the model fit. Among the resulting
coefficients of determination are , adjusted , an F-Statistic and a S.E. equation. Whereas, and even more so adjusted tell us to what extent the observed values in the data given can be replicated by our lag augmented vector autoregression model, the reported F-statistic renders the importance of our explanatory variables. Further, the reported S. E. equation provides a measure of the difference between predicted values and actual ones. Here, small values indicate a tight fitting model.
Table 3
Model fit tests
R-squared Adj. R-squared F-statistic S. E. equation
LAGIEM 0.998 0.998 23474.790 0.014
LPHELIX 0.997 0.997 12707.820 0.011
LEUA 0.959 0.958 873.749 0.050
LEGIX 0.996 0.996 8726.524 0.016
As Table 3 illustrates, adjusted R-squared ranges from 0.958 for the LEUA equation, over 0.996 in LEGIX and 0.997 in LPHELIX to 0.998 for the LAGIEM equation. These values indicate that the model fits well. Additionally, the displayed standard errors of the equation show us that the estimated model fits tight. The reported F-statistics are
statistically significant at a 1% significance level, meaning that all regression equations’ explanatory variables are jointly substantial.
23 Table 4
LM tests for serial correlation
Lags LM-Stat Probability
2 6.660 0.979
5 6.741 0.978
11 7.535 0.962
Respective probabilities are calculated using a chi-square distribution with 16 degrees of freedom
Table 4 shows us the LM-statistics for different amount of lags with their corresponding probabilities. As it can be obtained, the residuals of our data series exhibit no serial correlation at the 5% significance level. The vector autoregression model fits well and a LA-VAR Wald test can be deployed. Thus, the different time series of data can be examined for granger causality.
The lag augmented Wald test constitutes the applied granger causality test. The null hypothesis of the LA-VAR Wald test states that the tested independent variable does not granger cause the dependent variable tested for. Variations of this null hypothesis are tested for each possible dependent/independent variable combination possible. Thus, we can obtain decisive results regarding the directions in which granger causality occurs.
Beside the hypotheses stated earlier, the applied LA-VAR Wald test allows for testing joint hypotheses, thus the following joint hypotheses are tested as well:
: The lagged values of the price of electricity, the lagged values of the prices of European CO2 Emission Allowances and the lagged values of gas prices do not hold predictive value over alternative energy companies’ stock prices.
24
: The lagged values of alternative energy companies’ stock prices, the lagged values of the price of electricity and the lagged values of gas prices do not hold predictive value over the prices of European CO2 Emission Allowances.
: The lagged values of alternative energy companies’ stock prices, the lagged values of the price of electricity and the lagged values of the prices of European CO2 Emission Allowances do not hold predictive value over the price of gas.
The following table shows the results of the applied LA-VAR Wald test.
Table 5
LA-VAR Wald test Dependent variable
LAGIEM LPHELIX LEUA LEGIX
LAGIEM - 0.848 0.8399 0.0437**
LPHELIX 0.099* - 0.124 0.600
LEUA 0.510 0.273 - 0.891
LEGIX 0.994 0.170 0.929 -
***, **, *, designate statistical significance at 1%, 5%, and 10% levels of significance.
25 Table 6
LA-VAR Wald test (Joint) Dependent variable
LPHELIX, LEUA, LEGIX LAGIEM, LEUA, LEGIX LAGIEM, LPHELIX, LEGIX LAGIEM, LPHELIX, LEUA
LAGIEM 0.375 - - -
LPHELIX - 0.107 - -
LEUA - - 0.606 -
LEGIX - - - 0.917
***, **, *, designate statistical significance at 1%, 5%, and 10% levels of significance.
The results of the LA-VAR Wald test employed as granger causality test in our lag augmented VAR model show that past movements in the Ardour Global Alternative Energy Index – Europe, explain electricity prices. Further, the applied test determines that Egix gas prices granger cause values of the AGIEM index. Lastly, the tests
26
Consequently, in order to shine additional light on the dynamic properties of the data used, impulse response functions are utilized. These enable us to examine reactions of our estimated system to shocks of one standard deviation in each variable. Individual responses of each variable implemented, to shocks in each variable are illustrated in Figure 4.
Figure 4. Impulse response functions (y-axis: value of the natural logarithm of the respective variable, x-axis: number of days after the simulated shock; the dashed lines display initial values deviated by two
27
The by two standard errors deviated values shown in the graphs of figure 4 (dashed lines) will be utilized as confidence intervals, allowing us to evaluate whether the displayed responses to simulated shocks are significant.
The alternative energy companies’ stock prices as represented by the AGIEM index respond significantly positive to a shock to itself for a time span of up to 25 days. In accordance with the findings of the previously conducted Wald test for granger causality a positive significant response of the AGIEM index values to a shock in gas prices depicted by the Egix index can be found up to eight days after the simulated shock. Contrary to the results of the conducted granger causality test, the generated impulse responses show the AGIEM index significantly responding to a shock in European Emission Allowances prices, as indicated by the EUA variable. This response is slightly positive for up to ten days. Further, the Ardour Global Alternative Energy Index – Europe reacts slightly positive to a shock of one standard deviation magnitude in European electricity prices as constituted by the Phelix for a time span of seven days. Altogether, the AGIEM index appears to be susceptible to shocks in the other variables of our system. Nonetheless, the response to a shock in gas prices remains the greatest and is thus approximately consistent with the findings of the previously conducted LA-VAR Wald test. The Egix index responds positively to a shock in itself over a time span of 23 days, further a slightly positive significant response for four days to a shock in European electricity prices can be found. Regarding the Egix, no significance response to shocks in the AGIEM index and EUA prices can be concluded. European Emission Allowance prices do not respond significantly to shocks in the AGIEM index, nor to shocks in
28
This finding is in line with the results of the executed granger causality test, which did not exhibit significance regarding both variables’ empirical relationship as well.
Altogether, these findings question the importance of gas in the production costs of European electricity producers. An unexpected result is displayed by the impulse response test for the relationship between electricity prices and European Emission Allowances. Despite the results of the LA-VAR Wald test, indicating no significance, European electricity prices react positive to a shock in EUA prices, significant for 20 days. This finding supports the initially made assumption that CO2 Emission Allowances have an impact on electricity prices due to their relevance in electricity production from conventional measures.
The findings of Cludius, Hermann & Matthes (2013) regarding the impact of the merit order effect in a scenario of increased power generation from renewables are in line with the results elaborated by us. Past values of the AGIEM index influence Phelix electricity prices as determined by the conducted LA-VAR Wald test, additionally the response of Phelix electricity prices to a shock in the AGIEM index is significant. The previously calculated correlation coefficient of -0.904 specifies the direction of the impact a strengthening of alternative energy companies has on European electricity prices. In Cludius, Hermann & Matthes (2013), additional generation from renewable energies does have a price reducing effect on electricity prices. The results elaborated in this work indicate the same price reducing effect induced by a strengthening of alternative energy companies’ financial performance, as measured by the AGIEM index. A positive empirical relationship between additional generation from renewable energies and a financial strengthening of alternative energy companies seems natural.
29
On a side note, the prediction made by Mulder & Scholtens (2013) that future electricity prices may become less related to the marginal costs of conventional power plants cannot be validated by the analysis conducted. Even though no empirical evidence for a direct relationship between gas prices and electricity prices has been found, neither in the conducted granger causality test, nor in using impulse response functions, a direct effect of gas on the AGIEM index and thus extended on Phelix electricity prices cannot be denied. Evidence for the electricity price being related to the stock performance of alternative energy companies is factual.
In contrast to the first determined granger causality, the used AGIEM index and CO2 Emission Allowance Prices do not have a granger causal impact on each other. This result is in line with the assumption made by Oberndorfer (2009) who started his analysis from the premise that EUA price changes do not have an influence on
alternative energy firms, since corporations with main activities in renewables do exhibit a lower exposure to ETS regulation. This assumption can be confirmed for the scenario and variables tested. However, a shock to EUA prices did have an impact on Phelix electricity prices in the performed impulse response test. This outcome signals that despite an increase in electricity production from renewable measures, European
electricity prices are still affected by CO2 Emission Allowance prices. The assumption of a potential direct effect on the profitability of activities in renewable energies of
increasing EUA prices as stated by Da Silva, Moreno & Figueiredo (2016) is refuted by our results for the scenario and variables tested. Here, EUA prices did not have a direct effect on the profitability of alternative energy companies as represented by our AGIEM index. The reasoning behind this potential effect is that conventional power generation would be more expensive to operate, leaving more room for renewable energy
generation (associated increase in stock market returns). This potential cause does not take effect in our sample, which might be explained by an overvaluation of the effect, EUA prices have on conventional electricity producers’ profitability. Nonetheless, a shock in EUA prices did cause a response in European electricity prices.
30
However, each of these variables responded significantly to a one standard deviation shock in the Physical Electricity Index. Thus, dependence of European Emission Allowance prices, AGIEM index performance and Egix gas prices cannot be refuted. It seems there is an empirical relationship between prices of the Phelix index and the other variables implemented in our system of a non-granger causal type. Unlike the other variables in the system, AGIEM index performance responds significantly to shocks in any other variable, indicating sensitivity. Nevertheless only Egix gas prices do granger cause values of the AGIEM index as determined by the applied LA-VAR Wald test. The findings of Deane, Ciaráin & Gallachóir (2017) of an empirical relationship between gas prices and electricity prices could not be confirmed. A potential reasoning for not finding this empirical relationship might be deficient price convergence for gas- and/or electricity prices within the region tested for.
Conclusion
The impact of strengthening renewable energy generation has already been investigated in multiple scenarios with different focuses. Thus, a range of various
insights has been elaborated, of which some have been proven to be valid, while others to be void in the setting chosen for this work. In order to investigate the interrelation between alternative energy companies’ stock prices in Europe, European CO2 Emission Allowance prices, European electricity prices and European gas prices, a vector
autoregression model including these variables is developed.
The results of the consequently employed tests for granger causality indicate that the financial performance of the companies included in the AGIEM index does hold power in predicting price movements of European electricity, these prices are represented by the Physical Electricity Index (Phelix). Further, the same tests determined significant
31
Subsequently utilized impulse response functions confirmed both of the previously mentioned findings but also gave additional insight regarding the relationship between European CO2 Emission Allowance prices and European electricity prices, as Phelix electricity prices responded significantly to a simulated shock in EUA prices. By having elaborated these results, some previously developed findings could be confirmed and even transferred to a European scale.
The assumption of renewable energies being unexposed to ETS regulation made by Oberndorfer (2009) can be verified, for granger causality but is contradicted by the findings elaborated by the conducted impulse response test. Hence, it remains unclear whether and how alternative energy companies are affected by ETS regulation and emphasize should be put on additional research. The findings of Cludius, Hermann & Matthes (2013) and of Sensfuß, Ragwitz & Genoese (2007), who found that an increase of renewable energies in the generation mix reduces the price of electricity as the merit order effect takes place could be confirmed by the results of the applied tests. As for Oberndorfer (2009), the potential direct effect as hypothesized by Da Silva, Moreno & Figueiredo (2016) cannot be confirmed nor denied for the scenario tested in. According to the conducted granger causality test, EUA prices cannot explain the movements in alternative energy companies’ stock prices, but cause a significant response in these when exposed to a shock. The conclusion elaborated by Deane, Ciaráni & Gallachóir (2017) could not be confirmed with the results found for the relationship between gas- and electricity prices. Yet statistical dependence of alternative energy companies’ stock prices on European gas prices has been determined by the applied granger causality test and the utilized impulse response functions. This finding might be explained by factors of the competition in European electricity generation, but ultimately remains to be investigated. These perceptions might be of use to policy makers, investors and other stakeholders.
The limitations of this work are easily distinguishable as variables utilized needed to be proxied. As such, a data collection focused on increasing the accuracy of the
32
Further, potential repercussions of an increased share of renewable energies in Europe’s generation mix with a focus on exposure to weather conditions, should be investigated. Additionally, it would turn out worthwhile to employ research on the influence of electricity price convergence on the determined effects. Conducting
33
References
Brooks, C., 2008. Introductory econometrics for finance. Cambridge University Press, United Kingdom.
Cludius, J., Hermann, H., Matthes, F., 2013. The merit order effect of wind and photovoltaic electricity generation in germany 2008-2012. Centre for Energy and Environmental Markets Working Paper 3-2013.
Da Silva, P., Moreno, B., Figueiredo, N., 2016. Firm-specific impacts of co2 prices on the stock market value of the spanish power industry. Energy Policy 94, 492-501.
Deane, J., Ciaráni, M., Gallachóir, B., 2017. An integrated gas & electricity model of the eu energy system to examine supply interruptions. Applied Energy 193, 479-490.
European Commission, 2017. The eu emissions trading system (eu ets). Retrieved 22.05.2017 from https://ec.europa.eu/clima/policies/ets_en
European Energy Exchange AG (EEX), 2017. Eex reference price egix 2017-05-02. Retrieved 18.05.2017 from
https://www.eex.com/de/marktdaten/erdgas/terminmarkt/egix/egix-download
Eurostat, 2016. Electricity production, consumption and market overview. Retrieved 22.05.2017 from
34
Gospodinov, N., Herrera, A., Pesavento, E., 2013. Unit roots, cointegration, and pretesting in var models. Advances in Econometrics 32, 81-115.
Granger, C., 1969. Investigating causal relations by econometric models and cross-spectral methods. Econometrica 37, 424-438.
Henriques, I., Sadorsky, P., 2008. Oil prices and the stock prices of alternative energy companies. Energy Economics 30, 998-1010.
Mulder, M., Scholtens, B., 2013. The impact of renewable energy on electricity prices in the netherlands. Renewable Energy 57, 94-100.
Oberndorfer, U., 2009. Eu emission allowances and the stock market: evidence from the electricity industry. Ecological Economics 68, 1116-1126.
Pesaran, M., Shin, Y., 1998. Generalized impulse response analysis in linear multivariate models. Economics Letters 58, 17-29.
Sensfuß, F., Ragwitz, M., Genoese, M., 2007. The merit-order effect: a detailed analysis of the price effect of renewable electricity generation on spot market prices in germany. Working paper sustainability and innovation No. S7/2007.
35
S-Network Global Indexes, 2017. Industry definition. Retrieved 10.04.2017 from http://ardour.snetglobalindexes.com/sites/snardour/industry-definition
36
Appendix
37
39
40
