**3. Wind capacity, energy sources and European public policies**

Wind energy growth in the last decade in Europe was mainly driven by several factors such as: energy demand growth; the commitments made to greenhouse gas reduction under the Kyoto protocol directives; improvements in renewable energy technology; and the reduction of the marginal cost of wind power generation over the past 15 years, approaching the cost of conventional energy sources (Pechak et al., 2011). For these reasons, wind power has reg‐ istered a strong impulse since the late 1990s and early 2000s. As a consequence, due to the lack of data before 1998 for almost all European countries, this study uses panel data for the time span 1998-2009, for the following countries: Austria, Belgium, the Czech Republic, Den‐ mark, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, the Netherlands, Norway, Poland, Portugal, Spain, Sweden and the United Kingdom. These countries are part of a group that is driven by long-term energy goals under European directives (EU di‐ rective, 2009). Not all countries have the same number of observations due to sporadic miss‐ ing values, which leads to an unbalanced panel. The remaining countries of the EU27 did not provide available data for wind power installed capacity in the considered time span.

Panel data techniques have several advantages, such as: (i) they allow a more accurate statis‐ tical inference; (ii) they provide more informative data and variability; (iii) they increase the number of observations and degrees of freedom; and (iv) they allow for controlling individ‐ ual heterogeneity and unobserved characteristics of errors which are not detectable in timeseries or cross-sectional models (Baltagi, 2005 and Hsiao, 2006).

### **3.1. Wind capacity**

**2.1. Intermittency and wind power overcapacity**

54 New Developments in Renewable Energy

geted by Yang et al. (2012) and Zhang and Li (2012).

led to fast, but inefficient, wind energy deployment.

tor of 16.3% between 2007 and 2010 (Yang et al., 2012).

**2.2. Backup and energy storage**

Although the issue of renewable intermittency is far from new in the literature, the rele‐ vance of this topic together with the phenomenon of overcapacity requires much more re‐ search. The main reasons and impacts of non-constant generation of wind energy are analyzed by authors such as Albadi and El-Saadany (2010), and Green and Vasilakos (2010). Gonzalez et al. (2004) focus on Ireland, Gül and Stenzel (2005) on Scandinavia, the United Kingdom and the United States, Caralis et al. (2008) on Greece and the Chinese case is tar‐

Intermittency in renewables can be analyzed by using the capacity factor. This is the ratio, for a certain period of time, of the energy generated to the energy that would have been gen‐ erated in operation by operating total continuous power during the same period (Denholm et al., 2005). Boccard (2009) summarizes that capacity factor depends on: (i) wind variability; (ii) the shadowing phenomenon; and (iii) the intensive focus on subsidy policies. The shad‐ owing phenomenon comes from installing too many wind turbines in a limited area to save costs on land use. Moreover, the short distance between wind farms compromises the indi‐ vidual performance of each farm. The vast use of public financial support policies may have

Acker et al. (2007) noted that a seasonal influence in the capacity factor can be observed. Caralis et al. (2008) analyzed the capacity factors in Greece and suggest that spatial disper‐ sion of wind farms benefits the wind power capacity factor. They concluded that the accu‐ mulation of too many wind farms is not always the optimal solution because it may impair the efficiency of each individual wind farm. More recently, Yang et al. (2012) and Zhang and Li (2012) assessed wind power growth in China, which was driven by three main factors: (i) the perception that China benefits from large wind resources; (ii) the adoption of incentives and subsidies that support the investment in wind power; and (iii) the reduction in wind capital costs. The authors note that more attention to the efficiency of wind turbine alloca‐ tion in China is needed. In fact, one-third of wind turbines were idle, causing a capacity fac‐

It is important to seek new ways to deal with wind speed variability, both in the short and long term. Examples could be additional energy sources to backup power in windless peri‐ ods or energy storage devices (Purvins et al., 2011). To ensure a secure energy supply, it is necessary to mix wind power with other energy sources, including fossil fuels. Pearce (2009) suggests a solar photovoltaic system mixed with combined heat and power to overcome in‐ termittency in California without resorting to energy storage. Moreno and Martínez-Val (2011) argue that thermal power plants are no longer so important in base load energy gen‐ eration, turning them into backup sources to substitute renewables. These authors support that by 2020, backup with combined cycle gas turbine plants needs to grow to 8 or 9 Giga‐ watts. The literature (e.g. Archer and Jacobson, 2007) also mentions another method to smooth wind variability. These authors found that by interconnecting multiple wind parks

For a better approach to the issue of intermittency and overcapacity, it proved necessary to make the concept of overcapacity operational. To do so, a variable which emulates wind overcapacity (*WOCAPc,t*) was created.

*WOCAPc,t* is the dependent variable and represents the ratio of idle capacity in a year to the hypothetical maximum energy that could be produced in a year, in a continuous full-power operation. This ratio was computed from raw data, and can be done in two different ways: (i) through idle capacity; and (ii) through capacity factor. Accordingly, for option (i) the re‐ sult is:

$$\text{WOCAP}\_{c,t} = \frac{\text{IDCAP}\_{c,t}}{\text{TOTALCAP}\_{c,t}}.\tag{1}$$

21366.3836

14781.2648

(147*\**8760) - (277*\**1000)

(29*\**8760) - (49*\**1000)

115.3790

23.4064

*CAPc,t* was:

as:

in these regions.

**3.2. Variables**

measures and public policies as follows.

For Finland and Latvia, which are the countries with the lowest wind installed capacity (147 MW and 29 MW respectively) with electricity output of 277 GWh and 49 GWh, in 2009 *ID‐*

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

Hence, wind overcapacity ratios (*WOCAPc,t*) for these two countries respectively are given

Our computations indicate that 82.89 %, 77.85 %, 78.49 % and 80.71 % of the wind installed capacity was idle during the year, i.e., a capacity factor of 17.11 %, 22.15 %, 21.51 % and 19.29 % respectively for Germany, Spain, Finland and Latvia. These values are relatively high. Indeed, it is surprising that this issue has not been addressed earlier with more em‐ phasis in the literature. Average *WOCAPc,t* values for all countries of our panel for the time span 1998-2009 are presented in Figure 1. Wind overcapacity average values are in line with other authors who addressed capacity factors, like Boccard (2009) and Yang et al. (2012). For example, in Denmark and Portugal, the average *WOCAPc,t* is 0.7790 and 0.7840 respectively, and according to (4) the capacity factor is 0.2210 and 0.2160. It denotes that Nordic countries (e.g. Norway, Sweden, Finland, Denmark, the United Kingdom and Ireland) as well as southern Europe (e.g. Portugal, Spain and Greece) have less idle capacity and therefore more capacity factors than continental countries. This may be because of higher wind speeds

Several causes for idle capacity are suggested by the normative literature. Following this closely, the impact of variables with different natures is controlled for, such as: conventional energy sources; other renewable sources; socio-economic drivers; and energy efficiency

<sup>25777</sup> ≈ 0.8289, (7)

http://dx.doi.org/10.5772/52159

57

<sup>18988</sup> ≈0.7785. (8)

<sup>8760</sup> ≈ 115.3790 *MW* , (9)

<sup>8760</sup> ≈ 23.4064 *MW* . (10)

<sup>147</sup> ≈ 0.7849, (11)

<sup>29</sup> ≈0. 8071. (12)

In equation (1) *TOTALCAPc,t* is the total of wind installed capacity. *IDCAPc,t* denotes the idle capacity of wind power in a year. In other words, *IDCAPc,t* represents the difference between maximum possible wind electricity generation during the year (8760 hours) and the amount of electricity actually generated. *TOTALCAPc,t* and *IDCAPc,t* are expressed in Megawatts (MW) and this last one is computed as follows:

$$\text{IDCAP}\_{c,t} = \frac{\{\text{WINDCAP}\_{c,t} \, ^\ast 8760\} \cdot \{\text{TOTELECGN}\_{c,t} ^\ast 1000\}}{8760} \,\tag{2}$$

where *TOTELECGENc,t* is the total electricity generated in a year, in Gigawatts *per* hour (GWh). *TOTELECGENc,t* is multiplied by 1000 to convert to same units.

Regarding option (ii) *WOCAPc,t* can be computed as the difference between 1 and the capaci‐ ty factor (*CFc,t*) as follows:

$$\text{WOCAP}\_{c,t} = \mathbf{1} \text{ - CF}\_{c,t} \,. \tag{3}$$

The capacity factor is computed as follows:

$$\text{CF}\_{c,t} = \frac{\text{TOTELECGEN}\_{c,t} \* 1000}{\text{TOTECAP}\_{c,t} \* 8760} \,\text{.}\tag{4}$$

In expressions (3) and (4) *CFc,t* is the ratio of actual wind power to maximum capacity in a year. For example, for Germany and Spain, which are the leader countries in terms of wind installed capacity, in 2009 the total installed capacity was respectively 25777 MW and 18988 MW. Electricity output was 38637 GWh and 36851 GWh. From here, following equation (2) for Germany and Spain, *IDCAPc,t* in 2009 was:

$$\frac{(25777^\*8760) \cdot (38637^\*1000)}{8760} \approx 21366.3836 \text{ MW} \,\text{A} \,\text{W} \,\text{-} \,\tag{5}$$

$$\frac{\text{(18988\*8760)} \cdot \text{(36851\*1000)}}{8760} \approx 14781.2648 \text{ MW.} \tag{6}$$

In accordance with equation (1), wind overcapacity ratios (*WOCAPc,t*) for Germany and Spain are given as:

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward http://dx.doi.org/10.5772/52159 57

$$\frac{21366.3836}{25777} \approx 0.8289\,,\tag{7}$$

$$\frac{14781.2648}{18988} \approx 0.7785.\tag{8}$$

For Finland and Latvia, which are the countries with the lowest wind installed capacity (147 MW and 29 MW respectively) with electricity output of 277 GWh and 49 GWh, in 2009 *ID‐ CAPc,t* was:

$$\frac{(147^\*8760) \cdot (277^\*1000)}{8760} \approx 115.3790 \text{ MW}\_{\prime} \tag{9}$$

$$\frac{(29^{\ast}8760)\cdot(49^{\ast}1000)}{8760} \approx 23.4064\text{ MW.}\tag{10}$$

Hence, wind overcapacity ratios (*WOCAPc,t*) for these two countries respectively are given as:

$$\frac{115.3790}{147} \approx 0.7849\,,\tag{11}$$

$$\frac{23.4064}{29} \approx 0.8071.\tag{12}$$

Our computations indicate that 82.89 %, 77.85 %, 78.49 % and 80.71 % of the wind installed capacity was idle during the year, i.e., a capacity factor of 17.11 %, 22.15 %, 21.51 % and 19.29 % respectively for Germany, Spain, Finland and Latvia. These values are relatively high. Indeed, it is surprising that this issue has not been addressed earlier with more em‐ phasis in the literature. Average *WOCAPc,t* values for all countries of our panel for the time span 1998-2009 are presented in Figure 1. Wind overcapacity average values are in line with other authors who addressed capacity factors, like Boccard (2009) and Yang et al. (2012). For example, in Denmark and Portugal, the average *WOCAPc,t* is 0.7790 and 0.7840 respectively, and according to (4) the capacity factor is 0.2210 and 0.2160. It denotes that Nordic countries (e.g. Norway, Sweden, Finland, Denmark, the United Kingdom and Ireland) as well as southern Europe (e.g. Portugal, Spain and Greece) have less idle capacity and therefore more capacity factors than continental countries. This may be because of higher wind speeds in these regions.

#### **3.2. Variables**

*WOCAPc,t* is the dependent variable and represents the ratio of idle capacity in a year to the hypothetical maximum energy that could be produced in a year, in a continuous full-power operation. This ratio was computed from raw data, and can be done in two different ways: (i) through idle capacity; and (ii) through capacity factor. Accordingly, for option (i) the re‐

*TOTALCAPc*,*<sup>t</sup>*

In equation (1) *TOTALCAPc,t* is the total of wind installed capacity. *IDCAPc,t* denotes the idle capacity of wind power in a year. In other words, *IDCAPc,t* represents the difference between maximum possible wind electricity generation during the year (8760 hours) and the amount of electricity actually generated. *TOTALCAPc,t* and *IDCAPc,t* are expressed in Megawatts

where *TOTELECGENc,t* is the total electricity generated in a year, in Gigawatts *per* hour

Regarding option (ii) *WOCAPc,t* can be computed as the difference between 1 and the capaci‐

. (1)

<sup>8760</sup> , (2)

*WOCAPc*,*<sup>t</sup>* =1 - *CF <sup>c</sup>*,*<sup>t</sup>* . (3)

<sup>8760</sup> ≈ 21366.3836 *MW* , (5)

<sup>8760</sup> ≈14781.2648 *MW* . (6)

*TOTCAPc*,*t\**<sup>8760</sup> . (4)

*WOCAPc*,*<sup>t</sup>* <sup>=</sup> *IDCAPc*,*<sup>t</sup>*

*IDCAPc*,*<sup>t</sup>* <sup>=</sup> (*WINDCAPc*,*t\**8760) - (*TOTELECGEN <sup>c</sup>*,*t\**1000)

*CF <sup>c</sup>*,*<sup>t</sup>* <sup>=</sup> *TOTELECGEN <sup>c</sup>*,*t\**<sup>1000</sup>

In expressions (3) and (4) *CFc,t* is the ratio of actual wind power to maximum capacity in a year. For example, for Germany and Spain, which are the leader countries in terms of wind installed capacity, in 2009 the total installed capacity was respectively 25777 MW and 18988 MW. Electricity output was 38637 GWh and 36851 GWh. From here, following equation (2)

In accordance with equation (1), wind overcapacity ratios (*WOCAPc,t*) for Germany and

(GWh). *TOTELECGENc,t* is multiplied by 1000 to convert to same units.

(MW) and this last one is computed as follows:

The capacity factor is computed as follows:

for Germany and Spain, *IDCAPc,t* in 2009 was:

(25777\*8760) - (38637\*1000)

(18988*\**8760) - (36851*\**1000)

ty factor (*CFc,t*) as follows:

Spain are given as:

sult is:

56 New Developments in Renewable Energy

Several causes for idle capacity are suggested by the normative literature. Following this closely, the impact of variables with different natures is controlled for, such as: conventional energy sources; other renewable sources; socio-economic drivers; and energy efficiency measures and public policies as follows.

**Figure 2.** Average conventional energy sources share for the time span 1998-2009

To assess the impact of renewables on wind overcapacity, our option is to use variables rep‐ resenting the most common renewable energy sources such as hydropower, renewable waste and solar energy. Figure 3 presents the growth rate of renewables' share in the 19 countries under analysis. Regarding hydropower, the effect of the capacity factor of hydro‐ power (*CFHYDc,t*) is controlled. This variable was computed similarly to the nuclear capacity factor, according to expression (4) to avoid multicollinearity problems. In the context of the Europe 20-20-20 targets, renewable waste and solar energy have been increasingly used to generate electricity (Münster and Meibom, 2011). To assess the impact of these two energy sources on wind overcapacity, the effect of the share of waste (*WASTSHc,t*) and solar (*SOLSHc,t*) in the total electricity generated is controlled. We also sought to ascertain the im‐ pact of installing more wind power over the years through the growth rate of wind power installed capacity (*WINDGRc,t*). It is expected that the overcapacity of wind power will be

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

http://dx.doi.org/10.5772/52159

59

In Figure 3, values suggest that, in Europe, there has generally been large growth in the share of electricity generated from renewable energy sources. Negative values may indicate that in these countries hydro power is becoming less important in the energy portfolio. For example, Portugal, Germany, Denmark and Ireland have average rates of growth of 10.75%, 9.57%, 10.63% and 11.55% respectively, which indicates huge support in electricity genera‐

Potential socio-economic drivers such as population density or economic development were controlled for in order to assess their effect on wind overcapacity (see Figures 4 and 5 for average values). According to Caralis et al. (2008) and Boccard (2009), the spatial dispersion of wind farms may be an important driver for a greater or lesser capacity factor. To control for the effect of spatial dispersion of wind farms, a proxy, the variable *POPDENSc,t* is used.

*3.2.2. Other renewable energy sources*

positively influenced by more wind power plants.

tion from renewables in the early 2000s.

*3.2.3. Socio-economic drivers*

**Figure 1.** Average *WOCAPc,t* for the time span 1998 – 2009

#### *3.2.1. Conventional energy sources*

To assess the impact of conventional energy sources on wind overcapacity, the shares of fos‐ sil energy sources in total electricity generation across European countries were used (see Figure 2 for average values). The variables are for coal-based power plants (*COALSHc,t*), gasfired (*GASSHc,t*) and oil power plants (*OILSHc,t*). The literature (e.g. Luickx et al., 2008; Øster‐ gaard, 2008; Larraín et al., 2010; and Purvins et al., 2011) argues that these variables are the main sources used to backup wind power, especially coal and gas. In fact, gas turbines can be used as a backup source for wind power in windless periods because their startup times are in the order of a few minutes while other conventional power plants may take much lon‐ ger (Kehlhofer et al., 2009). It is expected that these variables will be highly significant in ex‐ plaining wind overcapacity. Nuclear power is also part of conventional energy sources. The impact of nuclear capacity factor (*CFNUCLc,t*) in wind overcapacity (computed according to (4)) is controlled. Nuclear power still has great importance in Europe, despite its capacity factor reduction by 7.9% between 1998 and 2009. The toxic waste that comes from nuclear power and the fact that it is difficult to treat as well as risk of disaster have recently brought the debate to Germany to reduce its share of nuclear power in electricity generation.

Figure 2 suggests that the 19 countries included in our study still have a large share of con‐ ventional energy sources in total electricity generation, except Nordic countries and Portu‐ gal, which have been at the forefront of the support in renewables, namely wind and solar energy.

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward http://dx.doi.org/10.5772/52159 59

**Figure 2.** Average conventional energy sources share for the time span 1998-2009

#### *3.2.2. Other renewable energy sources*

**Figure 1.** Average *WOCAPc,t* for the time span 1998 – 2009

To assess the impact of conventional energy sources on wind overcapacity, the shares of fos‐ sil energy sources in total electricity generation across European countries were used (see Figure 2 for average values). The variables are for coal-based power plants (*COALSHc,t*), gasfired (*GASSHc,t*) and oil power plants (*OILSHc,t*). The literature (e.g. Luickx et al., 2008; Øster‐ gaard, 2008; Larraín et al., 2010; and Purvins et al., 2011) argues that these variables are the main sources used to backup wind power, especially coal and gas. In fact, gas turbines can be used as a backup source for wind power in windless periods because their startup times are in the order of a few minutes while other conventional power plants may take much lon‐ ger (Kehlhofer et al., 2009). It is expected that these variables will be highly significant in ex‐ plaining wind overcapacity. Nuclear power is also part of conventional energy sources. The impact of nuclear capacity factor (*CFNUCLc,t*) in wind overcapacity (computed according to (4)) is controlled. Nuclear power still has great importance in Europe, despite its capacity factor reduction by 7.9% between 1998 and 2009. The toxic waste that comes from nuclear power and the fact that it is difficult to treat as well as risk of disaster have recently brought

the debate to Germany to reduce its share of nuclear power in electricity generation.

Figure 2 suggests that the 19 countries included in our study still have a large share of con‐ ventional energy sources in total electricity generation, except Nordic countries and Portu‐ gal, which have been at the forefront of the support in renewables, namely wind and solar

*3.2.1. Conventional energy sources*

58 New Developments in Renewable Energy

energy.

To assess the impact of renewables on wind overcapacity, our option is to use variables rep‐ resenting the most common renewable energy sources such as hydropower, renewable waste and solar energy. Figure 3 presents the growth rate of renewables' share in the 19 countries under analysis. Regarding hydropower, the effect of the capacity factor of hydro‐ power (*CFHYDc,t*) is controlled. This variable was computed similarly to the nuclear capacity factor, according to expression (4) to avoid multicollinearity problems. In the context of the Europe 20-20-20 targets, renewable waste and solar energy have been increasingly used to generate electricity (Münster and Meibom, 2011). To assess the impact of these two energy sources on wind overcapacity, the effect of the share of waste (*WASTSHc,t*) and solar (*SOLSHc,t*) in the total electricity generated is controlled. We also sought to ascertain the im‐ pact of installing more wind power over the years through the growth rate of wind power installed capacity (*WINDGRc,t*). It is expected that the overcapacity of wind power will be positively influenced by more wind power plants.

In Figure 3, values suggest that, in Europe, there has generally been large growth in the share of electricity generated from renewable energy sources. Negative values may indicate that in these countries hydro power is becoming less important in the energy portfolio. For example, Portugal, Germany, Denmark and Ireland have average rates of growth of 10.75%, 9.57%, 10.63% and 11.55% respectively, which indicates huge support in electricity genera‐ tion from renewables in the early 2000s.

#### *3.2.3. Socio-economic drivers*

Potential socio-economic drivers such as population density or economic development were controlled for in order to assess their effect on wind overcapacity (see Figures 4 and 5 for average values). According to Caralis et al. (2008) and Boccard (2009), the spatial dispersion of wind farms may be an important driver for a greater or lesser capacity factor. To control for the effect of spatial dispersion of wind farms, a proxy, the variable *POPDENSc,t* is used.

**Figure 3.** Average rate of growth in renewables' share in total electricity generation (including Hydro) for the time span 1999-2009

**Figure 5.** Average natural logarithm of GDP *per capita* for the time span 1998-2009

the eco-tax on energy consumption or CO2 emissions and other eco-taxes.

Norway holds the largest value for this indicator.

*3.2.4. Energy efficiency measures and public policies*

Latvia is the country with the smallest average natural logarithm of GDP *per capita*, while

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

http://dx.doi.org/10.5772/52159

61

Energy policies and measures have been widely used to promote and support the deploy‐ ment of renewables. Data from the Mesures d'Utilisation Rationnelle de l'Energie (MURE) database was collected, which provides information concerning the amount and the impact of these measures in order to control for the influence of energy policies on overcapacity. The variables are the cumulative amount of measures taken per household and in the indus‐ trial and tertiary sectors, as presented in Figure 6. Firstly, the total energy policies and meas‐ ures carried out in a year (*ALLPOLc,t*) were considered. It is expected that the total measures may have a positive impact on wind power overcapacity. For a deeper analysis of public policies, the total energy policies are divided into seven individual types to assess the influ‐ ence of each type individually: (i) Legislative/normative (*NORMPOLc,t*) stands for mandato‐ ry standards for buildings, regulations for heating systems and hot water systems, regulations in the field of building and mandatory standards for electrical appliances; (ii) legislative/informative (*INFOPOLc,t*) aims to inform about energy efficiency, mandatory standards in buildings and electrical appliances; (iii) fiscal/tariff (*FISCPOLc,t*) measures in‐ clude tax exemptions/reductions in retrofitting investments; (iv) incentives/subsidies (*FIN‐ POLc,t*) includes feed-in tariffs, grants and loans. A positive effect of these measures on *WOCAPc,t* is expected, due to their contribution to renewables' deployment which may posi‐ tively influence wind overcapacity; (v) information/education (*EDUPOLc,t*) measures aim to provide campaigns by energy agencies and energy suppliers; (vi) co-operative measures (*COOPPOLc,t*) include voluntary programs; and (vii) cross-cutting measures (*CUTPOLc,t*) are

**Figure 4.** Average population density for the time span 1998-2009

This variable assesses the effect of available and suitable land area for wind park installation in countries with greater or lesser population density. The Netherlands, followed by the United Kingdom, reveal the largest population density in the panel.

Regarding economic development, the natural logarithm of Gross Domestic Product (GDP) *per capita* (*LNGDPPCc,t*) is used to measure the capacity of European countries to invest in more efficient energy generation technologies. It is expected that more developed countries will have greater available financial resources to invest in more efficient energy sources, such as offshore wind parks.

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward http://dx.doi.org/10.5772/52159 61

**Figure 5.** Average natural logarithm of GDP *per capita* for the time span 1998-2009

Latvia is the country with the smallest average natural logarithm of GDP *per capita*, while Norway holds the largest value for this indicator.

#### *3.2.4. Energy efficiency measures and public policies*

**Figure 3.** Average rate of growth in renewables' share in total electricity generation (including Hydro) for the time

This variable assesses the effect of available and suitable land area for wind park installation in countries with greater or lesser population density. The Netherlands, followed by the

Regarding economic development, the natural logarithm of Gross Domestic Product (GDP) *per capita* (*LNGDPPCc,t*) is used to measure the capacity of European countries to invest in more efficient energy generation technologies. It is expected that more developed countries will have greater available financial resources to invest in more efficient energy sources,

span 1999-2009

60 New Developments in Renewable Energy

**Figure 4.** Average population density for the time span 1998-2009

such as offshore wind parks.

United Kingdom, reveal the largest population density in the panel.

Energy policies and measures have been widely used to promote and support the deploy‐ ment of renewables. Data from the Mesures d'Utilisation Rationnelle de l'Energie (MURE) database was collected, which provides information concerning the amount and the impact of these measures in order to control for the influence of energy policies on overcapacity. The variables are the cumulative amount of measures taken per household and in the indus‐ trial and tertiary sectors, as presented in Figure 6. Firstly, the total energy policies and meas‐ ures carried out in a year (*ALLPOLc,t*) were considered. It is expected that the total measures may have a positive impact on wind power overcapacity. For a deeper analysis of public policies, the total energy policies are divided into seven individual types to assess the influ‐ ence of each type individually: (i) Legislative/normative (*NORMPOLc,t*) stands for mandato‐ ry standards for buildings, regulations for heating systems and hot water systems, regulations in the field of building and mandatory standards for electrical appliances; (ii) legislative/informative (*INFOPOLc,t*) aims to inform about energy efficiency, mandatory standards in buildings and electrical appliances; (iii) fiscal/tariff (*FISCPOLc,t*) measures in‐ clude tax exemptions/reductions in retrofitting investments; (iv) incentives/subsidies (*FIN‐ POLc,t*) includes feed-in tariffs, grants and loans. A positive effect of these measures on *WOCAPc,t* is expected, due to their contribution to renewables' deployment which may posi‐ tively influence wind overcapacity; (v) information/education (*EDUPOLc,t*) measures aim to provide campaigns by energy agencies and energy suppliers; (vi) co-operative measures (*COOPPOLc,t*) include voluntary programs; and (vii) cross-cutting measures (*CUTPOLc,t*) are the eco-tax on energy consumption or CO2 emissions and other eco-taxes.

**Variable Definition Source Obs. Mean SD Min. Max.**

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

World Bank, World Development Indicator Database

MURE DATABASE

MURE DATABASE

MURE DATABASE

MURE DATABASE

MURE DATABASE

MURE DATABASE

MURE DATABASE

MURE DATABASE

Notes: MURE DATABASE stands for Mesures d'Utilisation Rationnelle de l'Energie (MURE II Database); co-ordinated by the Institute of Studies for the Integration of Systems and the Fraunhofer Institute for Systems and Innovation Re‐ search ISI. IEA stands for International Energy Agency Data Services and EUROSTAT stands for Eurostat Statistics Data‐ base available at http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/search\_database with the code

Figure 6 indicates that there has been major support for renewables through energy policies

and measures especially in Germany, France, Finland and the United Kingdom.

228 9.7194 0.6774 7.9737 10.6431

http://dx.doi.org/10.5772/52159

63

228 29.8553 18.9586 0 82

228 6.9035 6.0468 0 36

221 1.0905 1.8367 0 7

228 3.2807 3.2434 0 13

228 8.6754 7.0079 0 26

228 5.5526 4.7159 0 22

218 2.7456 3.0511 0 16

228 1.6404 3.5199 0 16

Logarithm of Gross Domestic Product per

Total of Accumulated Number of Renewable Energy Policies and Measures

Accumulated Number of Renewable Energy Policies and Measures – Normative/Legislative

Accumulated Number of Renewable Energy Policies and Measures – Tariff/

Accumulated Number of Renewable Energy Policies and Measures – Legislative/informative Legislative/

Accumulated Number of Renewable Energy Policies and Measures –

Accumulated Number of Renewable Energy Policies and Measures –

Accumulated Number of Renewable Energy Policies and Measures – Co-

Accumulated Number of Renewable Energy Policies and Measures – Cross-

**Table 1.** Variables definition, sources and summary statistics

*LNGDPPCc,t*

*ALLPOLc,t*

*NORMPOLc,t*

*FISCPOLc,t*

*INFOPOLc,t*

*FINPOLc,t*

*EDUPOLc,t*

*COOPPOLc,t*

*CUTPOLc,t*

nrg\_113a.

Capita

fiscal

informative

Educational

operative

cutting

Incentives/subsidies

**Figure 6.** Average number of accumulated energy policies and measures for the time span 1998-2009


On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward http://dx.doi.org/10.5772/52159 63


Notes: MURE DATABASE stands for Mesures d'Utilisation Rationnelle de l'Energie (MURE II Database); co-ordinated by the Institute of Studies for the Integration of Systems and the Fraunhofer Institute for Systems and Innovation Re‐ search ISI. IEA stands for International Energy Agency Data Services and EUROSTAT stands for Eurostat Statistics Data‐ base available at http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/search\_database with the code nrg\_113a.

**Table 1.** Variables definition, sources and summary statistics

**Figure 6.** Average number of accumulated energy policies and measures for the time span 1998-2009

Ratio of non-used output to the maximum possible output over a year

elect. gen. (TWh)

62 New Developments in Renewable Energy

elect. gen. (TWh)

elect. gen. (TWh)

elect. gen. (TWh)

elect. gen. (TWh)

*POPDENSc,t* Population density (people/km2)

capacity

Ratio of elect. gen. to coal (TWh)/total

Ratio of elect. gen. to gas (TWh)/total

Ratio of elect. gen. to oil (TWh)/total

Ratio of average plant output to the maximum possible output over a year

Ratio of average plant output to the maximum possible output over a year

Ratio of elect. gen. to waste (TWh)/total

Ratio of elect. gen. to solar (TWh)/total

Yearly growth rate of wind installed

*WOCAPc,t*

*COALSHc,t*

*GASSHc,t*

*OILSHc,t*

*CFNUCLc,t*

*CFHYDc,t*

*WASTSHc,t*

*SOLSHc,t*

*WINDGRc,t*

**Variable Definition Source Obs. Mean SD Min. Max.**

World Bank, World Development Indicators Database

EUROSTAT 221 0.7956 0.0543 0.5947 0.9912

IEA 227 0.2940 0.2483 0 0.9636

IEA 227 0.2161 0.1761 0.0015 0.6339

IEA 227 0.0564 0.0770 0.0001 0.4243

IEA 228 0.4324 0.4164 0 0.9659

IEA 228 0.2884 0.1231 0.0948 0.6223

IEA 227 0.0308 0.0348 0 0.1486

IEA 227 0.0004 0.0018 0 0.0210

EUROSTAT 223 50.5234 98.1664 -7.1429 1000

228 139.6083 115.6765 14.5655 489.6442

Figure 6 indicates that there has been major support for renewables through energy policies

and measures especially in Germany, France, Finland and the United Kingdom.

#### **3.3. Data**

In this chapter data from several sources such as the Eurostat database, International Energy Agency (IEA), World Bank and MURE Database is used. Table 1 presents the variables, their definition, sources and summary statistics for the time span 1998-2009.

*POPDENSc,t LNGDPPCc,t NORMPOLc,t FISCPOLc,t FINPOLc,t EDUPOLc,t COOPPOLc,t CUTPOLc,t*

http://dx.doi.org/10.5772/52159

65

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

*POPDENSc,t* 1

*LNGDPPCc,t* 0.1445 1

among variables is not a problem.

**Table 2.** Correlation Matrix

*NORMPOLc,t* 0.1375 0.2511 1

*FISCPOLc,t* 0.557 0.3284 0.1434 1

*FINPOLc,t* 0.215 0.2488 0.3877 0.3986 1

*Mean VIF* 2.36

**Table 3.** Variance Inflation Factor (VIF)

*EDUPOLc,t* -0.0983 0.3927 0.1509 0.2376 0.4939 1

*COOPPOLc,t* 0.2422 0.4507 0.1037 0.3808 0.3564 0.3814 1

**Variables VIF 1/VIF** *COALSH c,t* 2.33 0.4292 *GASSH c,t* 2.42 0.414 *OILSH c,t* 1.81 0.5522 *CFNUCL c,t* 2.76 0.363 *CFHYD c,t* 2.92 0.3423 *WASTSH c,t* 1.95 0.5131 *WINDGR c,t* 1.11 0.8993 *SOLSH c,t* 1.64 0.6092 *POPDENS c,t* 3.47 0.2881 *LNGDPPC c,t* 2.22 0.4514 *NORMPOL c,t* 2.37 0.4226 *FISCPOL c,t* 2.41 0.415 *FINPOL c,t* 3.29 0.304 *EDUPOL c,t* 2.19 0.4556 *COOPPOL c,t* 2.66 0.3758 *CUTPOL c,t* 2.16 0.4621

*CUTPOLc,t* 0.275 0.3259 0.0897 0.1559 0.4795 0.3185 0.2131 1

Notwithstanding, the Variance Inflation Factor (VIF) test for multicollinearity among variables was performed. Individ‐ ual values are below 5 for all individual tests and 2.36 for mean VIF (Table 3), which reinforces that multicollinearity

#### **3.4. Methods**

To make a proper empirical analysis the panel dataset structure was analyzed, which gener‐ ally has a complex nature of term error composition. Several methods were applied: (i) visu‐ al analysis of data; (ii) test for first-order autocorrelation in panel data; (iii) test for the presence of groupwise heteroskedasticity; and (iv) test for contemporaneous correlation. Stata v11.2 econometric software was used.

The correlation matrix (Table 2) values suggest that correlation coefficients are low and do not suggest the existence of collinearity among the variables.



Notwithstanding, the Variance Inflation Factor (VIF) test for multicollinearity among variables was performed. Individ‐ ual values are below 5 for all individual tests and 2.36 for mean VIF (Table 3), which reinforces that multicollinearity among variables is not a problem.

#### **Table 2.** Correlation Matrix

**3.3. Data**

64 New Developments in Renewable Energy

**3.4. Methods**

*WOCAPc,t* 1

*COALSHc,t* -0.0365 1

*GASSHc,t* -0.1244 -0.1455 1

*OILSHc,t* -0.2115 0.071 0.2375 1

*CFNUCLc,t* 0.2218 -0.1512 -0.0172 -0.4296 1

*CFHYDc,t* 0.0557 -0.4607 -0.2358 -0.2613 0.1269 1

*WASTSHc,t* 0.0412 -0.0863 0.0188 -0.2497 0.2778 0.4466 1

*WINDGRc,t* 0.4853 -0.0079 -0.1081 -0.0593 -0.0462 -0.0219 -0.1532 1

*SOLSHc,t* 0.0203 -0.0003 0.0637 -0.0357 0.1309 -0.088 0.0184 -0.0723 1

*POPDENSc,t* 0.1344 0.1529 0.5052 -0.0721 0.3865 -0.33 -0.0177 -0.0968 0.0592 *LNGDPPCc,t* -0.1772 -0.3668 0.0691 -0.0704 0.1725 0.3844 0.301 -0.2477 0.0294 *NORMPOLc,t* 0.0936 -0.2008 0.2193 0.1654 0.0503 -0.1347 -0.0908 -0.109 0.4542 *FISCPOLc,t* -0.0087 -0.1839 0.2779 -0.2763 0.4015 -0.1103 0.0589 -0.0683 -0.068 *FINPOLc,t* -0.1705 -0.1179 -0.027 -0.3426 0.5317 0.169 0.088 -0.0178 0.1358 *EDUPOLc,t* -0.0832 -0.2672 0.0633 -0.3603 0.1555 0.3154 0.252 -0.0749 0.0879 *COOPPOLc,t* -0.008 -0.1613 0.2452 -0.2366 0.4922 0.3567 0.5535 -0.1681 0.0922 *CUTPOLc,t* 0.0634 0.0865 -0.0564 -0.2202 0.2274 0.175 -0.0224 -0.082 0.249

Stata v11.2 econometric software was used.

not suggest the existence of collinearity among the variables.

In this chapter data from several sources such as the Eurostat database, International Energy Agency (IEA), World Bank and MURE Database is used. Table 1 presents the variables, their

To make a proper empirical analysis the panel dataset structure was analyzed, which gener‐ ally has a complex nature of term error composition. Several methods were applied: (i) visu‐ al analysis of data; (ii) test for first-order autocorrelation in panel data; (iii) test for the presence of groupwise heteroskedasticity; and (iv) test for contemporaneous correlation.

The correlation matrix (Table 2) values suggest that correlation coefficients are low and do

*WOCAPc,t COALSHc,t GASSHc,t OILSHc,t CFNUCLc,t CFHYDc,t WASTSHc,t WINDGRc,t SOLSHc,t*

definition, sources and summary statistics for the time span 1998-2009.


**Table 3.** Variance Inflation Factor (VIF)

As part of the empirical research using panel dataset techniques, several tests to detect com‐ mon panel phenomena in errors structure were performed (see Table 4 for results). The Wooldridge test with the null hypothesis of no first-order autocorrelation to detect serial correlation in the idiosyncratic errors of panel-data (Wooldridge, 2002) was performed. This test follows a normal distribution N(0,1) in Ordinary Least Squares (OLS) estimator. Fur‐ thermore, a modified Wald test was applied to search for the presence of groupwise hetero‐ skedasticity in the residuals of a fixed effect (FE) regression model, which assumes homoskedasticity across cross-sections. The modified Wald Test has χ<sup>2</sup> distribution and tests the null hypothesis of: *σ<sup>c</sup>* <sup>2</sup> =*σ* <sup>2</sup> for*c* =1, …, *N* where*σ* <sup>2</sup> is the variance of the *c*country (Greene, 2000). As stated by Marques and Fuinhas (2012b), if one considers that European countries are guided by common energy guidelines, one might expect the presence of con‐ temporaneous correlation in our panel. In order to detect this phenomenon, or rather, test the null hypothesis of cross-section independence, Pesaran (2004), Frees (1995 and 2004), and Friedman (1937) tests were performed. While Pesaran follows a standard normal distri‐ bution, the Frees statistic test uses Frees Q-distribution and Friedman uses Friedman's chisquare distributed statistic. Frees and Friedman perform only with available data for all cross-sections. Hausman's statistics test the null hypothesis that the difference of coefficients between fixed-effects and random-effects is not systematic.

The OLS estimator proves to be consistent when there is no presence of multicollinearity among the explanatory variables and when the regressors are exogenous. It is optimal

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

homoskedastic following*E*(*ε*)=0. Therefore, in our case it may be useful to benchmark re‐ sults of our panel estimation. Moreover, we apply the panel fixed-effects (FE) and ran‐ dom-effects estimators (RE). Using the fixed-effects estimator appears to be appropriate in studying the impact of variables that vary over time. It explores the different variables within groups that have their own characteristics, in our case European countries. The fixed-effects estimator assumes that something time-invariant within groups can affect the dependent variable and cannot be correlated with other groups. In turn, random effects assume that variation across groups is random and not correlated to the dependent and

<sup>2</sup> | *NT* and when the errors are

http://dx.doi.org/10.5772/52159

67

*β<sup>k</sup> Xk* ,*c*,*<sup>t</sup>* + *dt* + *εc*,*<sup>t</sup>* , (13)

.

when there is no serial auto-correlation following *V* (*ε*)=*σε*

*WOCAPc*,*<sup>t</sup>* = *α* + ∑

skedastic with no serial correlation. The dummy for time is denoted by *dt*

**4. Empirical evidence of the drivers of wind overcapacity**

*k*=1 *k*

where the error term is *εc*,*<sup>t</sup>* =*α<sup>c</sup>* + *uc*,*t* with *α<sup>c</sup>* uncorrelated with the regressors and *εc*,*t* homo‐

The estimation results are shown in Table 5. Conventional standard errors (CSE) are provid‐ ed, as are robust standard errors (RSE) to deal with the presence of heteroskedasticity. Mod‐ els *I* and *II* represent pooled OLS; models *III* and *IV* are panel fixed-effect estimators (FE); and models *V* and *VI* stand for random-effect estimators (RE). The error term is*εc*,*<sup>t</sup>* =*α<sup>c</sup>* + *uc*,*t*.

A battery of diagnostic tests was applied to test the quality of the estimators. The Breusch-Pagan Lagrange multiplier (LM) is provided to test whether the RE estimator is more suitable than the OLS estimator. The results show that the null hypothesis of varian‐ ces across groups being equal to zero is rejected, so there is a significant difference across groups. Accordingly, the RE estimator is more suitable than Pooled OLS. The Hausman test to choose the most appropriate estimator between FE and RE was applied. The null hypothesis assumes that the difference in coefficients is not systematic, thus accepting RE over FE estimator (Greene, 2008). The Hausman test accepts the null hypothesis, thus the errors *αc* are uncorrelated with the regressors. As a consequence, the discussion will be based on RE estimator with RSE (*VI*). In other words, it seems that differences across countries influence *WOCAPc,t*, so the panel RE estimator is more appropriate than FE to

independent variables.

our analysis.

The generic model to estimate is:


Note: \*\*\*, \*\* denote significance at 1 and 5% significance levels

#### **Table 4.** Specification tests and statistics

According to table 4 results, the Wooldridge test value (3.48) does not reject the null hypoth‐ esis of no first-order autocorrelation. Accordingly, the autoregressive (AR1) estimator is not suitable. The modified Wald test value (749.41) suggests rejection of the null hypothesis of errors homoskedasticity within cross-sections. Therefore, the presence of groupwise hetero‐ skedasticity is confirmed. As far as the presence of contemporaneous correlation is con‐ cerned, with the exception of the Pesaran test for random effects (-2.061), generally the null hypothesis of no contemporaneous correlation was not rejected, suggesting that there is spa‐ tial independence across European countries. This is not surprising given the technical na‐ ture of our research into the interaction of conventional sources and renewables with wind overcapacity instead of common policy guidelines.

The OLS estimator proves to be consistent when there is no presence of multicollinearity among the explanatory variables and when the regressors are exogenous. It is optimal when there is no serial auto-correlation following *V* (*ε*)=*σε* <sup>2</sup> | *NT* and when the errors are homoskedastic following*E*(*ε*)=0. Therefore, in our case it may be useful to benchmark re‐ sults of our panel estimation. Moreover, we apply the panel fixed-effects (FE) and ran‐ dom-effects estimators (RE). Using the fixed-effects estimator appears to be appropriate in studying the impact of variables that vary over time. It explores the different variables within groups that have their own characteristics, in our case European countries. The fixed-effects estimator assumes that something time-invariant within groups can affect the dependent variable and cannot be correlated with other groups. In turn, random effects assume that variation across groups is random and not correlated to the dependent and independent variables.

The generic model to estimate is:

As part of the empirical research using panel dataset techniques, several tests to detect com‐ mon panel phenomena in errors structure were performed (see Table 4 for results). The Wooldridge test with the null hypothesis of no first-order autocorrelation to detect serial correlation in the idiosyncratic errors of panel-data (Wooldridge, 2002) was performed. This test follows a normal distribution N(0,1) in Ordinary Least Squares (OLS) estimator. Fur‐ thermore, a modified Wald test was applied to search for the presence of groupwise hetero‐ skedasticity in the residuals of a fixed effect (FE) regression model, which assumes

for*c* =1, …, *N* where*σ* <sup>2</sup>

(Greene, 2000). As stated by Marques and Fuinhas (2012b), if one considers that European countries are guided by common energy guidelines, one might expect the presence of con‐ temporaneous correlation in our panel. In order to detect this phenomenon, or rather, test the null hypothesis of cross-section independence, Pesaran (2004), Frees (1995 and 2004), and Friedman (1937) tests were performed. While Pesaran follows a standard normal distri‐ bution, the Frees statistic test uses Frees Q-distribution and Friedman uses Friedman's chisquare distributed statistic. Frees and Friedman perform only with available data for all cross-sections. Hausman's statistics test the null hypothesis that the difference of coefficients

According to table 4 results, the Wooldridge test value (3.48) does not reject the null hypoth‐ esis of no first-order autocorrelation. Accordingly, the autoregressive (AR1) estimator is not suitable. The modified Wald test value (749.41) suggests rejection of the null hypothesis of errors homoskedasticity within cross-sections. Therefore, the presence of groupwise hetero‐ skedasticity is confirmed. As far as the presence of contemporaneous correlation is con‐ cerned, with the exception of the Pesaran test for random effects (-2.061), generally the null hypothesis of no contemporaneous correlation was not rejected, suggesting that there is spa‐ tial independence across European countries. This is not surprising given the technical na‐ ture of our research into the interaction of conventional sources and renewables with wind

**Pooled OLS Fixed Effects (FE) Random Effects (RE)**

distribution and tests

is the variance of the *c*country

homoskedasticity across cross-sections. The modified Wald Test has χ<sup>2</sup>

<sup>2</sup> =*σ* <sup>2</sup>

between fixed-effects and random-effects is not systematic.

Note: \*\*\*, \*\* denote significance at 1 and 5% significance levels

overcapacity instead of common policy guidelines.

) 749.41\*\*\* Pesaran test -1.717 -2.061\*\* Frees test 0.744 1.191 Friedman test 7.053 4.605

the null hypothesis of: *σ<sup>c</sup>*

66 New Developments in Renewable Energy

Wooldridge test F(N(0,1)) 3.48

**Table 4.** Specification tests and statistics

Modified Wald test (χ <sup>2</sup>

$$\text{WOCAP}\_{c,t} = \alpha + \sum\_{k=1}^{k} \beta\_k X\_{k,c,t} + d\_t + \varepsilon\_{c,t} \tag{13}$$

where the error term is *εc*,*<sup>t</sup>* =*α<sup>c</sup>* + *uc*,*t* with *α<sup>c</sup>* uncorrelated with the regressors and *εc*,*t* homo‐ skedastic with no serial correlation. The dummy for time is denoted by *dt* .

#### **4. Empirical evidence of the drivers of wind overcapacity**

The estimation results are shown in Table 5. Conventional standard errors (CSE) are provid‐ ed, as are robust standard errors (RSE) to deal with the presence of heteroskedasticity. Mod‐ els *I* and *II* represent pooled OLS; models *III* and *IV* are panel fixed-effect estimators (FE); and models *V* and *VI* stand for random-effect estimators (RE). The error term is*εc*,*<sup>t</sup>* =*α<sup>c</sup>* + *uc*,*t*.

A battery of diagnostic tests was applied to test the quality of the estimators. The Breusch-Pagan Lagrange multiplier (LM) is provided to test whether the RE estimator is more suitable than the OLS estimator. The results show that the null hypothesis of varian‐ ces across groups being equal to zero is rejected, so there is a significant difference across groups. Accordingly, the RE estimator is more suitable than Pooled OLS. The Hausman test to choose the most appropriate estimator between FE and RE was applied. The null hypothesis assumes that the difference in coefficients is not systematic, thus accepting RE over FE estimator (Greene, 2008). The Hausman test accepts the null hypothesis, thus the errors *αc* are uncorrelated with the regressors. As a consequence, the discussion will be based on RE estimator with RSE (*VI*). In other words, it seems that differences across countries influence *WOCAPc,t*, so the panel RE estimator is more appropriate than FE to our analysis.


Globally, the estimation results provided in Table 5 reveal consistency despite some dif‐ ferences between significance levels. By examining the variables in descending order, for fossil fuels, the effect of *COALSHc,t* and *GASSHc,t* proved to be negative and statistically significant at 5% and 1% respectively. On the other hand, variable *OILSHc,t* does not seem to be significant statistically. This result is in line with expectations, revealing that there is backup for wind power using fossil fuels like coal and gas to overcome intermittency. Oil power plants are not generally used for backup, so these results may reveal our model's

On the Public Policies Supporting Renewables and Wind Power Overcapacity: Insights into the European Way Forward

http://dx.doi.org/10.5772/52159

69

*CFNUCLc,t*, *CFHYDc,t* and *SOLSHc,t* coefficients reveal no statistical relationship between the capacity factor of nuclear and hydro or the share of solar energy and wind power overcap‐ acity. The effects of variables *WINDGRc,t*, *POPDENSc,t*, *LNGDPPCc,t* and *ALLPOLc,t* are posi‐ tive and statistically significant. Therefore, it is assumed that they are important drivers in explaining wind overcapacity. Results from disaggregated policies are presented in table A. 1. None of the individual energy policies proves to be significant in explaining wind over‐ capacity except *NORMPOLc,t* and *FISCPOLc,t*. However, it is worth noting that there is no in‐ clusion of the legislative/informative policies (*INFOPOLc,t*) due to their identical nature and the fact that they could create collinearity problems. Contrary to expectations, financial poli‐

Exclusion tests were run for the explanatory variables (*FINPOLc,t*, *EDUPOLc,t*, *COOP‐ POLc,t* and *CUTPOLc,t*), which do not reveal a statistical significance, following the parsi‐ monious principle. The results are shown in Table A.2. In fact, the models maintain robustness among the estimators for all coefficients with or without these individual pol‐ icies. This set of variables has no influence either on the ratio of non-used wind capacity (*WOCAPc,t*) or on the remaining model. Nevertheless, given that consistency and robust‐ ness are crucial properties, a subsection is opened to provide additional analysis on the

According to Huber (1973), as further evidence of the robustness of results, it is appropriate to apply the robust regression (RREG) estimator to cope with possible outliers from our da‐ taset. These outliers can impair the stability and reliability of results. Such as in Marques and Fuinhas (2012b), robustness is analyzed by providing the robust regression with Huber

As shown from this additional assessment of the robustness of results, the variables main‐ tain their signs, though with small differences in significance levels. In general, the robust

cies have no statistical relation to wind overcapacity.

**5. Consistency and robustness of empirical evidence**

and Tukey weight functions, as presented in table 6.

regression validates the main results of the estimations.

robustness.

reliability of results.

Notes: OLS - Ordinary Least Squares. RE – Random Effects. FE – Fixed Effects. CSE – Conventional standard errors. The Ftest has normal distribution N(0,1) and tests the null hypothesis of non-significance of all estimated parameters. The Wald test has χ <sup>2</sup> distribution and tests the null hypothesis of non-significance of all coefficients of independent varia‐ bles. The LM test has χ <sup>2</sup> distribution and tests the null hypothesis of non-relevance of individual effects in the RE mod‐ el. The Hausman test has χ <sup>2</sup> distribution and tests the null hypothesis of the difference in coefficients not being systematic between two selected estimators. Standard errors are reported in brackets. All estimates were controlled to include time effects, although they are not reported for reasons of simplicity. \*\*\*, \*\*, denote significance at 1 and 5% significance levels respectively for both coefficient estimators and test statistics.

**Table 5.** Regression results - Dependent Variable *WOCAPc,t*

Globally, the estimation results provided in Table 5 reveal consistency despite some dif‐ ferences between significance levels. By examining the variables in descending order, for fossil fuels, the effect of *COALSHc,t* and *GASSHc,t* proved to be negative and statistically significant at 5% and 1% respectively. On the other hand, variable *OILSHc,t* does not seem to be significant statistically. This result is in line with expectations, revealing that there is backup for wind power using fossil fuels like coal and gas to overcome intermittency. Oil power plants are not generally used for backup, so these results may reveal our model's robustness.

**OLS FE RE** *Ind. Variables CSE(I) RSE(II) CSE(III) RSE(IV) CSE(V) RSE(VI)*

> -0.0696 (0.1264)

> -0.1174 (0.1330)


0.0893 (0.0785)



0.0002\*\*\* (0.0000)

1.5059 (2.0191)

0.0021 (0.0014)

0.0878 (0.0563)

0.0009\*\* (0.0004)


Notes: OLS - Ordinary Least Squares. RE – Random Effects. FE – Fixed Effects. CSE – Conventional standard errors. The Ftest has normal distribution N(0,1) and tests the null hypothesis of non-significance of all estimated parameters. The

systematic between two selected estimators. Standard errors are reported in brackets. All estimates were controlled to include time effects, although they are not reported for reasons of simplicity. \*\*\*, \*\*, denote significance at 1 and 5%

distribution and tests the null hypothesis of non-significance of all coefficients of independent varia‐

distribution and tests the null hypothesis of non-relevance of individual effects in the RE mod‐

distribution and tests the null hypothesis of the difference in coefficients not being

*N* 218 218 218 218 218 218

*Wald (*χ *2)* 136.63\*\*\*

*LM (*χ *2)* 11.76\*\*\*



0.0893 (0.0619)



0.0002\*\*\* (0.0001)

1.5059 (0.8716)

0.0021 (0.0011)

0.0878 (0.0511)

0.0009\*\*\* (0.0003)







0.0049 (0.0386)

0.2666\*\* (0.1174)

0.0002\*\*\* (0.0000)

0.0001\*\*\* (0.0000)


0.0008\*\*\* (0.0002)

1.0593\*\*\* (0.0666)

1.1403 (1.8146) -0.0381\*\* (0.0192)




0.0049 (0.0415)

0.2666\*\* (0.0849)

0.0002\*\*\* (0.0000)

0.0001\*\*\* (0.0000)


0.0008\*\*\* (0.0002)

1.0593\*\*\* (0.0805)

1.1403 (0.9527)





0.0118 (0.0402)

0.2878\*\*\* (0.0836)

0.0002\*\*\* (0.0001)

0.0002\*\*\* (0.0000)


0.0008\*\*\* (0.0003)

1.0650\*\*\* (0.0746)

*R2* 0.4316 0.4316 0.3623 0.3623

significance levels respectively for both coefficient estimators and test statistics.

1.2293 (1.0963)

*COALSHc,t*

*GASSHc,t*

*OILSHc,t*

*CFNUCLc,t*

*CFHYDc,t*

*WASTSHc,t*

*WINDGRc,t*

*SOLSHc,t*

*POPDENSc,t*

*LNGDPPCc,t*

*ALLPOLc,t*

*CONST*

Wald test has χ <sup>2</sup>

bles. The LM test has χ <sup>2</sup>

el. The Hausman test has χ <sup>2</sup>


68 New Developments in Renewable Energy




0.0118 (0.0362)

0.2878\*\*\* (0.1084)

0.0002\*\*\* (0.0000)

0.0002\*\*\* (0.0000)


0.0008\*\*\* (0.0002)

1.0650\*\*\* (0.0617)

*F (N(0,1))* 7.09\*\*\* 4.82\*\*\*

*Hausman (*χ *2)* 30.93

**Table 5.** Regression results - Dependent Variable *WOCAPc,t*

1.2293 (1.8236) *CFNUCLc,t*, *CFHYDc,t* and *SOLSHc,t* coefficients reveal no statistical relationship between the capacity factor of nuclear and hydro or the share of solar energy and wind power overcap‐ acity. The effects of variables *WINDGRc,t*, *POPDENSc,t*, *LNGDPPCc,t* and *ALLPOLc,t* are posi‐ tive and statistically significant. Therefore, it is assumed that they are important drivers in explaining wind overcapacity. Results from disaggregated policies are presented in table A. 1. None of the individual energy policies proves to be significant in explaining wind over‐ capacity except *NORMPOLc,t* and *FISCPOLc,t*. However, it is worth noting that there is no in‐ clusion of the legislative/informative policies (*INFOPOLc,t*) due to their identical nature and the fact that they could create collinearity problems. Contrary to expectations, financial poli‐ cies have no statistical relation to wind overcapacity.

Exclusion tests were run for the explanatory variables (*FINPOLc,t*, *EDUPOLc,t*, *COOP‐ POLc,t* and *CUTPOLc,t*), which do not reveal a statistical significance, following the parsi‐ monious principle. The results are shown in Table A.2. In fact, the models maintain robustness among the estimators for all coefficients with or without these individual pol‐ icies. This set of variables has no influence either on the ratio of non-used wind capacity (*WOCAPc,t*) or on the remaining model. Nevertheless, given that consistency and robust‐ ness are crucial properties, a subsection is opened to provide additional analysis on the reliability of results.
