Household and industrial electricity demand in Europe

Energy Policy 122 (2018) 592–600

Contents lists available at ScienceDirect

Energy Policy journal homepage: www.elsevier.com/locate/enpol

Household and industrial electricity demand in Europe ⁎

T

Catia Cialani , Reza Mortazavi School of Technology and Business Studies, Dalarna University, 791 88 Falun, Sweden

A R T I C LE I N FO

A B S T R A C T

Keywords: Electricity demand Price elasticity Europe EU-29 GMM ML

This paper examines the electricity demand, and its determinants, in 29 European countries during the liberalization of the electricity market. Based on panel data for these countries for the years 1995–2015 and using a dynamic partial adjustment model, price elasticities are estimated for both residential and industrial electricity demand. These elasticities and effects of other variables on electricity consumption are estimated using both GMM (generalized method of moments) and ML (maximum likelihood) approaches. It is found that the price elasticities are very small, especially in the short run, while the income elasticities are relatively large, especially for households and in the long run.

1. Introduction During the last two decades, liberalization of the electricity sector (i.e., opening electricity markets to competition, thereby increasing freedom of choice for consumers) has spread globally. The movement started in a few countries (among others, the United Kingdom, Norway and Australia) in the early 1990s, and was subsequently embraced by other countries including Spain, Germany, and Italy. By the mid-1990s the European Union (EU) was also committed to the process (Directive 96/92/EC).1 Central objectives of the EU's liberalization policy were to increase welfare by reducing electricity prices for consumers, guarantee security of supply throughout the EU, promote energy efficiency and the use of renewable energy resources, and raise economic efficiency (Willems and Ehlers, 2008). However, contrary to expectations, electricity prices generally increased for most of European countries following liberalization from 1995 to 2015 (Eurostat, 2018). According to economic theory, this should have led to a reduction in electricity consumption, but more information about the interactions involved is required. In particular, knowledge of consumers’ sensitivity to changes in electricity prices is extremely important for activities such as reorganizing production, adjusting controls, planning energy or intermediate product storage systems, and provision of appropriate backup capacities or substitute energy sources (Kirschen et al., 2000). Thus, the purpose of this paper is to explore the short- and long-term elasticity of demand for electricity, and determinants of the demand, during the liberalization period in the EU-29 (EU-28 plus Norway). The findings are expected to facilitate efforts to plan and organize electricity supplies robustly and efficiently.

Despite extensive literature on diverse aspects of the electricity sector in the EU and elsewhere, we have found only two studies that include estimates of electricity demand in Europe using panel data. Moreover, these studies only estimated price elasticity in the short-run (ca. 10 years), one covering the period from 1994 to 2004 (Eskeland and Mideksa, 2010) and the other the period from 1990 to 2003 (Azevedo et al., 2011). Most of the other extant literature on electricity demand during the market liberalization period also covers short timeframes. In addition, previous studies focus on electricity demand of residential (household) consumers rather than industrial consumers. Thus, to extend understanding of the demand-side of the electricity market in Europe, thereby facilitating efforts to improve the efficiency of energy services, we identified the following needs. First, to review the empirical literature providing estimates of the price elasticity of electricity demand in panel settings, for both residential and industrial consumers, during liberalization of the EU's electricity sector. Second, to obtain new empirical evidence regarding household and industrial electricity demand in Europe and the associated short- and long-run price and income elasticities. Third, to obtain estimates of these parameters over a sufficiently long time (two decades) to identify other determinants of electricity demand. Finally, to analyze effects of determinants of both residential and industrial electricity demand. Our analysis is based on a dynamic partial adjustment approach to estimate electricity demand, using aggregate panel data for the EU-29 countries from 1995 to 2015, and both GMM (General Method of Moments) and ML (Maximum Likelihood) modeling. Thus, this paper contributes to the literature by providing an analysis of electricity demand in European countries covering a wider and more recent temporal

⁎

Corresponding author. E-mail address: [email protected] (C. Cialani). 1 https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A31996L0092. https://doi.org/10.1016/j.enpol.2018.07.060 Received 2 April 2018; Received in revised form 29 July 2018; Accepted 31 July 2018 0301-4215/ © 2018 Elsevier Ltd. All rights reserved.

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

− 1.27 in the short-run, and between − 0.19 and − 1.06 in the longrun. Regarding the econometric approach, most previous authors have employed either static models or dynamic partial adjustment models. Eskeland and Mideksa (2010) used a static model of residential electricity demand in 30 European countries to study effects of temperature changes on electricity consumption. Azevedo et al. (2011) also estimated residential electricity demand using static models applied to two panel datasets (one covering 15 EU countries from 1990 to 2003 and the other covering US states from 1990 to 2004), yielding short-run price elasticities of − 0.2 and − 0.21 to − 0.25, respectively. More recently, Cebula (2012) estimated residential electricity demand in US states between 2002 and 2005 using a two-stage least squares approach. Results include findings that residential electricity consumption declined with adoption of energy efficiency programs, increases in price, annual cooling degree days and per capita real disposable income. Authors who have used dynamic models for energy demand include Bernstein and Griffin (2006) and Paul et al. (2009), although they did not address the potential dynamic panel bias that arises by including the lag of consumption. Both studies estimate residential electricity demand in the USA. Using data covering 1977–2004, Bernstein and Griffin (2006) obtained short and long-run price elasticities of − 0.24 and − 0.32 respectively. The study by Paul et al. (2009) covered the years 1990–2004, and derived estimated short- and long-run price elasticities of − 0.13 and − 0.40 respectively. They claimed that attempts to account for the lag of consumption (which introduces dynamic panel bias) were unsuccessful and resulted in unstable estimates. Therefore, they only reported least squares dummy variable (LSDV) estimates. However, some recent studies have accounted for dynamic panel bias and used more advanced dynamic panel data models, e.g., panel cointegration, autoregressive distributed-lag (ARDL) or (GMM) estimators. Narayan et al. (2007) used a panel cointegration technique to estimate residential electricity consumption in G7 countries, obtaining group-mean estimates indicating an elastic price effect (− 1.56) and inelastic income (0.245) effect. In contrast, Dergiades and Tsoulfidis (2008), Hung and Huang (2015), and Nakajima (2010) found residential electricity demand to be income inelastic and price elastic. Dergiades and Tsoulfidis (2008) also estimated residential electricity demand in the USA using the ARDL panel cointegration approach and, in contrast to Narayan et al. (2007), detected a significant price effect. They estimated a short-run price elasticity of − 0.39 and long-run income and price elasticities of 0.27 and − 1.07, respectively. Bernstein and Madlener (2015) analyzed residential electricity demand in 18 OECD countries from 1981 to 2008 using panel cointegration and Granger causality testing. They found a short-run price elasticity of − 0.1, and a long-run elasticity of − 0.39. Lower values (− 0.07 and −

period (during the liberalization process) than previous analyses, at aggregate level and using two distinct econometric approaches. Moreover, to the best of our knowledge, no previous study has examined aggregate industrial electricity demand in Europe. Our study is the first attempt to fill this gap. Our analysis is important for formulating future energy policies, determining future energy requirements and investments, and regulating the activities in the electric market. The results of our analysis will also allow us to draw lessons from the liberalization era in the European countries for both residential and industrial categories of consumers. The remainder of this paper is organized as follows. Section 2 presents the literature review, and Section 3 a short overview of the electricity market liberalization process. Data and econometric models are presented in Section 4, while results are discussed in Section 5. Finally, concluding remarks are presented in Section 6. 2. Literature review This section reviews the empirical literature on electricity demand, particularly literature including estimates of price and income elasticities. Some studies focus on a single company or country, while others consider data from several countries. Here, we are mainly interested in studies based on aggregate datasets. There are large variations in estimated short- and long-run price elasticities presented in recent studies, likely due to differences in the time periods covered, and in both the types of datasets (time series vs. panel data) and econometric approaches used. We also briefly discuss some methodologies that have been applied. 2.1. Household electricity demand There is extensive empirical literature on household electricity demand, which is generally estimated by one of two approaches. The first is to use aggregate data, usually including data on price and income variables along with various other factors such as climate and urbanization. Filippini (1999), García-Cerruti (2000), Hondroyiannis (2004), Holtedahl and Joutz (2004) and Narayan and Smyth (2005) use this kind of specification for analyzing residential electricity demand. In the second approach, survey data are used to estimate residential electricity demand and consider effects of potential explanatory variables, such as housing characteristics, use of appliances, and household demographics. Baker et al. (1989), Leth-Petersen (2002), Larsen and Nesbakken (2004), and Filippini, and Pachauri (2004) all use this method. Previous studies on household electricity demand have provided widely varying indications of its responsiveness to price and income. As summarized in Table 1, recent estimates of the price elasticities for household electricity demand vary between − 0.05 and

Table 1 Previously published estimates of short- and long-run price elasticities of household electricity demand obtained from panel data models. Study

Time period

Panel

Price elasticity Short-run

Narayan et al. (2007) Dergiades and Tsoulfidis (2008) Tanishita (2009) Paul et al. (2009) Eskeland and Mideksa (2010) Nakajima and Hamori (2010) Azevedo et al. (2011) Bernstein and Madlener (2011) Alberini and Filippini (2011) Blazquez et al. (2013) Okajima and Okajima (2013) Hung and Huang (2015)

1978–2003 1956–2006 1986–2006 1990–2004 1994–2005 1975–2005 1990–2004 1990–2003 1981–2008 1995–2007 2000–2008 1990–2007 2007:01–2013:12

G-7 US Japan US Europe US US EU-15 OECD US Spain Japan Taiwan

593

− 0.39 0.5–0.9 − 0.13 − 0.2 − 0.14 to − 0.33 − 0.21 to − 0.25 − 0.2 − 0.05 to − 0.06 − 0.08 to − 0.15 − 0.07 − 0.4 − 1.14 to − 1.13 − 0.85 to − 1.27

Long-run − 1.06 − 1.07 1.0–2.07 − 0.40

− 0.39 − 0.44 to − 0.73 − 0.19 − 0.49

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

respectively, confirms the findings of Dilaver and Hunt (2011). In contrast, Jamil and Ahmad (2011) found both price (− 1.22) and income (1.61) effects to be elastic for manufacturing electricity demand in Pakistan.

0.19) were obtained by Blazquez et al. (2013), who applied a FE (Fixed Effect) estimator and the Blundell-Bond GMM estimator to a Spanish panel dataset. Alberini and Filippini (2011) estimated dynamic models of residential electricity demand in the USA and obtained slightly larger short-run (− 0.08 to − 0.15) and long-run (− 0.44 to − 0.73) elasticities. Hung and Huang (2015) estimated dynamic residential electricity demand in 19 Taiwanese counties, and found that price elasticity was higher in summer months than in non-summer months (ranging from − 1.149 to − 1.130 and from − 0.85 to − 1.27, respectively). In contrast, income elasticity was lower in non-summer months than in summer months (ranging from 0.16 to 0.310 and 0.737 to 0.890, respectively). Nakajima (2010) employed panel-dynamic ordinary least squares to estimate residential electricity demand in Japan, obtaining long-run price and income elasticity estimates of − 1.127 and 0.602, respectively.

3. Overview of the EU's energy market A long-term priority of the European Union is to establish and maintain “a competitive single EU electricity market” (European Commission, 2005).2 By means of three legislative packages (1996, 2003, 2009), the EU has gradually opened this sector for competition, aiming develop an internal European electricity market. As already mentioned, this liberalization was intended to reduce prices and increase the sector's efficiency. The first step towards a pan-European liberalized wholesale energy market was taken in 1996 with implementation of EU Directive 96/92/EC, which defined common rules for the generation, transmission and distribution of electricity, with the intention to create an efficient supranational European market. Subsequent electricity market directives (e.g. 2003/54/EC3 and 2009/72/EC4) have also addressed emission targets for the electricity sector and specified paths to integrate renewable energy. Directives adopted in 2003 established common rules for internal markets for electricity, intended to ensure electricity supply to all consumers, full market opening, higher service standards and business efficiency as well as supply security and lower electricity prices. EU directives implemented in 2004 and 2007 stipulated that electricity markets must be fully opened and allow all electricity consumers, both industrial (from the 1rst July 2004) and residential (from the 1rst July 2007), to choose their electricity suppliers, regardless of national boundaries.

2.2. Industrial electricity demand Literature on industry-level electricity demand is very scarce. We have found no studies that provide estimates of industrial consumer's electricity demand at aggregate (country) panel level. Therefore, as summarized in Table 2, we consider here short- and long-run price elasticities for industrial electricity demand obtained from analyses of panel data covering industrial consumers within regions or single countries, and some obtained from analyses of time series data. The reported elasticities (both short- and long-run) range between − 0.002 and − 0.43. The industrial sector's responsiveness to changes in the price of electricity and income (both absolute and relative to the residential sector's responsiveness) is controversial. Woodland (1993), Doms and Dunne (1995) and Bjørner et al. (2001) provided data for the industrial sector but not the residential sector. However, Beenstock et al. (1999) covered both residential and industrial demand, Kamerschen and Porter (2004) analyzed industrial, residential and aggregate demand, and Bose and Shukla (1999) estimated residential, industrial, agricultural and commercial demand. Beenstock et al. (1999) used quarterly data for Israel to compare three dynamic econometric methodologies for analyzing households’ and industrial companies’ demand for electricity: a Dynamic Regression Model and two co-integration approaches (OLS and Maximum Likelihood). Jamil and Ahmad (2010), Dilaver and Hunt (2011), Inglesi-Lotz and Blignaut (2011), and Arisoy and Ozturk (2014) all found that the industrial sector is less responsive than the residential sector to changes in price and income. More specifically, Jamil and Ahmad (2010) found that the manufacturing sector in Pakistan was less responsive to changes in price (− 0.17) and income (0.42) in the longrun. Dilaver and Hunt (2011) reached a similar conclusion from estimates of long-run industry electricity demand in Turkey based on the Structural Time Series Method (STSM; yielding price and income elasticities of − 0.16 and 0.15, respectively). More recent analysis of electricity demand in Turkey by Arisoy and Ozturk (2014), providing estimated long-run income and price elasticities of 0.79 and − 0.014,

4. Empirical framework Energy demand is a function of the quantity of energy available from a given carrier (e.g. electricity) or carriers and economic variables (such as price and income) and other factors (e.g. climatic factors). Demand for energy can arise for numerous reasons. Households consume energy to satisfy various needs and are assumed to allocate their income to energy and other goods that meet competing needs in a manner that maximizes their satisfaction from total expenditure. Industries and commercial users require energy as a production input and aim to minimize total production costs (or maximize cost-effectiveness). Therefore, households and productive users of energy have differing motivations for consuming energy, and any analysis of energy demand should treat them separately. In this section, we present key variables that may influence residential and industrial electricity demand, then briefly discuss selected summary statistics drawn from our dataset, and finally present the econometrics approach applied to analyze the data. 4.1. Data sources and variable measurements

Table 2 Previously published short- and long-run price elasticities of industrial electricity demand. Study

Time period

Panel

Price elasticity Short-run

Beenstock et al. (1999) Bjørner et al. (2001) El-Shazly (2013) Bojnec and Papler (2011)

1962–1994

Israel

1983–1996

Denmark2949 Egypt Slovenia

1982–2010 1993–2010

Data for the analysis were collected from several sources and compiled into a single data file. More specifically, we used panel data provided by Eurostat5 covering 29 countries (see Appendix A), from 1995 to 2015. This sample of countries was selected because Eurostat provides complete national datasets on their energy consumption, and nearly complete datasets on the other variables of interest. To fill gaps

Long-run

− 0.002 to − 0.44 − 0.69 to − 0.21

2

http://ec.europa.eu/competition/sectors/energy/overview_en.html. https://eur-lex.europa.eu/legal-content/en/TXT/?uri= CELEX:32003L0054. 4 https://eur-lex.europa.eu/legal-content/en/ALL/?uri=CELEX %3A32009L0072. 5 For details see http://ec.europa.eu/eurostat/data/database. 3

0.050 − 0.23 to − 0.43

594

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

in the latter datasets, we use data from the World Bank.6 In this section, we first briefly describe the variables and discuss our expectations, based on existing literature, regarding signs of the control variables (which are assumed to be most relevant for this analysis). It should be noted that choosing the optimal price variable as a determinant of electricity demand is not straightforward. Contributions of taxes and levies to overall electricity prices for household and industrial consumers vary substantially among European countries, so we decided to use the prices paid by the consumers, including all taxes and levies. We consider the following variables:

Table 3 Descriptive statistics. Variablea elect

electind

pricehh

– Electricity price for household consumers (pricehh). Following Eurostat, this variable is defined as the “Average national price in Euro per kWh including taxes and levies applicable for the first semester of each year for medium size household consumers (Consumption Band Dc with annual consumption between 2500 and 5000 kWh). Until 2007 the prices are referring to the status on 1st January of each year for medium size consumers (Standard Consumer Dc with annual consumption of 3500 kWh).” – Electricity price for industrial consumers (priceind). Eurostat defines this variable as the “Average national price in Euro per kWh without taxes applicable for the first semester of each year for medium size industrial consumers (Consumption Band Ic with annual consumption between 500 and 2000 MWh). Until 2007 the prices are referring to the status on 1st January of each year for medium size consumers (Standard Consumer Ie with annual consumption of 2000 MWh). We also apply this definition, apart from including taxes and levies. – Electricity consumption by households (electhh). This is the quantity of electricity consumed by households, including all use of electricity (in billion kWh) for space and water heating and all electrical appliances. – Electricity consumption by industry (electind). This is the total energy consumption in all industrial sectors, in billion kWh. – Gross Domestic Product GDP (gdp). This is the value of the total final output of goods and services produced by an economy within a specified temporal period in billions of Euro (real value, base year 2010). – Gross Domestic Product GDP per capita (gdpcap). This indicator, used as a proxy of household income (expressed in euros, real value, base year 2010), is simply calculated by dividing real GDP by the average population in a specific year. – Population (pop). This refers to the population of a Member State, on 31st December in a specified year, as transmitted by that Member State to Eurostat. – Heating degree day (hdd) is a weather-based technical index designed to describe the energy requirements for heating buildings. – Cooling degree day (cdd) is a weather-based technical index designed to describe energy requirements for cooling buildings (air-conditioning).

priceind

gdpcap

gdp

pop

cdd

hdd

Mean

Std.dev.

min

max

Obs.

Overall Between Within Overall Between Within Overall Between Within Overall Between Within Overall Between Within Overall Between Within Overall Between within Overall Between Within Overall Between Within

27.82

38.55 38.97 4.23 50.75 51.38 4.78 0.05 0.04 0.04 0.04 0.02 0.03 16.74 16.74 3.06 4637.43 4666.87 665.23 22.06 22.41 0.87 0.17 0.17 0.04 1.24 1.24 0.22

0.40 0.58 − 2.64 0.40 0.48 16.03 0.04 0.06 0.04 0.02 0.06 0.01 2.09 4.17 6.97 4.89 6.09 − 4505.34 0.38 0.40 13.53 0.00 0.00 − 0.03 0.34 0.50 2.42

163.10 139.30 51.62 239.31 222.92 61.21 0.31 0.24 0.27 0.26 0.15 0.23 84.40 73.23 40.04 29,943.12 25,290.39 6210.17 82.54 81.80 21.38 0.78 0.68 0.30 6.21 5.56 3.74

609 29 21 609 29 21 609 29 21 609 29 21 609 29 21 609 29 21 609 29 21 609 29 21 609 29 21

37.82

0.13

0.09

23.90

1557.43

17.23

0.10

3.00

a The following variables were rescaled as follows: gdp, electhh and electind were divided by one billion; pop by one million and gdpcap, cdd and hdd by 1000. b Within variation – variation over time or given individual (time-variant). Between variation – variation across individual (time-invariant). Overall variation: variation overtime and individuals.

within country variation was lower than the between country variation, as expected given the relatively short time period (1995–2015) considered. Formally, when the series are tested for stationarity the null hypothesis that all panels contain unit roots is rejected.7 The average household electricity consumption during the considered timeframe was about 28 Billion kWh and the average industrial electricity consumption about 37 Billion kWh. For both variables, the between country variation is much larger than the within (over time) variation. The average prices that residential and industrial consumers paid were 0.13 and 0.09 euro per kWh, respectively. For the price variables, the between and within country variations did not differ much. The average gdp per capita was about 24,000 euros, and the between country variation in this variable was much larger than the within country variation. 4.3. Econometric technique

The sign of both cooling and heating degrees-days is expected to be positive, since positive or negative deviations from the predetermined threshold temperatures are associated with needs for cooling and heating, respectively. Using a general climatological approach, the base temperature is set to a constant value of 15 °C for calculating hdd and a constant value of 24 °C for calculating cdd. Cooling or heating largely involve utilization of electric appliances such as air-conditioners or baseboard heaters, thus they both increase electricity consumption.

In this section, we propose a panel data model to explain electricity consumption as a function of several economic variables. We assume that the level of energy consumption in general and electricity in particular depends on price and income not only in the current period but also in the preceding period, i.e. energy demand in the current period is 7 Several so-called unit root tests have been developed for panel data. We applied a Fisher type (Maddala and Wu, 1999; Choi, 2001) unit root test implemented in Stata software, which indicated that the null hypothesis that all panels contain unit roots should be rejected. The output tables from the software can be found in Appendix C. According to Barbieri (2006, p.12) this type of test has several advantages: 1) it does not require a balanced panel, unlike IPS [Im, Pesaran and Shin] tests; 2) it can be carried out for any unit root test derived; 3) and different lag lengths can be used in the individual ADF [Augmented Dickey-Fuller] regressions.

4.2. Descriptive statistics Table 3 presents summary statistics of the variables. Generally, the 6

hh

Categoryb

For details see http://www.devdata.worldbank.org/dataonline. 595

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

in the literature to solve the endogeneity problem posed by the lagged electricity consumption variable. Anderson and Hsiao (1982) proposed a simple instrumental variable estimator, and two estimators based on the general method of moments (GMM) have been proposed by Arellano and Bond (1991) and Blundell and Bond (1998). For this study, we decided to estimate the dynamic demand as expressed in Eq. (5) using the GMM-AB (Arellano-Bond) estimator designed by Arellano and Bond (1991) designed for handling with small T and large N panels. However, the literature does not provide clear indications of appropriate thresholds for T and N. The GMM-AB estimator (using lagged variables as instruments) has been widely used but (as already mentioned) is especially suitable for datasets with small T and large N, i.e. few time periods and many panels. There are alternative GMM estimators for cases where both T and N are large (Alvarez and Arellano, 2003). However, the dataset considered here includes relatively few panels (29 European countries). Moreover, the efficiency of the GMM-AB estimator has been questioned recently. According to Allison et al. (2017) “While the AB approach provides consistent estimators of the coefficients, there is evidence that the estimators are not fully efficient, have considerable small-sample bias, and often perform poorly when the autoregressive parameter (the effect of a variable on itself at a later point in time) is near 1.0” To address these issues, we also apply the Maximum Likelihood method, a less common method to estimate variables from dynamic panel data models, such as electricity demand here.

correlated with its level in the previous period. A common way to model such dynamic relationships is to use a partial adjustment model (Houthakker, 1980; Paul et al., 2009). Here, it is assumed that the desired level of electricity consumption for each country i in period t depends multiplicatively on price and gdp (Gross Domestic Product, used as a proxy for income) according to:

elecit* = α1 (priceit )α2 (gdpit )α3

(1)

It is also assumed that the adjustment from the actual to desired level of electricity consumption takes one period according to:

elecit = (elecit*)θ (elecit − 1)1 − θ

(2)

Inserting the desired level of electricity consumption from 1 into 2 gives:

elecit = [α1 (priceit )α2 (gdpit )α3]θ (elecit − 1)1 − θ

(3)

Taking logarithms:

ln elecit = θ ln(α1) + α2 θ ln(priceit ) + α3 θ ln(gdpit ) + (1 − θ)ln(elecit − 1) (4) Simplifying, adding the two climatic variables hdd (heating degreedays) and cdd (cooling degree-days) and an error term νit to capture effects of unobserved factors, the equation can be rewritten as a simple dynamic panel data model:

ln electit = β1 + β2 ln elecit − 1 + β3 ln priceit + β4 ln gdpit + β5 popit + β6 cddit + β7 hddit + β8 yeart + νit

5. Results and discussion

(5)

Here, β2 = 1 − θ = 0 implies instantaneous adjustment and no dependence of electricity consumption on its lagged value. The short- and β long-run price elasticities are then β3 and 3 respectively. The error

The results of our modeling are presented in Table 4. Since the countries are very different in size the heteroscedasticity robust standard errors are estimated. Most of the parameter estimates are statistically significant and the coefficients generally have the expected signs. In particular, the coefficients for prices are negative while those for GDP and GDP per capita are positive, as predicted by economic theory. As already stated, GDP and GDP per capita are used as determinants of electricity consumption by the industrial sector and households, respectively. This is because we believe that industrial and household consumers are likely to be more strongly influenced by aggregate and per capita income, respectively. However, values of these coefficients are very small and do not differ much between household and industrial consumers. We obtain short-run price elasticity values of − 0.041 and − 0.044 for the residential sector using the GMM-AB and ML approaches, respectively, while corresponding values for the industrial sector are − 0.029 and − 0.052, respectively. Our results suggest that a 1% increase in the price of electricity will, ceteris paribus, result in an approximately 0.03% or 0.05% decline in industrial consumption of electricity (according to the GMM-AB and ML analyses, respectively), and an approximately 0.045% decline in household consumption of electricity. These results indicate that electricity demand is very priceinelastic in the short run (or more strictly was inelastic during the focal period in the EU-29 countries). Moreover, the values are lower than those reported in previous studies (see Table 1), including studies based on European data, e.g., analyses of residential electricity demand by Eskeland and Mideksa (2010), Azevedo et al. (2011). Similarly, industrial price elasticities are lower than values reported by previous authors, such as Bjørner et al. (2001), Bojnec and Papler (2011), who investigated electricity consumption by Danish and Slovenian industries, respectively. However, we cannot compare our results for the full spectrum of industrial consumers considered here with previous reports because no other studies have covered all the European countries. The long-run price elasticities we obtain are also relatively small. Therefore, from an energy policy perspective, our results indicate that price increases would be very blunt instruments to discourage residential and industrial electricity consumption, i.e. the increases may have to be unfeasibly large to have substantial effects.

1 − β2

term, νit , can be decomposed into νit = ui + eit . ui is the panel unit specific error and eit is the overall error that varies over time and panel units. It is assumed that E (ui ) = E (eit ) = E (ui eit ) = 0 . Eq. (5) is used to model electricity consumption at both aggregate household and industry levels. For the latter, the population variables are not included as explanatory variables as we assume that they are irrelevant for energy consumption in the industrial sector. It should also be noted that the proxy for residential electricity demand is GDP per capita while for the industrial sector it is GDP. In some previous studies, corresponding equations have included prices of gas (as a substitute for electricity) as an explanatory variable. We exclude gas prices partly because of lack of data, and partly because of wide variations in the proportional contributions of gas to the energy consumed among European countries for historical and structural reasons (Ranci, 2011). The coefficient β2 captures the impact of past consumption on current consumption of electricity. Consequently, a positive and significant coefficient is consistent with the hypothesis that electricity consumption has a habitual component. Moreover, since electricity consumption and most of the regressors are logarithmic, the coefficients are directly interpretable as demand elasticities. Short- and long-run price elasticities can be obtained from Eq. (5). We expect electricity demand to be less responsive to price changes in the short-run than in the long-run, as the stock of electrical appliances or behavioral habits concerning electricity consumption cannot be changed immediately. Therefore, we do not expect immediate consumption to respond immediately to changing prices, and short-run electricity consumption may deviate from optimal long-run consumption. In our panel data analysis, we specify a model with country-specific fixed effects to account for unobserved heterogeneity. However, since inclusion of a lagged dependent variable in the explanatory variables violates the strict exogeneity assumption, estimation of the dynamic panel data demand as stated in Eq. (5) using a fixed effect is not appropriate. Several instrumental variable estimators have been proposed 596

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

addition, the low price elasticity of electricity demand suggests that electricity is a necessary good, i.e. consumers would continue buying it despite price increases because it is essential for their daily lives. The finding that income elasticities are higher than price elasticities suggests that households and industry are more responsive to income changes than price changes. Moreover, households seem to be more responsive than industrial consumers. While our short-run price elasticities are lower than most values found in literature, the long-run elasticities we obtain are in the middle of the reported range. Nevertheless, estimated elasticities should be compared cautiously since they might depend on the type of model applied (static vs dynamic), type of data used (macro vs micro), stage of economic cycle (expansionary or recessionary), degree of economic development of the countries and econometric techniques used. The coefficients of the lagged variable are significant, carry the expected signs in all models and are used for computation of the longrun elasticities. We obtain long-run price elasticities of − 0.189 and − 0.302 for the residential sector, according to the GMM-AB and ML modeling, respectively, and values of − 0.118 and − 0.198, respectively, for the industrial sector. Corresponding values of long-run income elasticity are 0.862 and 0.913, respectively, for the residential sector and 0.730 and 0.640, respectively, for the industrial sector. The positive sign of the lagged electricity demand variable confirms that electricity demand is influenced by “long-term habit inertia” (Agnolucci, 2010) or a “memory effect” (Gam and Rejeb, 2012), i.e., demand for electricity in a particular period is related to the demand for electricity in the preceding period. These findings show that long-run elasticities are greater than the short-run elasticities as we expected. Interestingly, the GMM-AB approach yielded lower price elasticity for the industrial sector than the ML approach, possibly because of number of variables and timeframe used that could have affected these estimators. The coefficients of the two climate variables, hdd and cdd (indicating demand for electricity for heating and cooling, respectively), show the expected sign but they are not all significant. Elasticities obtained for heating degree-days for the residential sector are 0.034 and 0.033 (according to the GMM- and ML-based analyses, respectively), which are significant at the 5% and 1% probability levels, respectively, while corresponding elasticities for the industrial sector are 0.052 and 0.027, respectively (significant in both cases at the 1% probability level). The values for cooling degree days are higher: e.g. 0.073 for the residential sector according to the GMM-based approach (significant at the 10% level probability level) and 0.176 for the industrial sector according to the ML-based approach (significant at the 5% probability level). The higher sensitivity of electricity demand to cold than to hot weather can be explained by the climatic conditions in Europe, where periods of heating are generally longer than periods when cooling is necessary. The relatively low impact of hdd could also be due to other energy sources being more frequently used than electricity for heating in some European countries. Our results are partially in line with previous research (Blazquez et al., 2013). Finally, we find that the population significantly and positively affects electricity consumption in the residential sector, as expected. Generally, the price elasticities found in this study are lower than those obtained in previous studies based on panel data. Although both the price and income elasticities we obtain are significant, the point estimates are below one, suggesting that electricity consumption is inelastic with respect to price and income, confirming results obtained in other analyses based on aggregate data. It should be noted that we solely relied on aggregate data, and a common drawback could be the accordingly low variation of price and other variables, which may explain the relatively small estimates of elasticities in this study. Moreover, the parameter estimates obtained from the GMM-based approach were similar to those obtained from the ML-based approach, except for the climatic parameters. Our results suggest that reliance on a single method may be unwise and that different approaches can be applied. Overall, we can argue that residential and industrial electricity demand are relatively inelastic in the short

Table 4 Estimated elasticities of electricity demand of the residential and industrial sectors. SR and LR refer to short-run and long-run, respectively. Residentiala

Industriala

GMM-ABb

MLc

ln gdpcapit

0.785*** (0.026) − 0.041** (0.020) 0.186***

0.855*** (0.017) − 0.044*** (0.012) 0.133***

ln popit

(0.048) 0.237***

(0.022) 0.131***

(0.085) 0.073* (0.038) 0.034** (0.014) − 0.002* (0.001) − 0.041** (0.020) − 0.189* (0.101)

(0.048) 0.068 (0.046) 0.033*** (0.010) − 0.001 (0.001) − 0.044*** (0.012) − 0.302*** (0.090)

ln elecit − 1

ln priceit

cddit hddit yeart SR price elasticity LR price elasticity

ln elecit − 1

ln priceit ln gdpit

cddit hddit yeart SR price elasticity LR price elasticity

GMM-ABb

MLc

0.753*** (0.063) − 0.029* (0.016) 0.180***

0.739*** (0.030) − 0.052*** (0.013) 0.167***

(0.036)

(0.032)

0.121 (0.094) 0.052*** (0.018) − 0.002* (0.001) − 0.029* (0.016) − 0.118* (0.061)

0.176** (0.084) 0.027*** (0.010) − 0.001 (0.001) − 0.052*** (0.013) − 0.198*** (0.050)

a Robust standard errors in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01. A set of year dummies were also included but not presented to save space. b The xtabond routine (based on the Arellano-Bond estimator) in Stata was used for GMM-based estimation. c The conditional mixed process (cmp) estimator in Stata (Roodman, 2011) was used for the ML-based estimation. The Arellano-Bond test for serial correlation in the first-differenced residuals, AR(2) test, for the Residential sector (GMM-AB) gives a p-value of 0.054, only marginally higher than the conventional 5% significance level for rejecting the null hypothesis of no autocorrelation. For the Industrial sector the corresponding pvalue is 0.447. The Sargan test of overidentifying restrictions gives p-values of 0.676 and 0.522 for the Residential and Industrial sectors, respectively, indicating that the null hypothesis that overidentifying restrictions are valid should not be rejected. However, the number of instruments exceeds the number of countries, which casts some doubt on the reliability of these test results. Based on the Wald statistic the GMM-AB models are significant (p values of the statistic practically zero) in the sense that the null hypothesis that all the coefficients of the explanatory variables are simultaneously zero can be strongly rejected. However, since we want to compare the GMM-based estimates with ML-based estimates we also use a generalized R-squared goodness of fit measure. This is simply the squared correlation between a dependent variable and its predicted value, also used previously in dynamic panel data modeling (Bloom et al., 2001). We obtain R-squared values of 0.639 and 0.549 for the GMM-AB Residential and Industrial sector models, respectively. The ML models are also significant in the sense that the null hypothesis that all the coefficients of the explanatory variables are simultaneously zero can be strongly rejected (pvalues of the Wald statistic are practically zero). Again, since we want to compare the GMM-AB estimates with ML estimates we also use a generalized Rsquared goodness of fit measure, obtaining R-squared values of 0.924 and 0.686 for the Residential and Industrial sectors, respectively. ML is preferred to GMMAB based on this generalized R-squared criterion. Moreover, the ML-based estimates are generally more statistically significant, indicating that it may be a more efficient approach, as previously discussed. This is also manifested in narrower confidence intervals for the estimated elasticities.

In contrast, we find that the demand for electricity is responsive to income as expressed by gdp and gdpcap. We obtain short-run elasticity values of 0.133 and 0.186 for residential consumers, respectively, according to GMM-AB and ML analyses, and similar values (0.167 and 0.180, respectively) for industrial consumers. Our results corroborate the positive sign and magnitude of elasticity obtained in other studies based on panel data by Paul et al. (2009), Eskeland and Mideksa (2010), Alberini and Filippini (2011), Bernstein and Madlener (2011), Jamil and Ahmad (2011) and Blazquez et al. (2013). Since this elasticity is below unity, income growth apparently results in a less than proportional increase in electricity demand. As the demand for electricity increases when income increases, electricity is a normal good. In 597

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

that electricity consumption changes more rapidly in the industrial sector than in the residential sector. Therefore, an increase in electricity prices will have very little impact on residential and industrial electricity consumption unless it is very large. Additionally, the relatively high long-run income elasticity we obtained indicates that further increases in per capita income of European households will most likely translate into significant, although less than proportional, increases in electrical equipment and hence electricity consumption. Moreover, although there have been substantial improvements in equipment in recent years, there is still margin for European consumers to acquire more and better appliances as their living standards increase. Thus, to limit the growth rate of electricity consumption, policymakers should consider the possibility of introducing higher energy efficiency standards for electrical appliances. Another interesting finding from our study is that electricity demand is more sensitive to cold than to hot weather, as European consumers react more strongly to hdd than cdd, which reflects the climatic conditions in Europe where heating is generally required for longer periods than cooling. Furthermore, the results of the study can be used for national-level forecasting, which is important for planning future investments. A limitation of the study is that the national-level analysis was restricted to the period between 1995 and 2015, due to lack of official data covering other periods. Thus, further analyses covering longer periods, and seasonal changes in household and industrial demand for electricity, would be valuable.

run, probably due to the cost of immediately adjusting the stock of electrical appliances or behavioral habits in the use of electricity in response to a change in price. Conversely, residential electricity demand is more elastic to price changes in the long run, possibly because consumers have more opportunity to adapt their behavioral habits and replace their electrical appliances and equipment over longer timeframes. The slightly higher price-elasticity of industrial demand may be related to other factors, such as competitive pressures to cut costs, including use of electrical energy. The liberalization of the electricity market in Europe does not seem to have affected residential and industrial consumer behavior. 6. Conclusions In this study, we estimate the electricity demand of both residential and industrial consumers in 29 countries during the liberalization of the European electricity market, using aggregate data covering the period 1995–2015, and dynamic partial adjustment modeling with two approaches: GMM-AB and ML. The aim is to contribute to the literature on residential and industrial electricity demand analysis using aggregate data by analyzing the impact of prices on electricity demand over longer timeframes than previous studies. We anticipate that the new estimates of price and income elasticities based on aggregate panel data for both households and industrial consumers will be useful for European policymakers. Our findings indicate that both residential and industrial electricity consumptions are price and income inelastic. However, the residential sector is less sensitive to price changes, in absolute terms, than the industrial sector in both the short and long run. Our results also suggest

Acknowledgements The authors want to thank reviewers for their helpful comments.

Appendix A. List of countries

Austria Belgium Bulgaria Croatia Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Malta Netherlands Norway Poland Portugal Romania Slovakia Slovenia Spain Sweden United Kingdom

598

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

Appendix B. Summary of short- and long-run price and income elasticities with 95% confidence interval in parentheses

Short-run

pricehh priceind

gdpcap

gdp

Long-run

GMM-AB

ML

GMM-AB

ML

−0.041 (−0.080 −0.002) −0.029 (−0.060 0.002) 0.186 (0.091 0.280) 0.180 (0.109 0.251)

−0.044 (−0.067 −0.020) −0.052 (−0.078 −0.026) 0.133 (0.089 0.176) 0.167 (0.104 0.231)

−0.189 (−0.388 0.010) −0.118 (−0.238 0.002) 0.862 (0.397 1.328) 0.730 (0.461 1.000)

−0.302 (−0.48 −0.125) −0.198 (−0.296 −0.100) 0.913 (0.591 1.236) 0.640 (0.440 0.841)

Appendix C. Unit root test specifications based on augmented Dickey-Fuller tests

Ho: All panels contain unit roots Ha: At least one panel is stationary AR parameter: Panel-specific Panel means: Included Time trend: Not included Drift term: Included

Number of panels = 29 Number of periods = 21 Asymptotics: T - > Infinity Cross-sectional means removed ADF regressions: 1 lag

Fisher-type unit-root test for log(electhh) Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared

P Z L* Pm

Statistic 200.8449 − 9.2574 − 9.9268 13.2628

p-value 0.0000 0.0000 0.0000 0.0000

P Z L* Pm

Statistic 178.4299 − 8.5530 − 8.8439 11.1816

p-value 0.0000 0.0000 0.0000 0.0000

P Z L* Pm

Statistic 154.4277 − 7.0943 − 7.3123 8.9531

p-value 0.0000 0.0000 0.0000 0.0000

P Z L* Pm

Statistic 219.0847 − 9.9696 − 11.0162 14.9563

p-value 0.0000 0.0000 0.0000 0.0000

P Z L* Pm

Statistic 113.6905 − 5.0443 − 4.9464 5.1707

p-value 0.0000 0.0000 0.0000 0.0000

P Z L* Pm

Statistic 318.9447 − 13.9583 − 16.3356 24.2281

p-value 0.0000 0.0000 0.0000 0.0000

Fisher-type unit-root test for log(pricehh) Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared Fisher-type unit-root test for log(electind) Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared Fisher-type unit-root test for log(priceind) Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared Fisher-type unit-root test for log(GDP/Cap) Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared Fisher-type unit-root test for cdd Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared

599

Energy Policy 122 (2018) 592–600

C. Cialani, R. Mortazavi

Fisher-type unit-root test for hdd Inverse chi-squared(58) Inverse normal Inverse logit t(149) Modified inv. chi-squared

Statistic 316.4962 − 13.8636 − 16.2110 24.0008

P Z L* Pm

p-value 0.0000 0.0000 0.0000 0.0000

manufacturing technology usage. Technol. Forecast. Soc. Change 49 (3), 297–310. El-Shazly, A., 2013. Electricity demand analysis and forecasting: a panel cointegration approach. Energy Econ. 40, 251–258. Eskeland, G.S., Mideksa, T.K., 2010. Electricity demand in a changing climate. Mitig. Adapt. Strateg. Glob. Change 15 (8), 877–897. Eurostat, 2018. Available at 〈http://ec.europa.eu/eurostat〉. Filippini, M., 1999. Swiss residential demand for electricity. Appl. Econ. Lett. 6 (8), 533–538. Filippini, M., Pachauri, S., 2004. Elasticities of electricity demand in urban Indian households. Energy Policy 32 (3), 429–436. Gam, I., Rejeb, J.B., 2012. Electricity demand in Tunisia. Energy Policy 45 (C), 714–720. García-Cerruti, L.M., 2000. Estimating elasticities of residential energy demand from panel county data using dynamic random variables models with heteroskedastic and correlated error terms. Resour. Energy Econ. 22 (4), 355–366. Holtedahl, P., Joutz, F.L., 2004. Residential electricity demand in Taiwan. Energy Econ. 26 (2), 201–224. Hondroyiannis, G., 2004. Estimating residential demand for electricity in Greece. Energy Econ. 26 (3), 319–334. Houthakker, H.S., 1980. Residential electricity revisited. Energy J. 1 (1), 29–41. Hung, M.-F., Huang, T.-H., 2015. Dynamic demand for residential electricity in Taiwan under seasonality and increasing-block pricing. Energy Econ. 48, 168–177. Inglesi-Lotz, R., Blignaut, J.N., 2011. Estimating the price elasticity of demand for electricity by sector in South Africa. S. Afr. J. Econ. Manag. Sci. 14 (4), 449–465. Jamil, F., Ahmad, E., 2010. The relationship between electricity consumption, electricity prices and GDP in Pakistan. Energy Policy 38 (10), 6016–6025. Kamerschen, D.R., Porter, D.V., 2004. The demand for residential, industrial and total electricity, 1973–1998. Energy Econ. 26 (1), 87–100. Kirschen, D.S., Strbac, G., Cumperayot, P., de Paiva Mendes, D., 2000. Factoring the elasticity of demand in electricity prices. IEEE Trans. Power Syst. 15 (2), 612–617. Larsen, B.M., Nesbakken, R., 2004. Household electricity end-use consumption: results from econometric and engineering models. Energy Econ. 26 (2), 179–200. Leth-Petersen, S., 2002. Micro econometric modelling of household energy use: testing for dependence between demand for electricity and natural gas. Energy J. 23 (4), 57–84. Maddala, G.S., Wu, S., 1999. A comparative study of unit root tests with panel data and a new simple test. Oxf. Bull. Econ. Stat. 61 (0), 631–652. Nakajima, T., 2010. The residential demand for electricity in Japan: an examination using empirical panel analysis technique. J. Asian Econ. 21 (4), 412–420. Narayan, P.K., Smyth, R., 2005. The residential demand for electricity in Australia: an application of the bounds testing approach to cointegration. Energy Policy 33 (4), 467–474. Narayan, P.K., Smyth, R., Prasad, A., 2007. Electricity consumption in G7 countries: a panel cointegration analysis of residential demand elasticities. Energy Policy 35 (9), 4485–4494. Paul, A., Myers, E., Palmer, K., 2009. A partial adjustment model of U.S. electricity demand by region, season and sector, Resources for the Future Discussion Paper. Available at 〈https://papers.ssrn.com/sol3/papers.cfm?Abstract_id=1372228〉. Ranci, P., 2011. Energy Economics (Energia dell’Economia). Il Mulino, Bologna. Roodman, D., 2011. Fitting fully observed recursive mixed-process models with cmp. Stata J. 11 (2), 159–206. Willems, B., Ehlers, E., 2008. Cross-subsidies in the electricity sector. Compét. Regul. Netw. Ind. 9 (3), 201–228.

References Agnolucci, P., 2010. Stochastic trends and technical change: the case of energy consumption in the British industrial and domestic sectors. Energy J. 31 (4), 111–136. Alberini, A., Filippini, M., 2011. Response of residential electricity demand to price: the effect of measurement error. Energy Econ. 33 (5), 889–895. Allison, P.D., Williams, R., Moral-Benito, E., 2017. Maximum likelihood for dynamic panel models with cross-lagged effects. Socius 3, 1–17. Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58 (2), 277–297. Arisoy, I., Ozturk, I., 2014. Estimating industrial and residential electricity demand in Turkey: a time varying parameter approach. Energy 66, 959–964. Azevedo, I.M., Morgan, M.G., Lave, L., 2011. Residential and regional electricity consumption in the U.S. and E.U.: how much will higher prices reduce CO2 emissions? Electr. J. 24 (1), 21–29. Baker, P., Blundell, R., Micklewright, J., 1989. Modelling household energy expenditures using micro-data. Econ. J. 99 (397), 720–738. Barbieri, L., 2006. Panel unit root tests: a review. Ser. Rossa: Econ. - UCSC Piacenza 43, 1–53. Beenstock, K.M., Goldin, E., Nabot, D., 1999. The demand for electricity in Israel. Energy Econ. 21 (2), 168–183. Bernstein, M.A., Griffin, J., 2006. Regional Differences in Price-Elasticity of Demand for Energy. The Rand Corporation Technical Report NREL/SR-620-39512 available at 〈https://www.nrel.gov/docs/fy06osti/39512.pdf〉. Bernstein, R., Madlener, R., 2015. Short-and long-run electricity demand elasticities at the subsectoral level: a cointegration analysis for German manufacturing industries. Energy Econ. 48, 178–187. Bjørner, T.B., Togeby, M., Jensen, H.H., 2001. Industrial companies' demand for electricity: evidence from a micro panel. Energy Econ. 23, 595–617. Blazquez, L., Boogen, N., Fillippini, M., 2013. Residential electricity demand in Spain: new empirical evidence using aggregate data. Energy Econ. 36, 648–657. Bloom, N., Bond, S., Reenen, J.V., 2001. The dynamics of investment under uncertainty. The Institute for Fiscal Studies Working Paper, WP01/05. Available at 〈http://www. ifs.org.uk/wps/wp0105.pdf〉. Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Econ. 87 (1), 115–143. Bojnec, Š., Papler, D., 2011. Households' Perceptions on Electricity Consumption in Slovenia. In: Proceedings of the 8th International Conference, Economic Integration, Competition and Cooperation, 6–9 April, Opatija, University of Rijeka – Faculty of Economics, CD with Full papers. Available at 〈https://ssrn.com/abstract=2232082〉 or 〈http://dx.doi.org/10.2139/ssrn.2232082〉. Bose, R.K., Shukla, M., 1999. Elasticities of electricity demand in India. Energy Policy 27 (3), 137–146. Cebula, R.J., 2012. U.S. residential electricity consumption: the effect of states' pursuit of energy efficiency policies. Appl. Econ. Lett. 19 (15), 1499–1503. Choi, I., 2001. Unit root tests for panel data. J. Int. Money Financ. 20 (2), 249–272. Dergiades, T., Tsoulfidis, L., 2008. Estimating residential demand for electricity in the United States. Energy Econ. 30 (5), 2722–2730. Dilaver, Z., Hunt, L.C., 2011. Industrial electricity demand for Turkey: a structural time series analysis. Energy Econ. 33 (3), 426–436. Doms, M.E., Dunne, T., 1995. Energy intensity, electricity consumption and advanced

600