Residential energy demand for space heating in the Nordic countries: Accounting for interfuel substitution

Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews journal homepage: www.elsevier.com/locate/rser

Residential energy demand for space heating in the Nordic countries: Accounting for interfuel substitution Reza Fazeli a, Brynhildur Davidsdottir a, Jonas Hlynur Hallgrimsson b a b

School of Engineering and Natural Sciences, University of Iceland, Iceland Faculty of Economics, University of Iceland, Iceland

art ic l e i nf o

a b s t r a c t

Article history: Received 11 February 2015 Received in revised form 9 September 2015 Accepted 17 December 2015 Available online 8 January 2016

Reliable estimation of fuel consumption in residential sector is crucial for the future development of fuel supply system. Three approaches are compared in this study to capture climate impact, as well as and interfuel substitution between fuels, on residential fuel consumption for space heating in the Nordic countries. The first approach aims at directly estimating the demand for each energy carrier as a function of explanatory variables including heating degree days, fuels price and GDP per capita. The second approach is a twostage model combing an econometric model for total energy use for space heating, and the market share for energy carriers based on cross/price elasticities, while in the third approach a set of simultaneous equations models were estimated. Based on the results of the mean average error and root-mean-square error criteria for three approaches across the Nordic countries, it was found that the second and third approaches were able to capture the complementary and substitution effects between fuels. This finding confirms our hypothesis that the interfuel substitution is a key factor for estimating changes in fuel demand and should be accounted for when residential energy demand is projected. The results additionally have important implications for climate change policy, by exploring the impact of fuel price on residential energy demand for space heating. & 2015 Elsevier Ltd. All rights reserved.

Keywords: Residential energy demand Space heating Climate impact Interfuel substitution Renewable resources District heating Bioenergy

Contents 1. 2.

3. 4. 5.

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Trends of residential space heating in the Nordic countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Sweden. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Denmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Norway. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4. Finland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1. Sweden. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1. Approach 1: AR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.2. Approach 2: two-stage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.3. Approach 3: simultaneous equations model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.4. Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Denmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1. Approach 1: AR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2. Approach 2: two-stage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.3. Approach 3: simultaneous equations model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.4. Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3. Norway. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1. Approach 1: AR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

E-mail addresses: [email protected] (R. Fazeli), [email protected] (B. Davidsdottir), [email protected] (J.H. Hallgrimsson). http://dx.doi.org/10.1016/j.rser.2015.12.184 1364-0321/& 2015 Elsevier Ltd. All rights reserved.

1211 1214 1214 1214 1214 1215 1215 1217 1217 1217 1218 1218 1218 1218 1219 1219 1219 1219 1219 1220 1220

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

5.3.2. Approach 2: two-stage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.3. Approach 3: simultaneous equations model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.4. Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4. Finland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.1. Approach 1: AR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2. Approach 2: two-stage model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.3. Approach 3: simultaneous equations model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.4. Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix A. : Estimation results from AR and SEM models for estimating the fuel demand in Sweden . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix C. : Estimation results from AR and SEM models for estimating the fuel demand in Denmark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix D. : Estimation results from AR and SEM models for estimating the fuel demand in Norway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Appendix E. : Estimation results from AR and SEM models for estimating the fuel demand in Finland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1. Introduction Buildings represent the largest energy-consuming sector in the economy, with over one-third of all final energy and half of global electricity consumed there. They are also responsible for approximately one-third of global carbon emissions [1,2]. Due to cold climates in the Nordic countries space heating accounts for more than 60% of all energy use in buildings. Based on national statistics datasets [3–6], final energy use for space heating in the Nordic residential buildings increased from 614 TJ in 1990 to 682 PJ in 2013, driven mainly by growth in number of households and appliance ownership. Considering the concerns related to security of energy supply, growing energy demands, limitations of fossil fuels, and threats of disruptive climate changes, diversification and utilization of renewable energy resources are defined as the main strategies in the Nordic countries. In fact, Nordic countries are among the leading countries on successful development of renewable energy and energy efficiency [7]. In 1990, fuel oil, natural gas represented a significant share (43%) of total space heating demand in the residential buildings. However, since the 1990's, rapid development in district and biomass heating systems, resulted in the considerable reduction in the share of fossil fuels to less than 14% in 2010 [8]. One of the key benefits of the district heating system (DHS) is that it offers a considerable opportunity to incorporate different types of energy sources, for example bioenergy, solar energy and industrial waste heat. Nowadays, a well-developed infrastructure for district heating exists in Denmark, Finland and Sweden, while the district heating infrastructure in Norway is limited to large cities. Two explanations account for this different pattern of progress, which are the abundance of hydropower resources and a fairly distributed settlement pattern which slowed the expansion of district heating in Norway [2,9]. Recent studies focused on the developments of using renewable sources for residential energy consumption; progress made by the use of renewable energy in the European Union [10], technological options for large-scale biomass feedstock supply [11], methods to optimize the design and management of biomass-for-bioenergy supply chains [12], development of high solar fraction systems by innovative combination of optimized solar heating, cooling and storage technologies [13] and sustainable management of waste-to-energy facilities [14]. The growing literature on energy demand has offered different dimensions to the evaluation of its dynamics ranging from the choice of specifications to the methodological approaches as well as the underlying factors. Two main purposes of such studies are to identify the main drivers for energy demand and project the energy consumptions, which can be used for energy planning purposes [2,15]. Zhao and Magoules [16] reviewed the methodologies applied for the modeling and prediction of building energy consumption. These

1211

1220 1220 1220 1220 1221 1221 1221 1221 1221 1222 1222 1222 1225 1225 1225 1225

methods include engineering, statistical and artificial intelligence methods. The engineering methods use physical principles to analyze thermal dynamics and energy behavior on a building level or for sublevel components. Building energy simulation models such as CALPAS3 [17], DOE-2 [18], or FEDS and BEAMS [19,20], have been used to analyze the impact of climate change on the demand for energy in individual commercial buildings by Scott et. al., [21] and in groups of commercial and residential buildings in a variety of locations [22– 24]. Statistical regression models are generally micro-econometric studies aiming at estimating energy demand as a function of socioeconomic parameters, which are estimated based on the historical data. Several sub-categories have been identified, including crosssectional analysis [25–31], time-series [32,33], panel database [34,35] and cointegration analysis [36,37]. Artificial Neural Networks (ANN) are the most widely used artificial intelligence models in the application of building energy use predictions. ANN is good at solving non-linear problems and is an effective approach to this complex question. In the past twenty years, researchers have applied ANNs to analyze building energy consumption in a variety of settings, such as heating/cooling load [38,39], electricity consumption [40], optimization and estimation of usage parameters [41]. Undoubtedly, climate change is one of the most pressing concerns facing today society [42]. Most research on climate impact assessments on buildings has evaluated the impacts on total space heating and/or cooling due temperature changes induced by climate change. In general, climate projections are used as exogenous parameters on energy end-use or econometric models. The first studies on this subject date from the late 1980's. In an early study, Barthendu and Cohen [43] estimated the energy demand for heating (winter) and cooling (summer) in 2xCO2 scenarios1 in the region of Ontario, Canada, using regression analysis. Based on the effort by Baum, et al. [44], Table 1 provides a brief overview of 30 years of literature focused on climate change impacts on energy consumption in buildings. This review shows that not many studies tried to capture the climate and fuel substitution effects on fuel demand for space heating at the same time, except [31], where Mansur et al. have used cross-sectional data and a discrete-continuous choice model. They matched detailed climate data to household Energy Information Administration survey data. The climate variables entered into both the fuel choice and conditional consumption equations. The authors found that climate change will result in fuel switching in the United States: warmer summers increase the consumption 1

Scenarios for doubling of atmospheric CO2 concentrations.

1212

Table 1 An overview of the developed models to study the climate change impact on energy demand in buildings – Extended from [44]. Region, Country

Method

Explanatory variables

Finding on change in energy consumption (%)

Barthendu and Cohen [43]

State of Ontario, Canada

Econometric multivariate regression model (Degree-days and others)

Population, number of households, outdoor temperature

Baxter and Calandri [32]

State of California, USA

End-use energy models (heating and cooling of buildings and pumping and transport of water for farms and cities) Re-analysis of building energy consumption in EIA Annual Energy Outlook Estimate the linear, non-linear and non-parametric temperature response functions for a sample of UK households Regression of residential electricity consumption for Hong Kong

Building design, construction and operation, climate variables, such as annual average temperature, humidity and cloud cover

Heating energy:
31 to
45%; Cooling energy: þ 6 to þ 7% (compared to 1976–1983) during 2025–2065 Electricity will increase by about 7500 GWh (2.6%) and 2400 MW (3.7%) by 2010

Rosenthal and Gruenspecht United States [45] Henley and Peirson [46] United Kingdom

Lam [47]

Hong Kong

Considine [48]

United States

Vaage [49]

Norway

Olofsson and Andersson [39] Pardo et al.[50]

Umea, Sweden

Sarak and Satman [51]

Turkey

Amato et al., [19]

State of Massachusetts, USA

Scott et al.,[23]

United States

Mansur et al., [52]

United States

Christenson et al., [53]

Switzerland (four cities)

Fung et al., [54]

Hong Kong

Hadley et al., [55]

Northeastern states, USA

Bigano et al., [56]

OECD and (a few) nonOECD countries

Ruth and Lin [29]

State of Maryland, USA

De Cian et al., [57]

31 countries

Spain

Average price of fuel, square footage of building, degree days Monthly mean temperature, electricity price

Average monthly household income, household size, electricity price, CDD

A decrease of $5.5 billion (1991 dollars) in U.S. energy expenditures in 2010, as a result of 1 °C global warming Estimate the relationship between the temperature and household electricity consumption in UK

The elasticity of annual electricity demand with respect to cooling degree days is estimated to be 0.22, while prices are assumed exogenous. Econometric modeling of aggregate energy Real price, real disposable income, Heating Elasticity of the residential demand for natural gas, heat demand by sectors in U.S. degree-days (HDD), Cooling oil, electricity with respect to HDDs are 0.333, 0.262, and degree-days (CDD) 0.148, respectively. a discrete/continuous choice approach Energy price, household gross income, number of rooms Price elasticity of
1.29 in buildings, age of the building, type of building, climate Climate coefficient -0.294 A neural network model for six single-family indoor and outdoor temperature A correlation of 90–95% with real data when access to the buildings indoor and outdoor temperature difference was assumed. Econometric forecasting of daily electricity load Dummy variables for the days of the week and months, The coefficients for the HDD and CDD are 0.006 and in Spain HDD, CDD 0.005, respectively. Linear estimation of natural gas consumption overall heat transfer coefficient for the building, number Potential gas consumption in Turkey in 2023 could be as for residential heating in Turkey based on HDD of apartments, HDD, the efficiency of the heating system high as 14.92 Gm3 if 100% of residences use natural gas for space heating Econometric multivariate Fuel prices (electricity, natural gas, heating oil), hours of 2.1% and 1.2% increase in daylight, HDD, CDD per capita residential and regression model (Degree-days and others) commercial electricity consumption by 2020 Building models (FEDS and BEAMS) Building stock, temperature A net decrease in energy consumption in U.S. residential and commercial buildings ranging from about 5% in 2020 to as much as 20% in 2080, but with an increase of as much as 25% in temperature-sensitive electricity demand. Econometric analysis of RECS and CBECS Monthly average temperature, fuel price (electricity, fuel
2.8% for electricity only customers;
2% for gas customers; microdata (Discrete-continuous choice model) oil, natural gas, LPG), number of floors, number of
5.7% for fuel oil customers corresponding to a 1 °C household members, size of building, age of building, increase in January temperature in 2050 head householder age Degree-days method: HDD and CDD HDD, CDD, building design parameters HDD:
13 to
87% CDD: up to þ 20 times (2085 scenario) Yearly mean temperature For a 1 °C temperature rise, the electricity consumption Testing four types of functions (linear, quadwould increase by 9.2%, 3.0%, and 2.4% in domestic, radic, cubic and exponential) for energy concommercial and industrial sectors, respectively. sumption in Hong Kong For a low (1.21 °C) and a high (3.4 °C) temperature NEMS energy model, modified for changes in HDD and CDD, fuel prices and availability, floor area, degree-days equipment retail cost and efficiency, appliance saturation response to CO2 doubling, the cumulative cooling/heating demand until 2025 increases by 1.09 quadrillion Btu level (quads) and decreases by 0.82 quads, respectively. A dynamic panel analysis Real GDP, end-user prices and yearly average temperature residential demand responds negatively to temperature increases, while industrial demand is insensitive to temperature increases Econometric multivariate Daylight, fuel price, HDD, CDD Future energy prices and regional population changes regression model (Degree-days and others) may have larger impacts on future energy use than future climate A dynamic panel analysis

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Authors

Household fuel prices, average seasonal temperature levels, GDP per capita Ekici [38]

Elazıg, Turkey

transparency ratio, insulation thicknesses and orientations Energy intensity, price of heating energy and electricity, HDD, CDD

Shorr et al., [58]

13 states in the northeastern, United States

Dolinar et al., [59]

Slovenia (two cities)

Simulation of the indoor conditions and the energy use for heating and cooling

Building architecture and structure properties, Climatological conditions

Pilli-Sihvola et al., [60]

Five countries in Europe

HDD, CDD

Wang et al., [61]

Australia (five cities)

Auffhammer and Aroonruengsaw[62] Dowling [63]

California, USA

Econometric multivariate regression model (Degree-days and others) Software developed by coupling a frequency response building thermal model and a multi-zone ventilation model Panel data-based approach

EU-27

The POLES global energy model

GDP/ capita, energy prices, technology efficiency improvements, HDD, CDD, penetration of ACs, impact of CDD on AC penetration level

Labriet et al., [64]

Worldwide analysis

The techno-economic TIMES-WORLD and the general equilibrium GEMINI-E3 model are coupled with PLASIM-ENTS (a climate model)

Price of residential consumption, the aggregated energy price for residential sector, elasticity of substitution, technology shift population weighted HDD and CDD

Brown et al., [65]

50 states plus DC, USA

Nonlinear model for estimating electricity consumption in residential and commercial buildings

CDD with variable threshold temperature

Zhou et al., [66]

48 US states

Global Change Assessment Model (GCAM)

state level GDP, floor space area, building shell efficiency and surface ratio, satiation impedance per income, population-weighted HDD/CDDs, ‘‘share-weight’’ of technology

local climate and building fabric

Population, household income, mean daily temperature

Total energy consumption for the households may increase by up to 55% by the end of the century The EU27 residential and services sector heating demand is estimated to decrease between 74 Mtoe and 89 Mtoe in 2050, while residential and services sector cooling demand will increases by between 15 Mtoe and 32 Mtoe They estimated the changes in HDDs in 2100 and the variations of household total fossil energy and electricity consumption for several countries, including US and China. Higher exponent factor (1.5) indicates that space cooling is more climate-sensitive than is specified in the National Energy Modeling System (NEMS), which uses an exponent of 1.1 for commercial buildings. The results indicate that heating and cooling buildings energy and fuel use at the state level and that the 48 U.S. contiguous states exhibit a large spatial heterogeneity (ranges from
10% to þ 10% for total,
10% to þ 20% for electricity use and
20% to
5% for oil and gas use in the A2 scenario by the end of the century.

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

An Artificial neural network model for three different building samples Models for heating/cooling expense in 13 states in the northeastern United States

For an increase in summer temperature of 1%, the changes in electricity demand are
0.58%, þ0.54% and þ1.66% for cold, mild and hot countries, respectively. When ANN's outputs are compared with numerical results, average 94.8–98.5% accuracy is achieved By 2030 , total Heating/Cooling energy consumptions in high climate change scenario and with 3 actions (increased AC saturation, shift to lower-cost electric heating, and upgrades) are estimated to be 71.8, 73.9, 73.9, 53.7, MJ per Household, respectively. Heating:
14 to
32% Cooling:
3 to þ 418% in the next 50 years with temperature rise ( þ1 °C and þ 3 °C) During summer, electricity demand will increase 2.5–4% by 2050 compared with 2007 Total heating and cooling requirement:
19 to þ 61% by 2050

1213

1214

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

of electricity and fuel oil consumption, while warmer winters will result in less natural gas consumption for households. Considering the importance of climate and fuel substitution effects on fuel demand in residential sector, this study aims to bridge the knowledge gap by comparing three approaches to simultaneously account for these two effects. Due to the unavailability of detailed information on demographics, heating appliances and building characteristics, three econometric modeling techniques have been tested to estimate fuel demand for space heating in the Nordic countries at the national level. Besides in order to provide insights to the potential impact of fuel pricing in the context of climate and energy policy, the sensitivity of fuel demand to fuel price is studied. The paper first presents the main trends in fuels demand for space heating in residential buildings in the Nordic countries. Then three approaches for estimating residential fuel consumption are briefly described in Section 3. An overview of observed data is done in Section 4. Section 5 contains the estimation and evaluation of results, while the study concludes in Section 6.

2. Trends of residential space heating in the Nordic countries The share of fossil fuel consumption for space heating need in buildings in Sweden, Denmark and Finland was significant in 1990, but it declined rapidly with the development of district heating systems (Fig. 1). In addition to district heating, several supportive policies exist in favor of installing more biomass boilers in residential buildings as a replacement for oil heating systems [67]. The main advantages of wood pellets as one of the most important energy biomass commodity, include lower heating costs, the independence from fossil fuels and the eco-friendly characteristic of the pellets. Based on latest IEA dataset [8], Fig. 1 illustrates the changes in fuel consumption for residential space heating in the Nordic buildings. The replacement of fuel oil with other energy carriers, mainly district heating and biomass, is the common trend in all of the Nordic countries. In Sweden, total space heating decreased partially as district heating systems have higher efficiency compared to oil boilers. In Denmark as in Finland, fuel oil has been replaced by biomass and district heating, while it was substituted by electricity in Norway. 2.1. Sweden Sweden has a well-developed district heating sector, accounting for almost 50% of the heating market. Moreover, the prices of fuel oil and electricity rose relative to the district-heating price, since 1990, while the relative price of biomass was fairly constant.

This is consistent with an increase in the district-heating share since 1990, and a fall in the market shares of oil and electricity. The relationship between the fuel prices relative to district heating and market shares is illustrated in Fig. 2 for the energy carriers used for residential space heating in Sweden. District heating provides heat to about 82% of apartment buildings, and about 66% of commercial buildings. The share of biomass in district heating has increased gradually since the 1970s, from 2% to 62.9% in 2013 [68]. In 2013, biomass, municipal solid waste and peat accounted for 42%, 19% and 3% of district heating systems energy supply in Sweden, respectively [68]. 2.2. Denmark Since 1930s, Denmark has a long tradition in district heating. Denmark has a widespread district heating sector, supplying 63 PJ, representing more than 46% of the residential space heating demand in 2013 [4]. Different from the other Nordic Countries, Denmark is poor in forest resources, with around 490,000 ha covered by forest representing only 10% of land area [69]. The European Environment Agency (EEA) estimated the sustainable biomass potential at 105 PJ, with the major contribution of wastes (92 PJ) and forest residues (8.4 PJ) [70]. The changes in the fuel prices relative to fuel oil and market shares for the energy carriers used for space heating in the residential sector in Denmark, are illustrated in Fig. 3. Since 1990, the prices of biomass and district heating decreased relative to the fuel oil price, while the relative price of electricity was fairly constant. This trend resulted in an increase in the share of biomass and district heating since 1990, and a fall in the market shares of fuel oil and electricity. 2.3. Norway In Norway, electricity is the main source for heating the buildings with 81.2% in 2013, while the share of wood and wood pellets was 15%. Other energy sources including fuel oil and district heating are also used, even though they have only a minor role (district heating (2.3%) and fuel oil (1.6%)). Besides, according to IEA dataset biomass (from firewood and forestry residues) use for heating gradually increased since the 1990s, from approximately 22 PJ to 25 PJ in 2008 [8]. As historical data on the price of biomass for households in Norway is not available, so the relative price of electricity to fuel oil and market shares for electricity and fuel oil are illustrated in Fig. 4.

250 200 150 100 50 0 1990

2008

Sweden Fuel oil

1990

2008

1990

Denmark Natural gas

District heating

2008

Norway Electricity

1990

2008

Finland Biomass

Fig. 1. Space heating supply in the residential buildings across the Nordic countries [8].

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

1215

Relative price ratios 3

Market share 50% 45% 40% 35% 30% 25% 20% 15% 10% 5% 0%

2.5 2 1.5 1 0.5

Fuel oil

Electricity

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

0

Biomass

Fuel oil

Electricity

Biomass

District heating

Fig. 2. Relative price ratio and market share trends for residential space heating in Sweden.

0.3 0.25 0.2 0.15 0.1 0.05

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

0

Natural Gas

District heating

Electricity

Biomass

Fuel market share 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Relative price ratio

Natural Gas Electricity Fuel oil

District heating Biomass

Fig. 3. Relative price ratio and market share trends for residential space heating in Denmark.

As Fig. 4 depicts electricity became cheaper compared to fuel oil, which results in an increase in the share of electricity in residential heating since 2000.

heating was fairly constant. This trend is expected to be one of the motives for the increase in the share of biomass and electricity in residential heating since 2000, compared to fuel oil.

2.4. Finland District heating in Finland has become the most important form of space heating, covering almost 41% of the heating energy in buildings in 2013 and supplied about 66 PJ, of which 19.2 PJ came from renewable sources. More than 90% of the heating requirements of the buildings in large cities in Finland are covered by district heating [71]. In Finland, direct investment support is provided for individual biomass heating installations. [72]. The EEA estimates the Finnish biomass potential to be 402 PJ/year (9.6 Mtoe), of which 71 PJ will come from forestry by 2030 with the main resources from black liquor and wood waste [70]. Recently, subsidies are also available for the advancement to new heating systems which should further enhance the small scale use of pellet stoves in Finland [73]. The results of these polices can be observed in Fig. 5, as oil consumption has been substituted by biomass and district heating for residential space heating. The changes in the price of fuels relative to fuel oil and market shares from1990–2008 are illustrated in Fig. 5 for the energy carriers used for space heating in the Finnish residential sector. Since 2000, the prices of electricity and biomass decreased gradually relative to the oil price, while the relative price of district

3. Methodology This analysis focuses on the demand for energy carriers for residential space heating in the Nordic countries, using econometric approaches to study interfuel substitution effect and climate impacts. To integrate the impact of outdoor temperature in energy demand models, energy analysts often refer to concepts called heatingdegree-days (HDD) and cooling degree-days (CDD) [28,74]. The degree-day methodology presumes a V-shaped temperature–energy consumption relationship, primarily introduced by Jager [74]. As fully described by Erdogdu [36], the preliminary step of an econometric analysis is to run unit root tests on the time series data. In this study, the Augmented Dickey–Fuller (ADF) test is used to test for the presence of unit roots and establish the order of integration of the variables in the model. Based on ADF test, it can be concluded that whether the variable is stationary or not. The stationarity of time series is vital because correlation could persist in nonstationary time series and may result in spurious regression, as showed by Yule [75].

1216

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

0.8

1.8

0.7

1.6

Market share

1.2

0.5

1 0.4

0.8

0.3

0.6

0.2

0.4

0.1

0.2

Relative Price

1.4

0.6

0

0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Fuel oil market share

Electricity market share

Electricity price/fuel oil price

Fig. 4. Relative price ratio and market share trends for residential space heating in Norway.

Market share

Relative price ratios 45% 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

40% 35% 30% 25% 20% 15% 10% 5%

District heating

Electricity

Biomass

2008

2007

2005

2006

2004

2002

District heating Biomass

2003

2001

2000

1999

1998

1997

1995

1996

2008

2007

2006

2005

2004

2002

2003

2001

2000

1998

1999

1997

1996

0%

Electricity Fuel oil

Fig. 5. Relative price ratio and market share trends for residential space heating in Finland.

Three approaches (A1–A3) are compared for fuel demand estimation for residential space heating in the Nordic countries. HDD, GDP per capita and fuels' price are used as explanatory variables. Approach 1: As the first approach, a set of AutoRegressive models with eXogenous variable (ARX) [76] and autoregressive integrated moving average with exogenous variables (ARIMAX) are estimated to calculate the demand for each energy carrier without incorporating the interfuel substitution effect2 and the inputs for these models are HDD, fuels price and GDP per capita:
 1
φi :L :Ef ;c;t ¼ α0;i þ αHDD;i HDDc;t þ αf ;i FuelPricef ;c;t þ αGDPPC;i GDPPC c;t þ ϵt;i

ð1Þ

where, Ef ;c;t is the national demand of fuel f for residential space heating in country c3 in time t, α0;i is the constant term, FuelPricef,c,t is the national average price of fuel f in country c, GDPPCc,t is the GDP per capita in country c, αi and ϵt;i are the associated coefficient for explanatory variables and the error term in time t in equation i, respectively. The models developed using this approach intentionally don't take into account the substitution effects and their results will be used as a benchmark for comparison with the other two approaches. 2 In the case of Denmark and Norway, the moving average process of order one MA(1) was also added. 3 c can take S, D, N or F; S¼ Sweden, D¼ Denmark, N ¼ Norway, F¼ Finland.

Approach 2: A two-stage model combining an econometric model for total energy use for space heating, and the market share parameter is estimated for the demand of for individual fuels. Following the suggestion by Pardo et al., [50], an aggregate model for space heating energy demand is developed for each country, by adding M j;t as monthly dummy variable, and also introducing a one-order autoregressive process AR(1):   1
φj :L :ET;c;t ¼ β0;j þ βHDD;j HDDc;t þ βGDPPC;j GDPPC c;t þ

X12 j¼2

M j;t þ ϵt;j

ð2Þ

where, ET;c;t represents the total energy demand for residential space heating in country c in time t, β0;j is the constant term,βHDD;j is the sensitivity of heating demand to HDD, βGDPPC;j is the sensitivity of heating demand to GDP per capita and ϵt;j is the error term in time t in equation j. The parameters of this model is estimated for the estimation period. Then, in order to identify the specific fuel demand in the validation period, the market share of each fuel is estimated based on the cross/price elasticities of energy carriers. The cross/price elasticity parameter for each fuel is estimated based on the relative prices of the fuels using a least squared error approach suggested by Dean [77]4. In fact, all the fuels are 4 This was done using a linear least squares optimization in Excel. Read Dean [77] for more details of the calculation.

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Table 2 The changes in the average monthly temperature in °C/Decade at the capitals of the Nordic countries.

Table 3 The results of ADF unit root test for the Swedish time series. Variables

Oslo, Norway Summer Winter Year Copenhagen, Denmark Summer Winter Year Stockholm, Sweden Summer Winter Year Helsinki, Finland Summer Winter Year

1900–2010 0.06 0.11 0.10 1900–2010 0.12 0.13 0.14 1900–2010 0.14 0.08 0.12

1925–2010 0.06 0.15 0.11 1925–2010 0.09 0.16 0.13 1925–2010 0.13 0.14 0.15

1950–2010 0.18 0.39 0.25 1950–2010 0.17 0.25 0.19 1950–2010 0.34 0.39 0.32 1959-2010 0.22 0.48 0.31

1975–2010 0.31 0.68 0.45 1975–2010 0.22 0.35 0.29 1975–2010 0.79 0.83 0.69 1975-2010 0.65 0.82 0.53

assumed to have the same ability to switch with each other, which may not be the case in reality. Approach 3: A set of simultaneous equations models with endogenous switching are estimated for the residential fuel consumption based on the GDP per capita, HDD, fuel prices and demand for alternative sources, using the two-stage-least-square (TSLS) procedure. Based on the notation used by Fair [78], the matrix form of the model to be estimated is: AE þ BX ¼ U where U ¼ RU
1 þD where, E is a matrix of endogenous variables representing the demand for each fuel; X is a matrix of exogenous variables; U and D are matrixes of disturbance terms; A, B and R are coefficient matrices. Fair [79] discussed various methods for the estimation of simultaneous equations model with autoregressive residuals. In this study, the efficient estimation method5 introduced by Fair [79], was applied for the SEM with auto-regressive errors. All models have been estimated with monthly residential space heating consumption data in the estimation period, from January 1990 through December 2006. A sample from January 2007 through December 2008 has been used as validation data to assess the predictive power of the estimation models for each country.

1217

ADF test statistics Level

ES HDD Log (HDD) GDPPC Log (GDPPC) Log (fuel oil price) District heating price Electricity price Log (Oil demand) District heating demand Electricity demand Biomass demand


4.08 (90.07%)
3.26 (98.21%)
3.25 (92.22%)
5.13 (99.8%)
3.062 (96.9%)
2.10 (45.85%)
3.44 (95.07%)
1.56 (19.65%)
5.77 (99.9%)
6.52 (99.9%)
6.41 (99.9%)
6.44 (99.9%)

First difference


2.79 (93.9%)
5.23 (99.9%)

Kildsgaard, et al. [82] and Morck [83] are the basis for generating reliable monthly information for space heating demand at the national level for Norway, Finland, Sweden and Denmark, respectively. Considering the lack of disaggregated data, in this study an average response at the national level of each Nordic country was captured without neither taking into account regional differences in temperature within the countries, nor in energy systems (energy sources, efficiencies, distribution systems, characteristics of end users). Data on national average fuel price for residential consumers was collected from national statistical institutions of the Nordic countries7. As it was mentioned earlier, in this study HDDs are used as an explanatory variable in the regression model to represent the difference of the monthly average temperatures with the reference temperature (15.5 °C in this study). The monthly HDD at national level for each of the Nordic countries reported at Eurostat database [84] are used as the basis for our analysis. The data reported by UK MET office [85] was used to study the trends of average monthly temperature at the capitals of the Nordic countries. Table 2 shows the trends of changes in the average monthly temperature at the capitals of the Nordic countries in centigrade degree per decade. The rate of change has increased in the 1975–2010 period across all the capital cities compared to the long term trend.

4. Overview of data 5. Results The need for residential space heating in the Nordic countries is supported mainly by district heating, electricity, fuel oil, natural gas and biomass (firewood and wood pellets). The data used in this analysis are monthly time series on energy demand for space heating per capita, GDP per capita, HDD, and energy price over the period 1990–20086. The data on energy consumption for residential space heating is mainly obtained from the Nordic Energy Technology Perspectives (NETP) Model database developed by International Energy Agency (IEA) [8]. As there was limited access to monthly data for energy demand for space heating, the information from Dokka and Andersen [80], Kalema and Joutsi [81], 5 Fair [79] showed that all of the lagged left- and right-hand side variables must be included in the instrument list to obtain consistent estimates, when the simultaneous equations models with first order serially correlated errors are estimated through two-stage-least-square (TSLS) procedure. It's based on the assumption that the lagged variables will not be correlated with the error term at time t, since they were generated at an earlier point in time. Tests for autocorrelated errors should be conducted; as it affects the exogeneity assumption of the instrumental variables [90]. 6 Currently, the IEA dataset on residential space heating in the Nordic countries is only available over the period of 1990–2008.

In this section, the outcome of applying three approaches described in section four for estimating residential energy demand for space heating in the Nordic countries are discussed. Besides, results of ADF test on time series (stationarity test) are presented. 5.1. Sweden The result of ADF test on data for Sweden are presented in Table 3 The ADF statistics for all the variables except for fuel oil price at logarithm form and electricity price at level form are significant at 90% confidence level, which implies that there is no unit root problem in these variables and the variables are stationary. The ADF statistics for the first difference of the fuel oil price at logarithm form and electricity price at level form are significant at 90% confidence level, which leads to rejection of the null hypothesis. So, these two variables are integrated of order one, I(1). 7

By definition, it is paid end-use price that include all taxes to the consumer.

1218

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Table 4 Cross/price elasticities of energy carriers for residential space heating in Sweden. Fuel oil Fuel oil District heating Electricity Biomass

District heating

Electricity

Biomass

0.18

0.00 0.00

0.04 0.00 0.00

0.03

5.1.1. Approach 1: AR models The results of autoregressive models to estimate the fuel demand for residential space heating in Sweden are presented in Table 16 in Appendix A. Model 1 is estimated for fuel oil demand using the logarithm form of HDD, fuel oil price and GDP per capita8. The demand of district heating is estimated based on the HDD, district heating price and GDP per capita as the explanatory variables, in model 2. Model 3 is estimated for electricity demand based on HDD and first difference of electricity, while model 4 estimates biomass demand based on HDD and GDP per capita. The constant and explanatory variables are all significant, and the positive sign for HDD coefficient match prior expectation. The negative coefficient for GDP per capita in model 1, could be interpreted as a sign of fuel shift with the increase in GDP per capita. The coefficient of district heating price in model 2 is positive indicating that the increase in the price of district heating at current level did not result in reducing the demand, due to the good reputation of district heating as a reliable supply. In order to check for multicollinearity, the Variance Inflation Factors (VIF) was estimated for the explanatory variables in models 1–4. Because the VIF is considerably less than 5 for all variables, according to Marquardt [86], it is reasonable to think that multicollinearity is not present in models 1–4. Based on the value of DW statistic, the null hypothesis of no autocorrelation can be rejected. Because the coefficient of AR(1) is less than one, therefore models 1–4 are stationary. 5.1.2. Approach 2: two-stage model The second approach consists of two stages. First, an econometric model for total energy use for space heating is developed, and then, a market share elasticity parameter for individual fuels is estimated based on the relative price and market share change using a linear least squares model explained fully in Section 3 (Table 4). A cross elasticity of 0 means that the fuels cannot be substituted. An elasticity value of 0.01 means that a 100% increase in the price of fuel B relative to fuel A results in a 1% increase in the demand for fuel A (and an equal decrease in fuel B demand in absolute terms)9. Based on Table 4, the fuel oil was substituted more with district heating than with biomass. Besides, electricity was marginally substituted with district heating. The estimated coefficients for model 5 on total residential space heating demand in Sweden is presented in Table 18, Appendix B. To estimate the fuel demand in the validation period, the outcome of model 5 is combined with cross/price elasticities of energy carriers from Table 4. 5.1.3. Approach 3: simultaneous equations model The results of estimating the simultaneous equations model for the demand of fuel oil, district heating, electricity and biomass using Fair's approach [78] are presented in Table 16 in Appendix A. 8 The first difference of logarithm form of fuel oil price is included because it is integrated of order one (Table 3). 9 Note that the elasticities are defined with respect to the fuels on the vertical axis, so the 0.01 for residential fuel oil/district heating is the percent change in fuel oil demand from a 100% percentage change in the relative price between these fuels (price district heating/price fuel oil).

Table 5 Goodness-of-fit statistics to compare three approaches for residential fuel consumption for space heating at the validation period in Sweden. A1

A2

A3

Fuel oil demand MAE (%) RMSE

119.7 1.43

40.6 0.56

25.8 0.39

District heating demand MAE (%) RMSE

160.4 1.95

58.0 0.75

59.7 0.86

Electricity demand MAE (%) RMSE

185.0 2.21

86.6 1.17

41.1 0.54

Biomass demand MAE (%) RMSE

100.1 1.41

16.3 0.20

39.4 0.51

Table 6 The results of ADF unit root test on the Danish time series. Variables

ADF test statistics Level

ED HDD GDPPCD Fuel oil price Natural gas price District heating price Electricity price Fuel oil demand Natural gas demand District heating demand Electricity demand Biomass demand


29.85 (99.9%)
6.76 (99.9%)
2.32 (82.8%)
2.96 (94.33%)
3.83 (99.4%)
2.37 (84.3%)
1.55 (20.8%)
6.95 (99.9%)
8.04 (99.9%)
12.72 (99.9%)
5.27 (99.9%)
5.34 (99.9%)

First difference


3.63 (95.9%)


5.86 (99.9%)
6.18 (99.9%)

Based on the sign of significant coefficients at 1% level, there is a substitution effect between fuel oil and district heating, as well as fuel oil and biomass. In order to test for the stationarity of the AR model, it's necessary to look at the roots position. Because none of the roots lies outside the unit circle and AR coefficients are less than 1.0, therefore all the AR models are stationary. As discussed earlier, it's important to check for autocorrelation. Considering the value of DW statistics for all four equations, the errors are not autocorrelated. Therefore, the exogeneity of instrumental variables (the lag of dependent and explanatory variables) cannot be rejected. The J-statistics is less than 0.1, which is suitable. 5.1.4. Comparison In this section, the outcomes of three approaches for estimating the demand of each fuel in the validation period (January 2007 through December 2008), are compared in Table 5 in terms of mean average error (MAE) and root-mean-square error (RMSE) criteria. In overall, results from price elasticity based model (second approach) and simultaneous equation model (third approach) fit better than the first approach, confirming the importance of substitution and complementary effects between fuels. The difference between these two models are in how they quantify these effect. Models developed based on second approach only capture the substitution rises from the price elasticity, while the simultaneous equation models take into account the direct switch between fuels. According to Table 5, result from SEM fit better the data

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

compared to the other two approaches for fuel oil demand estimation, which imply that in addition to the relative price, there are other factors that drive the shift from fuel oil to district heating and biomass, such as the good reputation of district heating as a reliable supply for residential heating [87]. The positive coefficients of biomass and electricity in the last two columns of Table 5 imply that people tend to use biomass and electricity as complementary energy carriers for space heating in Sweden. Besides, it's important to highlight that the HDD was a significant factor in model 2 (A2) and in model 8 for electricity demand (A3). 5.2. Denmark Based on the results of ADF statistics for Danish time-series shown in Table 6, the residential space energy demand in Denmark (ED), HDD (HDD) and fuels demand at level form are significant at 90% confidence level, and the variables are stationary. The ADF statistics for the GDP per capita (GDPPCD), heating and electricity prices at level form are insignificant (confidence levelo90%), but the ADF statistics for the first difference of these variables are all significant, which imply that the variables are integrated of order one, I(1). 5.2.1. Approach 1: AR models The results of autoregressive models to estimate the fuel demand for residential space heating in Denmark are presented in Table 19 in Appendix C. Model 10 estimates the fuel oil demand using HDD and oil price and GDP per capita, while HDD and natural gas price are used in model 11 to estimate the natural gas demand. To estimate the district heating demand in Denmark, model 12 used HDD and the first difference of district heating price10 while model 13 estimates electricity demand as a function of HDD and the first difference of electricity price. According to Table 19 in Appendix C, the explanatory variables are all significant, and as expected the sign of the coefficients for HDD and fuel price are positive and negative, respectively. Because the VIF is considerably less than 5 for all variables, according to Marquardt [86], it is reasonable to think that multicollinearity is not present in models 10–14. Based on the value of DW statistic, the null hypothesis of no autocorrelation can be rejected. Because none of the roots lies outside the unit circle, therefore model 10–14 are stationary. 5.2.2. Approach 2: two-stage model The results for estimating the elasticity parameters between fuels for Denmark using the least squares optimization module developed in Excel, are presented in Table 7. Based on the findings, the biomass replaced the fuel oil and natural gas for space heating due to changes in relative prices. The fuel demand in Denmark in the validation period was calculated based on the outcome of the model 15 on total residential heating energy demand and the cross/price elasticities of energy carriers reported in Table 7. 5.2.3. Approach 3: simultaneous equations model Similar to the analysis for Sweden, as the third approach, a simultaneous equations model with endogenous switching is developed to estimate the effects of fuel substitution to support space heating demand in Denmark. The results of estimating the simultaneous equations model for the demand of oil, natural gas, district heating, electricity and biomass using Fair's approach [78] are reposted in Table 20 in Appendix C. Based on the sign of coefficients that are significant at 1% level, there is a substitution effect between fuel oil and district 10 The first difference of district heating price is included because it is integrated of order one, according to Table 6.

1219

Table 7 Cross/price elasticities of energy carriers for residential space heating in Denmark. Fuel oil Fuel oil Natural gas District heating Electricity Biomass

Natural gas

District heating

Electricity

Biomass

0.00

0.00 0.00

0.00 0.00 0.00

0.37 0.26 0.00 0.00

Table 8 Goodness-of-fit statistics to compare three approaches on fuel consumption for residential space heating at the validation period in Denmark. A1

A2

A3

Fuel oil demand MAE (%) RMSE

49.2 0.62

42.1 0.53

50.0 0.59

Natural gas demand MAE (%) RMSE

62.8 0.75

28.6 0.35

58.4 0.67

District heating demand MAE (%) 107.1 RMSE 1.48

9.0 0.12

79.3 1.05

Electricity demand MAE (%) RMSE

18.0 0.20

7.2 0.09

4.6 0.06

Biomass demand MAE (%) RMSE

175.3% 2.09

65.8% 0.71

121.6% 1.61

Table 9 The results of ADF unit root test on the Norwegian time series. Variables

EN Energy demand per capita HDD GDPPC Fuel oil price Electricity price Fuel oil demand Electricity demand Biomass demand

ADF test statistics Level

First difference


0.52 (51.0%)
2.18 (97.26%)
3.66 (99.9%)
2.04 (73.12%)
1.33 (38.84%)
2.97 (84.92%)
1.98 (95.43%)
2.17 (49.67%)
1.5 (48.01%)


8.56 (99.9%)


1.05 (73.84%)
15.26 (99.9%)
6.67 (99.9%)
8.48 (99.9%)
7.72 (99.9%)

heating, as well as fuel oil and biomass. Because none of the roots lies outside the unit circle and the AR coefficients are less than 1.0, therefore all the AR models are stationary. Considering the value of DW for models 17–20, the errors are not autocorrelated in all fuel demand equations. Therefore, the exogeneity of instrumental variables (the lag of dependent and explanatory variables) cannot be rejected. The J-statistics is less than 0.1, which is appropriate. 5.2.4. Comparison In this section, the prediction powers of the developed models for the demand of each fuel are assessed by comparing their output to actual data in the validation period and the MAE and RMSE criteria of the three proposed approaches for each fuel are presented in Table 8. It's obvious that the result from cross/price elasticity model (approach 2) fit better than the other approaches for almost all fuels (SEM result is slightly better for electricity demand). It's interesting because it implies that the fuel substitution was driven by the relative price. This finding could be of a

1220

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Table 10 Summary of AEG test.

Table 12 Goodness-of-fit statistics to compare three approaches for residential fuel consumption for space heating at the validation period in Norway.

Residuals ADF test statistic ADF critical value at three significance levels

Model 22 Model 23


3.25
3.53

1%

5%

10%


4.02
4.43


3.44
3.84


3.14
3.52

Table 11 Cross/price elasticities of energy carriers for residential space heating in Norway. Fuel oil Fuel oil Electricity

Electricity

A1

A2

A2

Fuel oil demand MAE (%) RMSE

46.3 0.52

61.6 0.78

31.2 0.42

Electricity demand MAE (%) RMSE

194.5 2.28

105.7 1.24

144.2 1.94

Biomass demand MAE (%) RMSE

118.8 1.42

16.9 0.31

79.3 0.98

0.71

great interest for decision makers when they are designing energy action plans for residential sector in Denmark. According to Table 18, the HDD is a significant variable; thus, it should also be included in projecting the residential energy demand. 5.3. Norway The results of ADF test on data for Norway are presented in Table 9, which show that the total energy demand (EN), electricity demand, biomass demand and fuels price at level form are insignificant at 90% confidence level. It means that there is a unit root problem in the variables and the variables are nonstationary. Based on the significance of ADF statistics for the first difference of these variables it can be concluded that the variables are integrated of order one, I(1). 5.3.1. Approach 1: AR models The results of estimating the demand of fuel consumption for residential space heating in Norway using the autoregressive approach are presented in Table 21. Model 21 estimates the demand of fuel oil based on HDD and the first difference of oil price, while models 22 and 23 used HDD to estimate the demand for electricity and biomass, respectively. The constant and explanatory variables are all significant, and the sign of the coefficients match the prior expectation. Because the VIF is considerably less than 5 for all variables, it is reasonable to assume that multicollinearity is not present in models 21–23. Based on the value of DW statistic, the null hypothesis of no autocorrelation can be rejected. Because none of the roots lies outside the unit circle, therefore all the ARIMAX models (21–23) are stationary. Considering the nonstationary characteristics of dependent variable in models 22 and 23, and according to Erdogdu [36], the next step in cointegration analysis is to test whether the residual has a unit root problem based on the Augmented Engle–Granger (AEG) test. The result of AEG test is presented in Table 10. It is clear that absolute value of ADF test statistic for both models are significant at 10% and the variables in both models are cointegrated. 5.3.2. Approach 2: two-stage model As the historical data for the price of biomass was not available, the price elasticity parameter was estimated between fuel oil and electricity. Based on our estimation, a 100% increase in the price of fuel oil relative to the price of electricity would result in a substitution of 71% of oil consumption with electricity. Combining the estimation of total residential heating demand (from model 24) and the cross/price elasticities of energy carriers reported in Table 11, the demand for each fuel in Norway in the validation period was calculated.

Table 13 The results of ADF unit root test on the Finnish time series variables. Variables

EF Log EF HDD GDPPC Fuel oil price DH Price Electricity price Biomass price Fuel oil demand District heating demand Electricity demand Biomass demand

ADF test statistics Level

First difference


1.46 (45.16%)
1.55 (49.62%)
11.045 (99.9%)
2.04 (72.87%)
1.83 (31.51%)
1.21 (9.49%) 0.03 (0.4%)
2.62 (72.9%)
1.86 (94.06%)
1.47 (45.53%)
2.09 (45.4%)
2.12 (46.9%)


5.77 (99.9%)
5.807 (99.9%)
0.54 (52.04%)
10.61 (99.9%)
13.50 (99.9%)
8.55 (99.9%)
10.74 (99.9%)
5.76 (99.9%)
6.12 (99.9%)
5.54 (99.9%)

5.3.3. Approach 3: simultaneous equations model Three simultaneous equations models with endogenous switching are developed to estimate the effects of fuel substitution to support space-heating demand in Norway and the results are presented in Table 22. Based on the sign of coefficients that are significant at 1% level, no substitution between oil, electricity and biomass was observed. In fact, the SEMs captured the complementary effect between fuels demand, which is unexpected. Because none of the roots lies outside the unit circle, therefore all the AR models are stationary. The J-statistics is less than 0.1, which is proper. Considering the value of DW statistics for all three models, the null hypothesis of no autocorrelation cannot be rejected for the three models. And therefore, the exogeneity of instrumental variables cannot be rejected. Therefore, the outcome of these models should be interpreted cautiously. 5.3.4. Comparison The outcomes of developed models for the demand of each fuel are compared in the validation period and the results of the MAE and RMSE criteria are accessible in Table 12. Based on these two criteria, none of the developed models provide satisfactory results. But it can be observed that including the substitution effect using the price elasticity parameter improved the forecasting power of the models. It should be emphasized that compared to the studied models for other countries, the models for Norway provide less capability for accurate estimation of fuels demand used for residential space heating. 5.4. Finland Based on the results of ADF statistics for Finnish time-series shown in Table 13, the fuel oil demand and HDD at level form are stationary (significant at 90% confidence level). On the other hand,

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Table 14 Cross/price elasticities of energy carriers for residential space heating in Finland. Fuel oil Fuel oil District heating Electricity Biomass

District heating

Electricity

Biomass

0.30

0.00 0.27

0.28 0.00 0.00

the ADF statistics for the energy demand (EF), fuels price, electricity demand and biomass demand at level form are insignificant even at 10% level of significance, which implies the variables are nonstationary at level. Based on the ADF statistics for the first difference of these variables they are all integrated of order one. 5.4.1. Approach 1: AR models Models 28–31 estimate the demand of fuel consumption for residential space heating in Finland using autoregressive models based on HDD, first difference of fuels' price and GDP per capita and the coefficients of these models are listed in Table 23. As the observation data for the demand for district heating, electricity and biomass were not stationary according to Table 13, the residual and estimated output of these models were tested for unit root problem11. According to Table 23, the constants and HDD are significant, and the sign of the coefficients match the prior expectation. Because the estimated VIFs are considerably less than 5 for all variables, it can be concluded that multicollinearity is not present in models 28–31. Based on the value of DW statistics for all four models except for model 30, the null hypothesis of no autocorrelation can be rejected. Because none of the roots lies outside the unit circle and the AR coefficients are all less than 1.0, therefore all the AR models are stationary. 5.4.2. Approach 2: two-stage model The price elasticity parameters for fuels used for residential space heating in Finland are calculated based on the relative prices of the fuels (Table 14). Based on the findings and due to the relative price, the district heating and biomass replaced the fuel oil, while part of the demand for distract heating is substituted with electricity. Model 32 estimates the total residential space heating demand based on HDD, GDP per capita and monthly dummy variables and the coefficients are reported in Table 18 in Appendix B12. Then, the fuel demand in Finland in the validation period was calculated based on the outcome of model 32 and the cross/price elasticities of energy carriers reported in Table 14. 5.4.3. Approach 3: simultaneous equations model Similar to the analysis for previous case studies, as the third approach, a set of simultaneous equations models with endogenous switching estimate the fuel demand for residential space heating in Finland accounting for the effects of fuel substitution and. the estimated coefficients are listed in Table 24 in Appendix E. Based on the sign of coefficients that are significant at 1% level, there is a substitution between fuel oil, electricity and biomass. Because none of the roots lies outside the unit circle (the AR coefficientso1.0), therefore all the AR models are stationary. 11 The results of ADF test for the residual of models 29–31 are
4.20,
2.76 and
3.03, respectively. Besides, the outcome of ADF test for the output of model 29 are
10.59,
5.97, and
9.75 respectively. Because all of them are significant at 90% confidence level, there is no unit root problem in the variables. 12 As the observation data for total energy demand was not stationary according to Table 18, therefore the residual and estimated output for district heating demand were tested for unit root problem. The ADF test on the residual and estimated output (
11.23,
3.34) shows that both of them are significant at 90% confidence level, which implies that there is no unit root problem in the variables. Moreover, because the coefficient of AR(1) is less than one, therefore AR model for total energy demand is stationary.

1221

Table 15 Goodness-of-fit statistics to compare three approaches for residential fuel consumption for space heating at the validation period in Finland. A1

A2

A3

Fuel oil demand MAE (%) RMSE

54.2 0.64

28.3 0.35

19.0 0.28

District heating demand MAE (%) RMSE

104.3 1.31

28.7 0.38

23.0 0.38

Electricity demand MAE (%) RMSE

54.5 0.69

14.4 0.17

6.5 0.09

Biomass demand MAE (%) RMSE

93.9 1.11

52.3 0.58

11.1 0.19

Considering the value of DW for all three models, the errors are not autocorrelated in all fuel demand equations. Therefore, the exogeneity of instrumental variables (the lag of dependent and explanatory variables) cannot be rejected. The J-statistics is less than 0.1, which is suitable. 5.4.4. Comparison In this section, the outcomes of three approaches for estimating the demand of each fuel are compared in the validation period in terms of the MAE and RMSE criteria (Table 15). It's obvious that the result from SEM fit the data better compared to the other two approaches for all fuels, which signifies the importance of substitution and complementary effects in fuel demand estimation. According to Table 24, the HDD is not a significant variable; but as expected, the sign of the HDD coefficient is positive.

6. Conclusions The major driving forces addressed for residential energy demand for space heating include fuel prices, GDP per capita, outdoor temperature, number of households, floor space per capita and specific energy consumption for residential heating. In this study, we have studied in particular interfuel substitution. Facilitating interfuel substitution is a crucial issue for policy makers, as governments around the world are concerned about local air pollution in addition to GHG emissions and seek to implement policies that are intended to control carbon emissions or steer economies toward or away from certain fuels. Three approaches have been proposed and compared in this study in order to estimate fuel demand for residential space heating in the Nordic countries focusing on the climate impact and interfuel substitution effect driven by changes in fuel prices. In accordance to both MAE and RMSE criteria, it seems that the price elasticity based models (second approach) and simultaneous equation models (third approach) deliver the closest estimation to observation data in the validation period for almost all of the fuels consumed for residential space heating. In fact, these two approaches were able to capture the complementary and substitution effect between fuels. This finding confirms our hypothesis that the interfuel substitution is a key factor to project the fuel demand. We found that substitution between fuels in Denmark, is caused by changes in relative prices, while in other Nordic countries, development and implementation of new district heating systems resulted in a shift from fossil fuels to the consumption of district heating. It was also observed that the HDD is a significant factor in the developed models and apart from Norway, the positive sign of the coefficients match the prior expectation. This

1222

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

conclusion emphasizes the need to include the interfuel substitution effect, when the aim is to project the short-term and long-term fuel demand for residential buildings. Despite the capability of econometric models, these models have several limitations. First of all, econometric analyses are complex and require considerable expertize to implement. Moreover, complexity doesn't always produce better results. They are also data intensive and assume the structure of the system is static. The implicit assumption of perfect competition is questionable even for developed nations [88]. Having said that, econometric models are useful in providing short term forecasts and in assessing the immediate effects of policy decisions, while due to their limitation to capture the structural changes of the system, their results for long term forecast should be interpreted carefully [89]. One specific limitation of this analysis was that the regional differences in temperature within the countries, nor in energy systems (energy sources, efficiencies, distribution systems, characteristics of end users) were not studied here. Despite the limitation of the developed framework, this study primarily enabled us to identify outdoor temperature and interfuel substitution as key factor for projecting fuel demand for residential sector. Our next step is to focus on the impact of building stock characteristics in the Nordic countries. Considering the significant share of old buildings in the Nordic countries, it's expected that the improvement in energy performance of new buildings as well as adoption of new heating and cooling technologies play a key

role in future fuel consumption in the residential sector in the Nordic countries.

Acknowledgments The authors would like to thank two anonymous reviewers for valuable comments on an earlier draft of this article. We wish to thank the Nordic Centre of Excellence for Strategic Adaptation Research (NORD-STAR) for providing funds for this study. The research presented in this paper contributes to the Nordic Centre of Excellence for Strategic Adaptation Research (NORD-STAR), which is funded by the Norden Top-level Research Initiative sub-programme “Effect Studies and Adaptation to Climate Change.”

Appendix A: Estimation results from AR and SEM models for estimating the fuel demand in Sweden See Tables 16 and 17.

Appendix B See Table 18.

Table 16 Summary of statistics and coefficients over the period 1990–2006, for models 1-4 (the t statistics are reported in parenthesis).

ci HDDt FuelPrice GDPPC AR(1) adjustedR2 F DW

Model 1 Fuel oil demand

Model 2 District heating demand

Model 3 Electricity demand

Model 4 Biomass demand

8.719nnn (4.15) 1.237nnn (12.64) 0.543nnn (4.63)
2.53nnn (
5.53) 0.412nnn (6.23) 0.824


7.315nnn (
6.57) 0.0069nnn (6.74)


2.56nnn (
5.76) 0.018nnn (20.69) 0.023nnn (4.30)


3.136nnn (
3.08) 0.009nnn (18.75)

170.72 1.94

0.114nnn (13.76) 0.0001nnn (4.64) 0.202nnn (2.72)

0.404nnn (6.02)

5.47e
5n (1.93) 0.436nnn (6.37)

0.916

0.844

0.832

539.67 1.94

364.46 1.66

328.64 1.52

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

Table 17 Summary of statistics and coefficients over the period 1990–2006, for models 6–9 (the t statistics are reported in parenthesis). Dependent variable

Model 6 Fuel oil demand

Model 7 District heating demand

Model 8 Electricity demand

Model 9 Biomass demand

C Fuel oil demand

0.35 (0.91)

0.540 (0.460)
2.023nnn (
3.83)


0.170 (
1.07) 0.612nnn (16.03)

District heating demand


0.50nnn (
3.88) 1.23nnn (11.19)

0.274 (0.824)
0.585nnn (
3.50)

Electricity demand Biomass demand HDD AR(1) adjustedR2 J-statistic DW

2.517nnn (4.82)

0.959nnn (5.09)


0.002 (
1.23) 0.719nnn (14.21) 0.971


0.003 (
1.17) 0.719nnn (5.74) 0.938

1.11nnn (18.18) 0.001n (2.57) 0.668nnn (13.07) 0.99

0.03 1.70

1.47e-25 1.70

3.9e-25 1.80

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.


0.002 (
1.06) 0.662nnn (11.94) 0.961 2.19e-18 1.75

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

1223

Table 18 Summary of statistics and coefficients over the period 1990–2006, for models 5, 15, 24 and 32 (the t statistics are reported in parenthesis).

βj nnn HDDt GDPPC M nnn 2 M nnn 3 M nnn 4 M nnn 5 M nnn 6 M nnn 7 M nnn 8 M nnn 9 M nnn 10 M nnn 11 M nnn 12 AR(1)nnn MA(1)nnn adjustedR F DW

2

Model 5 Sweden

Model 15 Denmark

Model 24 Norway

Model 32 Finland

43.67 (31.92)

27.73 (60.83)

19.22 (23.18)

2.89 (17.09)

0.0003n (1.45)
1.05e-4 (
0.95)
3.64 (
25.60)
13.16 (
70.22)
33.08 (
125.96)
35.70 (
105.63)
36.07 (
84.50)
40.04 (
84.13)
40.04 (
87.64)
36.52 (
97.10)
29.36 (
106.60)
15.49 (
77.01) 4.45 (31.98) 0.765 (14.46)

0.0011n (1.77) 0.90 (
7.32)
5.03 (
31.39)
15.70 (
71.60)
23.35 (
83.29)
27.91 (
79.94)
27.83 (
70.1)
27.85 (
71.71)
26.30 (
81.22)
16.68 (
69.99)
8.94 (
50.59)
3.70 (
30.62) 0.749 (16.35)


1.45e
5 (
0.99) 9.25e-6n (1.94)
0.04 (
11.40)
0.25 (
47.29)
0.91 (
121.92)
2.12 (
223.34)
2.24 (
185.80)
2.53 (
192.23)
2.24 (
181.01)
2.13 (
204.75)
1.23 (
155.64)
0.63 (
108.74)
0.35 (
89.38) 0.936 (29.59)

0.998

0.998

0.001nnn (2.49)
3.98e-6 (
0.32)
4.27 (
41.64)
10.05 (
70.15)
17.35 (
90.10)
19.31 (
80.43)
19.28 (
64.92)
19.17 (
57.46)
19.20 (
59.08)
19.10 (
70.11)
16.97 (
83.50)
7.73 (
50.71)
0.73 (
7.40) 0.625 (8.48) 0.267 (2.89) 0.998

13,417.35 1.54

94.74.74 1.59

5975.02 1.97

51,370.35 1.90

0.999

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

Table 19 Summary of statistics and coefficients over the period 1990–2006, for models 10-14 (the t statistics are reported in parenthesis). Dependent variable

Model 10 Fuel oil

Model 11 Natural gas

Model 12 District heating

Model 13 Electricity

Model 14 Biomass

ci

4.975 (1.76) 0.005nnn (7.30)
0.597n (
1.48)


1.72n (
1.86) 0.007nnn (21.05) 0.106 (0.87)


3.076nnn (
6.79) 0.018nnn (20.63)
0.051 (
0.38)


0.314nnn (
5.03) 0.002nnn (17.04)
0.001 (
0.13)


3.513n (
1.73) 0.005nnn (15.75)

HDDt FuelPrice GDPPC AR(1) MA(1) adjustedR2 F DW

0.42nnn (4.86)

0.581nnn (8.37)

6.97e-5 (1.48) 0.778nnn (16.28)

0.740nnn (15.00) 0.615nnn (9.24) 0.904

0.597nnn (10.12)

0.899

0.903

0.899

0.880

476.26 1.70

597.09 1.61

403.61 1.67

417.97 1.58

496.46 1.44

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

Table 20 Summary of statistics and coefficients over the period 1990–2006, for models 16-20 (the t statistics are reported in parenthesis). Dependent variable

Model 16 Fuel oil demand

Model 17 Natural Gas

Model 18 District heating demand

Model 19 Electricity demand

Model 20 Biomass demand

C

0.291 (0.31)


1.42nnn (
4.54)

0.272 (0.55)


0.15nnn (
2.77)


2.02nn (
2.49)

Fuel oil demand Electricity demand Biomass demand


3.867nnn (
5.80) 25.88nnn (7.50)

0.152nnn(15.93)

5.20nn (4.88)
0.320 (
1.52)


0.35n (
1.72)

0.157 (0.99)

0.82nnn (13.48) 0.994

0.007nnn (6.35) 0.52nnn (8.54) 0.915

0.76nnn (17.35) 0.985


0.007nnn (
0.27) 0.001nnn (3.02) 0.576nnn (9.78) 0.99

0.001nnn (4.16) 0.685nnn (13.10) 0.71

3.97e-23 0.97

7.29e-25 1.75

4.6e-23 1.51

7.1e-23 1.71

4.69e-24 1.83

HDD AR(1) adjustedR2 J-statistic DW

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

1224

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Table 21 Summary of statistics and coefficients over the period 1990–2006, for models 21-23 (the t statistics are reported in parenthesis). Dependent variable

Model 21 Fuel oil

Model 22 Electricity

Model 23 Biomass

ci


0.241 (
1.46)

HDDt DðOilPriceÞ AR(1) MA(1)

0.003nnn (8.56) 0.001 (0.14) 0.506nnn (6.92) 0.699nnn (12.19) 0.861


0.93nnn (
1.37) 0.010nnn (8..75)


0.394nnn (
1.12) 0.0006nnn (8.59)

0.548nnn (7.52) 0.652nnn (10.7) 0.868

0.570nnn (7.94) 0.654nnn (10.83) 0.867

310.97 1.74

442.64 1.67

441.25 1.64

adjustedR2 F DW

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

Table 22 Summary of statistics and coefficients over the period 1990–2006, for models 25–27 (the t statistics are reported in parenthesis). Dependent variable

Model 25 Fuel oil demand

Model 26 Electricity demand

Model 27 Biomass demand

C Fuel oil demand Electricity demand HDD GDPPC AR(1) Adjusted R2 J-statistic DW

0.078 (0.62)


0.361 (
0.57) 4.63nnn(6.83)


4.08nn (
3.99) 1.84nnn (3.14)

0.71nnn (15.78) 0.963 1.16e-21 1.48

0.002n (1.80) 6.3e-5nnn (3.93) 0.738nnn (12.36) 0.966 6.19e-16 1.11

0.216nnn (6.87)

0.71nnn (15.46) 0.972 2.13e-19 1.48

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

Table 23 Summary of statistics and coefficients over the period 1995–2006, for models 28-31 (the t statistics are reported in parenthesis).

ci HDDt Fuel Price GDPPC AR(1) Adjusted R2 F DW

Model 28 Fuel oil demand

Model 29 District heating demand

Model 30 Electricity demand

Model 31 Biomass demand


0.217 (
1.39) 0.005nnn (18.41) 0.023 (0.61)


3.558nn (
2.03) 0.009nnn (18.55)


1.833nnn (
4.63) 0.004nnn (26.09) 1.23 (1.60) 4.57e-5nnn (3.98)


2.74 (
1.52) 0.006nnn (14.78) 0.25 (1.41)

0.401nnn (4.87) 0.866 279.96 1.61

0.466n (1.77) 9.08e-5n (1.79) 0.40nnn (4.77) 0.865 207.80 1.58

0.834 238.26 1.28

0.418nnn (4.18) 0.862 184.84 1.55

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

Table 24 Summary of statistics and coefficients over the period 1995–2006, for models 33-36 (the t statistics are reported in parenthesis). Dependent variable

C Fuel oil demand District heating demand Electricity demand Biomass demand HDD AR(1) adjustedR2 J-statistic DW

Model 33 Fuel oil demand

Model 34 District heating demand

***

1.68 (3.97)
2.93*** (
2.71)

Model 35 Electricity demand

Model 36 Biomass demand


0.342*** (
2.71) 0.57*** (8.46)

0.02 (0.27)
0.24*** (
2.44) 0.778*** (14.34)


1.49* (
1.87) 2.44*** (5.01)

0.0002 (1.12) 0.80*** (13.40) 0.99

0.745*** (14.21) 0.999

8.17e-5 (1.22) 0.668*** (13.07) 0.99

0.76*** (14.17) 0.999

0.002 1.65

0.04 1.70

0.002 1.65

4.3e-24 1.62

One, two and three stars indicate statistical significance at 10%, 5% and 1% levels, respectively.

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

Appendix C: Estimation results from AR and SEM models for estimating the fuel demand in Denmark See Tables 19 and 20.

Appendix D: Estimation results from AR and SEM models for estimating the fuel demand in Norway See Tables 21 and 22.

Appendix E: Estimation results from AR and SEM models for estimating the fuel demand in Finland See Tables 23 and 24.

References [1] IEA. Transition to Sustainable Buildings-Strategies and oppurtunities to 2050. Paris, France; 2013. doi:10.1787/9789264202955-en. [2] Ürge-Vorsatz D, Cabeza LF, Serrano S, Barreneche C, Petrichenko K. Heating and cooling energy trends and drivers in buildings. Renew Sustain Energy Rev 2015;41:85–98. http://dx.doi.org/10.1016/j.rser.2014.08.039. [3] Swedish Energy Agency. Energy statistics energy commodity balance in 2013; 2014. [4] Danish Energy Agency. Energy statistics 2013; 2014. [5] Statistics Norway. Energy balance for Norway 2013; 2014. [6] Statistics Finland. Consumption of heating energy in residential buildings by year, Energy source and data; 2014. [7] Aslani A, Naaranoja M, Wong K-FV. Strategic analysis of diffusion of renewable energy in the Nordic countries. Renew Sustain Energy Rev 2013;22:497–505. http://dx.doi.org/10.1016/j.rser.2013.01.060. [8] IEA. Nordic energy technology perspectives. Paris, France; 2013. [9] Gebremedhin A, De Oliveira Granheim J. Is there a space for additional renewable energy in the Norwegian power system? Potential for reduced global emission? Renew Sustain Energy Rev 2012;16:1611–5. http://dx.doi. org/10.1016/j.rser.2011.11.032. [10] Scarlat N, Dallemand J-F, Monforti-Ferrario F, Banja M, Motola V. Renewable energy policy framework and bioenergy contribution in the European Union – An overview from National Renewable Energy Action Plans and Progress Reports. Renew Sustain Energy Rev 2015;51:969–85. http://dx.doi.org/ 10.1016/j.rser.2015.06.062. [11] Gabrielle B, Bamière L, Caldes N, De Cara S, Decocq G, Ferchaud F, et al. Paving the way for sustainable bioenergy in Europe: technological options and research avenues for large-scale biomass feedstock supply. Renew Sustain Energy Rev 2014;33:11–25. http://dx.doi.org/10.1016/j.rser.2014.01.050. [12] De Meyer A, Cattrysse D, Rasinmäki J, Van Orshoven J. Methods to optimise the design and management of biomass-for-bioenergy supply chains: a review. Renew Sustain Energy Rev 2014;31:657–70. http://dx.doi.org/10.1016/ j.rser.2013.12.036. [13] Drosou VN, Tsekouras PD, Oikonomou TI, Kosmopoulos PI, Karytsas CS. The HIGH-COMBI project: high solar fraction heating and cooling systems with combination of innovative components and methods. Renew Sustain Energy Rev 2014;29:463–72. http://dx.doi.org/10.1016/j.rser.2013.08.019. [14] Cucchiella F, D'Adamo I, Gastaldi M. Sustainable management of waste-toenergy facilities. Renew Sustain Energy Rev 2014;33:719–28. http://dx.doi.org/ 10.1016/j.rser.2014.02.015. [15] Pérez-Lombard L, Ortiz J, Pout C. A review on buildings energy consumption information. Energy Build 2008;40:394–8. http://dx.doi.org/10.1016/j. enbuild.2007.03.007. [16] Zhao H, Magoulès F. A review on the prediction of building energy consumption. Renew Sustain Energy Rev 2012;16:3586–92. http://dx.doi.org/ 10.1016/j.rser.2012.02.049. [17] Atkinson BA, BC S, WA H, Wilcox BA. National passive solar conference. Annu. Meet.–Am. Sect. Int. Sol. Energy Soc., Portland, OR; 1981. [18] Winkelmann F, Birdsall B, Buhl W, Ellington KL, Erdem A, Hirsch J, et al. DOE-2 supplement version 2.1E, LBL-34947. Berkeley CA; 1993. [19] PNNL. Facility energy decision system user's guide, Release 5.0, PNNL-10542. Richland, WA; 2002. [20] Elliott DB, Anderson D, Belzer DB, Cort KA, Dirks JA, Hostick DJ. Methodological framework for analysis of buildings-related programs GPRA metrics effort. Richland, WA; 2004. [21] Scott MJ, Hadley DL, Wrench LE. Effects of climate change on commercial building energy demand. Energy Sources 1994;16:339–54. [22] Loveland JE, Brown GZ. Impacts of climate change on the energy performance of buildings in the United States. Washington, DC; 1990.

1225

[23] Scott M, Dirks J, Cort K. The adaptive value of Energy Efficiency Programs in a Warmer World. In reducing uncertainty through evaluation 2005. International energy program evaluation conference; 2005 p. 671–682. [24] Huang YJ. The Impact of climate change on the energy use of the U.S. residential and commercial building sectors. Berkeley CA; 2006. [25] Belzer DB, Scott MJ, Sands RD. Climate change impacts on U.S. commercial building energy consumption: an analysis using sample survey data. Energy Sources 1996;18:177–201. [26] Sailor D. Relating residential and commercial sector electricity loads to climate —evaluating state level sensitivities and vulnerabilities. Energy 2001;26:645– 57. [27] Sailor DJ, Pavlova AA. Air conditioning market saturation and long-term response of residential cooling energy demand to climate change. Energy 2003;28:941–51. http://dx.doi.org/10.1016/S0360-5442(03)00033-1. [28] Amato AD, Ruth M, Kirshen P, Horwitz J. Regional energy demand responses to climate change: methodology and application to the commonwealth of Massachusetts. Clim Change 2005;71:175–201. http://dx.doi.org/10.1007/ s10584-005-5931-2. [29] Ruth M, Lin A-C. Regional energy demand and adaptations to climate change: methodology and application to the state of Maryland USA. Energy Policy 2006;34:2820–33. http://dx.doi.org/10.1016/j.enpol.2005.04.016. [30] Franco G, Sanstad AH. Climate change and electricity demand in California, vol. 87; 2006. doi: 10.1007/s10584-007-9364-y. [31] Mansur E, Mendelsohn R, Morrison W. Climate change adaptation: a study of fuel choice and consumption in the US energy sector. J Environ Econ Manag 2008;55:175–93. http://dx.doi.org/10.1016/j.jeem.2007.10.001. [32] Baxter LW, Calandri K. Global warming and electricity demand: a study of California. Energy Policy 1992:20. http://dx.doi.org/10.1016/0301-4215(92) 90081-C. [33] Franco G, Sanstad AH. Climate change and electricity demand in California. Clim Change 2008;87:139–51. http://dx.doi.org/10.1007/s10584-007-9364-y. [34] Deschênes O, Greenstone M. Climate change, mortality and adaptation: evidence from annual fluctuations in weather in the U.S; 2007. [35] Aroonruengsawat A, Auffhammer M. Impacts of climate change on residential electricity consumption: evidence from billing data; 2012. [36] Erdogdu E. Electricity demand analysis using cointegration and ARIMA modelling: a case study of Turkey. Energy Policy 2007;35:1129–46. http://dx.doi. org/10.1016/j.enpol.2006.02.013. [37] Bianco V, Manca O, Nardini S. Electricity consumption forecasting in Italy using linear regression models. Energy 2009;34:1413–21. http://dx.doi.org/ 10.1016/j.energy.2009.06.034. [38] Ekici B, Aksoy UT. Prediction of building energy consumption by using artificial neural networks. Adv Eng Softw 2009;40:356–62. [39] Olofsson T, Andersson S. Long-term energy demand predictions based on short term measured data. Energy Build 2001;33:85–91. [40] González PA, Zamarreño JM. Prediction of hourly energy consumption in buildings based on a feedback artificial neural network. Energy Build 2005;37:595–601. http://dx.doi.org/10.1016/j.enbuild.2004.09.006. [41] Hou Z, Lian Z, Yao Y, Yuan X. Cooling-load prediction by the combination of rough set theory and an artificial neural-network based on data-fusion techhttp://dx.doi.org/10.1016/j. nique. Appl Energy 2006;83:1033–46. apenergy.2005.08.006. [42] Stern NH. The economics of climate change: the Stern review. Cambridge and New York: Cambridge University Press; 2007. [43] Barthendu S, Cohen SJ. Impact of CO2-induced climate change on residential heating and cooling energy requirements in Ontario, Canada. Energy Build 1987;10:99–108. [44] Baum CF, Schaffer ME, Stillman S. Enhanced routines for instrumental variables/GMM estimation and testing; 2007. p.1–38. [45] Rosenthal DH, Gruenspecht HK. Effects of global warming on energy use for space heating and cooling in the United States. Energy 1995;16:77–96. [46] Henley A, Peirson J. Non-linearities in electricity demand and temperature: parametric versus non-parametric nethods. Oxf Bull Econ Stat 1997;59:149– 62. http://dx.doi.org/10.1111/1468-0084.00054. [47] Lam JC. Climatic and economic influences on residential electricity consumption. Energy Convers Manag 1998;39:623–9. [48] Considine TJ. The impacts of weather variations on energy demand and carbon emissions. Resour Energy Econ 2000;22:295–314. [49] Vaage K. Heating technology and energy use: a discrete continuous choice approach to Norwegian household energy demand. Energy Econ 2000;22:649–66. http://dx.doi.org/10.1016/S0140-9883(00)00053-0. [50] Pardo A, Meneu V, Valor E. Temperature and seasonality influences on Spanish electricity load. Energy Econ 2002;24:55–70. http://dx.doi.org/10.1016/ S0140-9883(01)00082-2. [51] Sarak H, Satman A. The degree-day method to estimate the residential heating natural gas consumption in Turkey: a case study. Energy 2003;28:929–39. http://dx.doi.org/10.1016/S0360-5442(03)00035-5. [52] Mansur E, Mendelsohn R, Morrison W. A discrete-continuous choice model of climate change impacts on energy; 2005. [53] Christenson M, Manz H, Gyalistras D. Climate warming impact on degreedays and building energy demand in Switzerland. Energy Convers Manag 2006;47:671–86. [54] Fung W, Lam K, Hung W, Pang S, Lee Y. Impact of urban temperature on energy consumption of Hong Kong. Energy 2006;31:2623–37. http://dx.doi. org/10.1016/j.energy.2005.12.009.

1226

R. Fazeli et al. / Renewable and Sustainable Energy Reviews 57 (2016) 1210–1226

[55] Hadley SW, Erickson DJ, Hernandez JL, Broniak CT, Blasing TJ. Responses of energy use to climate change: a climate modeling study. Geophys Res Lett 2006;33:L17703. http://dx.doi.org/10.1029/2006GL026652. [56] Bigano A, Bosello F, Marano G. Energy demand and temperature : a dynamic panel analysis; 2006. [57] De Cian E, Lanzi E, Roson R. The impacts of climate change on energy demand: a dynamic panel analysis; 2007. [58] Shorr N, Najjar R, Amato a, Graham S. Household heating and cooling energy use in the northeast USA: comparing the effects of climate change with those of purposive behaviors. Clim Res 2009;39:19–30. http://dx.doi.org/10.3354/ cr00782. [59] Dolinar M, Vidrih B, Kajfež-Bogataj L, Medved S. Predicted changes in energy demands for heating and cooling due to climate change. Phys Chem Earth A/B/ C 2010;35:100–6. http://dx.doi.org/10.1016/j.pce.2010.03.003. [60] Pilli-Sihvola K, Aatola P, Ollikainen M, Tuomenvirta H. Climate change and electricity consumption—Witnessing increasing or decreasing use and costs? Energy Policy 2010;38:2409–19. http://dx.doi.org/10.1016/j.enpol.2009.12.033. [61] Wang X, Chen D, Ren Z. Assessment of climate change impact on residential building heating and cooling energy requirement in Australia. Build Environ 2010;45:1663–82. [62] Auffhammer M, Aroonruengsawat A. Simulating the impacts of climate change, prices and population on California's residential electricity consumption. Clim Change 2011;109:191–210. http://dx.doi.org/10.1007/ s10584-011-0299-y. [63] Dowling P. The impact of climate change on the European energy system. Energy Policy 2013;60:406–17. http://dx.doi.org/10.1016/j.enpol.2013.05.093. [64] Labriet M, Joshi SR, Kanadia A, Edwards NR, Holden PB. Impacts of climate change on heating and cooling : a worldwide estimate from energy and macro-economic. SSES Annual Congress 2013; 2013. [65] Brown MA, Cox M, Staver B, Baer P. Climate change and energy demand in buildings Conclusions from the Literature. 2014 ACEEE Summer Study Energy Effic. Build; 2014 p. 26–38. [66] Zhou Y, Clarke L, Eom J, Kyle P, Patel P, Kim SH, et al. Modeling the effect of climate change on U.S. state-level buildings energy demands in an integrated assessment framework. Appl Energy 2014;113:1077–88. http://dx.doi.org/ 10.1016/j.apenergy.2013.08.034. [67] Swedish Energy Agency. Energy in Sweden 2012; 2012. [68] Swedish District Heating Association. Sweden DH statistics〈http://svenskfjarr varme.se/Statistik–Pris/Fjarrvarme/Energitillforsel/〉; 2013. [69] Sander B, Skøtt T. Bioenergy for electricity and heat: experiences from biomass-fired CHP plants in Denmark. Vattenfall; 2007. [70] EEA (European Environment Agency). How much bioenergy can Europe produce without harming the environment? Copenhagen, Denmark; 2006. [71] Finnish Energy Industries. District heating in Finland 2013; 2014. [72] Scarlat N, Dallemand J-F, Skjelhaugen OJ, Asplund D, Nesheim L. An overview of the biomass resource potential of Norway for bioenergy use. Renew Sustain Energy Rev 2011;15:3388–98. http://dx.doi.org/10.1016/j.rser.2011.04.028.

[73] Selkimäki M, Mola-Yudego B, Röser D, Prinz R, Sikanen L. Present and future trends in pellet markets, raw materials, and supply logistics in Sweden and Finland. Renew Sustain Energy Rev 2010;14:3068–75. http://dx.doi.org/ 10.1016/j.rser.2010.06.009. [74] Jager J. Climate and energy systems: a review of their Interactions. New York, NY: John Wiley & Sons; 1983. [75] Yule GU. Why do we sometimes get nonsense-correlations between timeseries?—a study in sampling and the nature of timeseries? J R Stat Soc 1926;89:1–63. [76] Box GEP, Jenkins GM. Time series analysis: forecasting and control. San Francisco: Holden-Day; 1970. [77] Dean M. Review of MED demand models. New Zealand: Wellington; 2011. [78] Fair RC. The estimation of simultaneous equation models with lagged endogenous variables and first oder serially correlated errors. Econometrica 1970;38:507–16. [79] Fair RC. Efficient estimation of simultaneous equations with auto-regressive errors by instrumental variables. Rev Econ Stat 1972;54:444–9. [80] Dokka TH, Andersen G. Marienlyst school – comparison of simulated and measured energy use in a passive house, Paper presented at the Nordic Passive House Conference 21–23. –From low energy buildings towards pluss energy developments, Trondheim Norway; October 2012. [81] Kalema T, Joutsi J. A passive house based on conventional solutions on the market. Conf. Passiv 2012 A; 2012. [82] Kildsgaard I, Widheden A, Jarnehammar A. Measured energy performance and the environmental assessment; 2013. [83] Morck O. Improved indoor climate and efficient energy use in educational buildings; 2012. [84] Eurostat. Heating degree-days by NUTS 2 regions – monthly data, code: nrg_esdgr_m; 2010. [85] UK MET office. 〈http://datamarket.com/data/set/1loo/average-monthly-tem peratures-across-the-world#!ds ¼ 1loo!1n6s ¼1cy.1ft.1mh.25m. 2mk&display¼line&s ¼-jq7〉; 2014 [accessed 13.01.14]. [86] Marquardt D. You should standardize the predictor variables in your regression models. J Am Stat Assoc 1980;75:87–91. [87] Ericsson K. Introduction and development of the Swedish district heating systems – Critical factors and lessons learned; 2009. [88] Craig PP, Gadgil A, Koomey JG. What can history teach us? A retrospective examination of long-term energy forecasts for the United States Annu Rev Energy Environ 2002;27:83–118. http://dx.doi.org/10.1146/annurev. energy.27.122001.083425. [89] Richter C. Some limitations of regional econometric models. Ann Reg Sci 1972;6:28–34. http://dx.doi.org/10.1007/BF01285511. [90] Baum CF. Econometric methods. Instrumental variables and two-stage least squares; 2012. [Chapter 15].