Effects of socio-demographic factors on residential electricity … · 2019. 11. 14. · i Effects of socio-demographic factors on residential electricity consumption in Korea Yoori

저 시-비 리- 경 지 2.0 한민

는 아래 조건 르는 경 에 한하여 게

l 저 물 복제, 포, 전송, 전시, 공연 송할 수 습니다.

다 과 같 조건 라야 합니다:

l 하는, 저 물 나 포 경 , 저 물에 적 된 허락조건 명확하게 나타내어야 합니다.

l 저 터 허가를 면 러한 조건들 적 되지 않습니다.

저 에 른 리는 내 에 하여 향 지 않습니다.

것 허락규약(Legal Code) 해하 쉽게 약한 것 니다.

Disclaimer

저 시. 하는 원저 를 시하여야 합니다.

비 리. 하는 저 물 리 목적 할 수 없습니다.

경 지. 하는 저 물 개 , 형 또는 가공할 수 없습니다.

http://creativecommons.org/licenses/by-nc-nd/2.0/kr/legalcode

http://creativecommons.org/licenses/by-nc-nd/2.0/kr/

Master’s Thesis in Engineering

Effects of socio-demographic factors on

residential electricity consumption in Korea

February 2019

Yoori Kim

Environmental, Energy and Engineering Economics

College of Engineering

Seoul National University



지도교수 허은녕

이 논문을 공학석사 학위논문으로 제출함

2019 년 2 월

서울대학교 대학원

에너지시스템공학부

김유리

김유리의 석사 학위논문을 인준함

2019 년 2 월

위 원 장 (인)

부위원장 (인)

위 원 (인)

i



Yoori Kim

Department of Energy Systems Engineering

The Graduate School of Seoul National University

Abstract

This study conducts empirical analyses on how population aging and other socio-

demographic factors affect residential electricity consumption per capita in 16

municipalities, excluding Sejong City, in the Republic of Korea during the period of 2000-

2016. By doing so, the study proves how each socio-demographic factor affects differently

to the residential electricity consumption in the Republic of Korea, considering its current

trends of fast-growing aging society. The study uses the following components of socio-

demographic factors: the share of population aged 65 and above for population aging, the

number of households, the share of population aged 15~64 for working age population, the

number of households, female to male ratio for gender difference, per capita income,

education level, dwelling types including the ratio of apartments to houses and the ratio of

apartments to multi-houses, and homeownership rate. Also, the effects of electricity price,

ii

heating degree days (HDDs), and cooling degree days (CDDs) variables on the residential

electricity consumption are discussed to produce more thorough empirical results. For

empirical analysis, the study chooses fully modified least squares (FMOLS) and dynamic

least squares (DOLS) estimators with panel cointegration methods and fixed effects and

random effects models for panel data estimations. Using panel data has the benefits of

accounting for the regional differences and time-variant effects, effectively controlling for

the possibly existing multi-collinearity problems.

The following hypotheses are tested: 1) Population aging and residential

electricity consumption exhibit a negative relationship; 2) Increase in household size

reduces residential electricity consumption per capita; 3) Education level, gender difference,

working age population, and homeownership do not have significant effects on residential

electricity consumption; 4) Heating degree days (HDDs) and cooling degree days (CDDs)

have positive relationships with residential electricity consumption.

The study chooses the following statistically significant variables for analysis:

Income, price, population aging, household size, heating degree days, (HDDs) and cooling

degree days (CDDs). The results of the panel cointegration test confirm the presence of a

long-run equilibrium relationships among the variables, and using FMOLS and DOLS

approaches, the cointegrating vectors are estimated. The results of FMOLS and DOLS

indicate that as population aging, electricity price, and household size increase, residential

electricity consumption decreases in the long run; as personal income, heating degree days

(HDDs), and cooling degree days (CDDs) increase, the residential electricity consumption

increase in the long run. The panel estimation results of fixed effects and random effects

model show that as population aging, electricity price, and household size increase,

iii

residential electricity consumption decreases; as personal income, heating degree days

(HDDs), and cooling degree days (CDDs) increase, the residential electricity consumption

increases.

The study makes the following conclusions from the empirical results. First, as

population aging continues to unfold in the Republic of Korea, residential electricity

consumption per capita will likely be reduced; however, this does not mean that population

aging is a desirable condition in the Republic of Korea. Rapid population aging signals the

lack of accessibility to and affordability of energy for the vulnerable elderly households in

the Republic of Korea, possibly leading to the worsened energy equity problems of

especially the vulnerable elderly groups that are short of the access to energy. Further,

population aging in the Republic of Korea will rather have varying effects on the volatility

of residential electricity consumption in the future considering that population aging will

progress at an unprecedented rate. Second, the negative relationship between household

size and residential electricity consumption can be interpreted that as the number of people

per household increases, the frequency of household sharing and using the home appliances

increases. Third, the positive relationship between heating degree days (HDDs) and

residential electricity consumption can be explained by the recent increase in the number

of buildings which operate heating system with electricity or by the increase in the use of

electric auxiliary heating devices attributable to cheap electricity prices of the Republic of

Korea. Fourth, increase in the residential electricity consumption by the increase in cooling

degree days (CDDs) means that more households nowadays own and use air-conditioners

at homes than in the past. Lastly, in terms of the relationship between increasing disposable

income and increasing residential electricity consumption, this can be attributed to the

iv

increased purchasing power of consumers which enables them to purchase and continue to

use the electrical appliances for cooling.

Recently, the Republic of Korea has officially entered the era of an “aged society,”

and it is predicted that the country will get older at a much faster pace. Given that the

policies must be designed and implemented in comprehensive and precise manners in an

aged society, it seems that sticking solely to the price policy to reduce the future residential

electricity consumption possess some critical limitations. Hence, the government must

consider the number of influential factors, including socio-demographic aspects associated

with population aging and temperature effects when designing effective policies that ensure

stabilizations of the residential electricity demand and supply.

Keywords: Socio-demographic factors, population aging, residential electricity

consumption, cooling degree days, heating degree days, panel cointegration, panel data

analysis

Student Number: 2017-22870

v

Contents

Abstract ................................................................................................................................ i

Contents .............................................................................................................................. v

List of Tables ..................................................................................................................... vii

List of Figures .................................................................................................................... ix

Chapter 1. Introduction ................................................................................................. 11

1.1 Research background and research objective ................................................. 11

1.2 Research hypotheses ....................................................................................... 18

1.2.1 Hypothesis I: Population aging and residential electricity consumption exhibit a

negative relationship ....................................................................................................... 18

1.2.2 Hypothesis II: Increase in household size reduces residential electricity

consumption per capita ................................................................................................... 19

1.2.3 Hypothesis III: Education level, gender difference, working age population, and

homeownership do not have significant effects on residential electricity consumption . 20

1.2.4 Hypothesis IV: Heating degree days (HDDs) and cooling degree days (CDDs)

have positive relationships with residential electricity consumption .............................. 21

1.3 Research design .............................................................................................. 23

Chapter 2. Overview of population aging and residential electricity consumption ...... 25

2.1 Overview of population aging in the world and Korea ................................... 25

2.2 Overview of residential electricity demand in Korea ..................................... 34

2.3 Comparison with Germany and Sweden on the effects of population aging on

residential energy consumption ................................................................................. 44

Chapter 3. Literature Review ........................................................................................ 47

3.1 Theoretical studies .......................................................................................... 47

3.1.1 Houthakker (1951) and subsequent studies on the demand for electricity ...... 47

3.1.2 Residential electricity demand model considering socio-demographic

characteristics .................................................................................................................. 51

3.2 Population aging ............................................................................................. 59

3.3 Household size ................................................................................................ 75

vi

3.4 Education, gender, homeownership, working age population, and dwelling type

77

3.5 Heating degree days (HDDs) and cooling degree days (CDDs) ..................... 79

3.6 Studies using panel estimation methods ......................................................... 87

Chapter 4. Data and research methods .......................................................................... 90

4.1 Data ................................................................................................................. 90

4.2 Methodology ................................................................................................. 121

4.2.1 Panel cointegration approach ........................................................................ 122

4.2.2 Tests for fixed effects and random effects models ........................................ 129

Chapter 5. Empirical results ........................................................................................ 135

5.1 Panel cointegration results ............................................................................ 135

5.2 Fully modified ordinary least squares (FMOLS) and dynamic least squares

(DOLS) results ......................................................................................................... 139

5.3 Results for tests of hypotheses with panel data ............................................. 144

5.4 Fixed effects and random effects results ....................................................... 150

Chapter 6. Conclusions ............................................................................................... 159

6.1 Conclusions ................................................................................................... 159

Bibliography.................................................................................................................... 163

Appendix ......................................................................................................................... 173

Trends of all data for 16 municipalities in Korea during 2000-2016 presented in Figure

A.1 ~ Figure A.34. ................................................................................................... 173

Abstract (Korean) .......................................................................................................... 211

vii

List of Tables

Table 2.1. Compound annual growth rate (CAGR) of final energy consumption by energy

source in Korea (1990-2016) ............................................................................................ 38

Table 3.1. Summary of the literature studies for Section 3.1 ......................................... 50

Table 3.2. Summary of literature on the relationship between population aging and

residential electricity consumption ................................................................................. 73

Table 3.3. Summary of literature on the relationship of heating degree days (HDDs) and

cooling degree days (CDDs) on residential electricity consumption ................................ 86

Table 4.1. Definition of the variables (16 municipalities in Korea excluding Sejong City,

2000-2016) ...................................................................................................................... 117

Table 4.2. Descriptive statistics of the data .................................................................. 118

Table 4.3. Results from Pearson’s correlation analysis ................................................ 119

Table 4.4. Compounded annual growth rate (CAGR) of the variables from 2000 to 2016

by regions ........................................................................................................................ 120

Table 5.1. Results of Pesaran’s (2004, 2015) cross-sectional dependence tests .......... 136

Table 5.2. Results of Pesaran’s (2007) panel unit root tests......................................... 137

Table 5.3. Pedroni’s (1999, 2004) and Kao’s (1999) panel cointegration tests for

individual models without the deterministic trends ........................................................ 138

Table 5.4. Results from the estimations with Fully Modified OLS (FMOLS)............. 142

Table 5.5. Results from the estimations with Dynamic Least Squares (DOLS) .......... 143

Table 5.6. Results of the Breusch-Pagan Lagrange Multiplier (LM) test for random

effects .............................................................................................................................. 145

Table 5.7. Results of Wald test for fixed effects ........................................................... 146

Table 5.8. Results of Hausman test for fixed effects and random effects models ........ 147

Table 5.9. Results of modified Wald test for heteroscedasticity .................................. 148

Table 5.10. Results of Wooldridge test for autocorrelation ...................................... 149

Table 5.11. Results from the estimations with fixed effects model .............................. 155

Table 5.12. Results from the estimations with random effects model.......................... 156

viii

Table 5.13. Results from the estimations with LSDV model by region (Reference region:

Seoul) .............................................................................................................................. 157

Table 5.14. Confirmation of the research hypotheses with the estimated results ......... 158

ix

List of Figures

Figure 1.1. Conceptualization of the socio-demographic factors influencing residential

energy consumption .......................................................................................................... 14

Figure 1.2. Conceptual framework of the research ........................................................ 24

Figure 2.1. Comparison of the rate at which aged society becomes super-aged society in

major countries .................................................................................................................. 28

Figure 2.2. Comparisons between the population pyramids of the world and Korea in

1950 ................................................................................................................................... 29


2000 ................................................................................................................................... 30


2050 ................................................................................................................................... 31

Figure 2.5. Annual trends of final energy consumption by energy source in Korea, 1990-

2016 ................................................................................................................................... 39

Figure 2.6. Annual trend of final energy consumption structures by energy sources in

Korea, 1990-2016 .............................................................................................................. 40

Figure 2.7. Annual trend of final electricity consumption in Korea, 1981-2016 ........... 41

Figure 2.8. Annual trend of electricity consumption in residential sector in Korea, 1993-

2016 ................................................................................................................................... 42

Figure 2.9. Monthly trend of electricity consumption in residential sector in Korea,

1999:01-2015:01 ............................................................................................................... 43

Figure 2.10. Energy consumption for housing by gender and age differences in Germany

........................................................................................................................................... 45

Figure 2.11. Energy consumption for housing by gender and age differences in Sweden

........................................................................................................................................... 46

Figure 3.1. The framework of application of household production theory on residential

electricity considering socio-demographic characteristics ................................................ 52

Figure 3.2. Steps to derive the demand for residential electricity consumption in Deaton

x

and Muellbauer (1980) ...................................................................................................... 55

Figure 4.1. Trends of residential electricity consumption per capita (MWh/person) of 16

municipalities in Korea, 2000-2016 .................................................................................. 91

Figure 4.2. Trends of residential electricity price (KRW/kWh) of 16 municipalities in

Korea, 2000-2016 .............................................................................................................. 94

Figure 4.3. Trends of personal income per capita (1000 KRW) of 16 municipalities in

Korea, 2000-2016 .............................................................................................................. 96

Figure 4.4. Trends of the share of population aged 65 and above (%) of 16 municipalities

in Korea, 2000-2016 ......................................................................................................... 98

Figure 4.5. Trends of working age population (%) of 16 municipalities in Korea, 2000-

2016 ................................................................................................................................. 100

Figure 4.6. Trends of household size of 16 municipalities in Korea, 2000-2016 ...... 102

Figure 4.7. Trends of the share of population with degrees in higher education (%) of 16

municipalities in Korea, 2000-2016 ................................................................................ 104

Figure 4.8. Trends of heating degree days (HDDs) of 16 municipalities in Korea, 2000-

2016 ............................................................................................................................... 106

Figure 4.9. Trends of cooling degree days (CDDs) of 16 municipalities in Korea, 2000-

2016 ............................................................................................................................... 108

Figure 4.10. Trends of attached house to apartment ratio (%) of 16 municipalities in

Korea, 2000-2016 .......................................................................................................... 110

Figure 4.11. Trends of detached house to apartment ratio (%) of 16 municipalities in

Korea, 2000-2016 ............................................................................................................ 112

Figure 4.12. Trends of female to male ratio (%) of 16 municipalities in Korea, 2000-2016

......................................................................................................................................... 114

Figure 4.13. Trends of homeownership ratio (%) of 16 municipalities in Korea, 2000-

2016 ................................................................................................................................. 116

11

Chapter 1. Introduction

1.1 Research background and research objective

Research Background

Since Thompson (1929) first proposed the concept of demographic transition,

scholars have begun to illustrate how demographic transition provides grounds for

threatening the sustainability in the world. For instance, rapid population growth

throughout the world, which is noticeable in developing countries, has driven the energy

demand and exerted pressures on natural resources, endangering socio-economic

sustainability. The problems caused by population surge have further been compounded by

population aging, with the trend being more pronounced in developed countries; in fact,

population aging has now become a serious issue affecting the policy agenda of the entire

world. Considered as one of the fastest-aging countries in the world, South Korea (referred

to as “Korea” hereinafter) is expected to have far more threatening effects of population

aging on the society in contrast to the other advanced countries. As of 2017, Korea has

officially become “aged society,” and US Bureau of Statistics forecast that it will only take

27 years for Korea to become a super-aged society from an aging society at the fastest pace

in the world.

It is undeniable that the increased life expectancy dedicated to improved medical

12

system, nutrition, sanitation, enhanced socio-economic well-being, and better access to

education has led to the enriched quality of life for all. Notestein (1954), Kirkwood (2001),

and Cowgill (1970) have even acknowledged that population aging denotes a sign of

progress and socio-economic success, touting the trend as “a great triumph of civilization,”

“the greatest triumph that our species has achieved,” and “one of the truest measures of

progress.” However in the late 1900s and early 2000s, scholars have started to consider that

population aging will present a daunting burden on the socio–economic endurance for it

results in dwindling labor force participation (O’Neill et al., 2010), arouses fears among

national governments and international agencies about the tremendous expenditure costs

associated with an aging society (Walker, 1990), and transfers the financial obligations to

future younger generations, especially in the welfare states (Judt, 2010).

With these shifting viewpoints, academia has recently been focusing on how

population aging and other socio-demographic matters affect the structural changes in

energy consumption behaviors (Shin, 2018). O’Neill and Chen (2002), Liddle (2004), and

Prskawetz et al. (2004) argue that socio-demographic factors accompanied by population

aging significantly affect the changes in volatility of energy consumptions in transport and

residential sectors, especially those regarding the residential electricity consumption.

Further, not only the roles of population aging but the roles other socio-demographic factors

on residential electricity consumption have led to heated debates in the academic fields

(Liddle, 2004; Brounen et al., 2012; Hong et al., 2018).

Given such wide range of academic opinions, the scholars have yet to reach a

consensus as to whether population aging has negative or positive effects on the residential

electricity consumption. Moreover, relevant studies are confined to conceptual and

13

qualitative analyses which lack empirical frameworks. In order to correctly interpret the

impacts of population aging on the residential electricity consumption and develop suitable

policies to address the issues caused by population aging, empirical research with concrete

theoretical backgrounds must be conducted using appropriate data and methods. Further,

previous studies with different viewpoints must be organized and classified in a comparable

manner to yield to a comprehensive and satisfying analysis.

14

Figure 1.1. Conceptualization of the socio-demographic factors influencing residential

energy consumption

Source: Reorganized Figure 1 in Frederiks et al. (2015)

15

As mentioned earlier, the roles of socio-demographic factors and population aging

combined have attracted attention in the academic fields (Liddle, 2004; Brounen et al.,

2012; Hong et al., 2018); further, scholars have started to address and examine the effects

of socio-demographic factors on the residential energy consumption and have confirmed

that these factors individually produce different effects on energy consumption in

residential sectors. Figure 1.1 presents the conceptualization of the various classified socio-

demographic factors influencing residential energy consumption, reorganized from

Frederiks et al. (2015). As seen above, the socio-demographic factors that affect the

residential energy consumption vary in number, and these can be categorized into numerous

predictors including population aging, working age population, education, homeownership,

dwelling type (the type of living quarters in which a person resides), gender, income, and

so forth. Referring to this framework, we primarily focus on the impacts of these socio-

demographic factors on the residential electricity consumption in Korea.

16

Research Objective

This study seeks to empirically analyze how the arising phenomena of population

aging and other socio-demographic factors affect residential electricity consumption in 16

municipalities, excluding Sejong City, in Korea during the period of 2000-2016.

Residential electricity consumption is chosen as the scope to be analyzed since, along with

the energy consumption in the transportation sector, residential energy consumption is one

of the main areas known to be affected by the phenomena of population aging and other

socio-demographic factors (O’neill and Chen, 2002; Liddle, 2004; Prskawetz et al., 2004).

Further, the study focuses on examining the key predictors of socio-demographic factors,

along with population aging, which have influence on residential electricity consumption

in Korea. In specific, the study refers to the framework developed by Frederiks et al. (2015),

which classify a number of socio-demographic factors influencing the residential energy

consumption into several predictors. Assuming that these predictors pose some influence

on the residential electricity consumption in Korea, the study uses the following

components of socio-demographic factors for empirical analysis: the share of population

aged 65 and above, the share of population aged 15~64 for working age population, average

household size, female to male ratio, per capita income, education level, dwelling types

including the ratio of apartments to houses and the ratio of apartments to multi-houses, and

homeownership rate. The study also discusses the effects of residential electricity price,

heating degree days (HDDs), and cooling degree days (CDDs) on the residential electricity

consumption to lead to more concrete empirical results. For empirical methods, the study

applies fully modified least squares (FMOLS) and dynamic least squares (DOLS)

17

estimators with panel cointegration methods and fixed effects and random effects models

for panel data estimations.

18

1.2 Research hypotheses

This section describes several hypotheses to be tested in the research. The

following research hypotheses mainly refer to those related to socio-demographic factors.

1.2.1 Hypothesis I: Population aging and residential electricity

consumption exhibit a negative relationship

The key hypothesis to be tested in this research is about the impact of population

aging, the main component of the socio-demographic factors, on the residential electricity

consumption in Korea. The first hypothesis states that population aging and residential

electricity consumption exhibit a negative relationship, meaning that as the share of

population aged 65 and above increases, the amount of residential electricity per capita

consumes decreases, and vice versa. The results of existing studies support this argument

(Lim et al., 2013; Noh and Lee, 2013; Keum et al., 2018; Brounen et al. 2012). They argue

that the people in different age structures behave differently when consuming energy or

electricity in the residential sectors because the older people usually have strict perceptions

on energy conservation than the other age groups do, and these typical perceptions on

energy savings make them to use less amount of electricity at homes (Lee et al. 2011).

19

1.2.2 Hypothesis II: Increase in household size reduces residential

electricity consumption per capita

The second hypothesis states that the increase in family or household size reduces

the residential electricity consumption per capita. This is because energy efficiency per

household tends to improve as the size of a household gets larger in general. Ota et al.

(2017) has explained this situation by the fact that the increase in the number of people in

a household reduces the needs to duplicate the usage of shared appliances or electricity (i.e.,

televisions, refrigerators, living room lights, etc.).

Yohanis et al. (2008) in their empirical research have even found that the

electricity consumption per capita decreases as the number of people residing increases in

the UK. Also, Druckman and Jackson (2008) have found that per capita electricity

consumption in the UK is negatively correlated to household size, suggesting that a

household with more people is generally more efficient regarding per capita energy

consumption. Referring to these supporting arguments, we hypothesize in our empirical

research that the increase in household size reduces residential electricity consumption per

capita in Korea.

20

1.2.3 Hypothesis III: Education level, gender difference, working

age population, and homeownership do not have significant

effects on residential electricity consumption

The third hypothesis of our research states that education level, gender difference

ratio, working age population, and homeownership rate do not pose significant effects on

residential electricity consumption in Korea. Numerous studies have tried to examine

whether these factors have influence on residential electricity consumption. Although

differing effects of the education level on domestic electricity consumption have been

reported, several studies have confirmed that these factors do not pose significant effects

on residential electricity consumption. For instance, Bedir et al. (2013) and Cramer et al.

(1985) have found that education levels do not significantly affect electricity consumption

in Dutch and US dwellings, respectively. For gender difference ratio, which is

predominantly reflected by male to female or female to male ratio, it does not pose

significant effects on residential electricity consumption. Homeownership rate as well does

not significantly affect residential electricity consumption since one, regardless of owning

or renting a house, must pay its bills for using electricity at homes. Considering the

arguments of these studies, the research hypothesis states that education level is not much

influential on the residential electricity consumption per capita in Korea.

21

1.2.4 Hypothesis IV: Heating degree days (HDDs) and cooling

degree days (CDDs) have positive relationships with

residential electricity consumption

The last hypothesis of our research states that heating degree days (HDDs) and

cooling degree days (CDDs) have positive relationships with residential electricity

consumption in Korea. Simply put, heating degree days refer to measurements calculated

to quantify the demand for energy consumed to heat a building. In specific, heating degree

days are measured relative to balance temperature, the outside temperature above which a

building needs no heating. Similarly, cooling degree days (CDDs) are measurements

designed to quantify the demand for energy consumed to cool a building. In specific,

cooling degree days are measured relative to balance temperature, the outside temperature

above which a building needs no cooling.

Numerous of studies have analyzed the impacts of heating degree days and

cooling degree days on the residential electricity consumption and have conflicting results

as to whether cooling degree days or heating degree days affect the increase in the

residential electricity consumption. In this research, the hypothesis is established that both

heating and cooling degree days will significantly affect the increase in residential

electricity consumption. Nowadays, demand for heating has recently shifted from city gas

to electricity in Korea. Also, for cooling, more people today have air-conditioners installed

at their homes and use them during the summer; also, summer has become hotter today

than summer in the past due to climate change from global warming. With these reasons,

the study makes the fourth hypothesis heating degree days and cooling degree days exhibit

22

positive relationships with residential electricity consumption in Korea.

23

1.3 Research design

Order of the thesis

The order of the thesis is as follows: The first chapter begins with the introduction,

in which research background and objective are discussed. Then, the four main research

hypotheses to be tested in the analysis are described. For the last part of the introduction,

research design, which consists of the order of the thesis and the conceptual framework of

the research, is explained. In the second chapter, brief overviews of the phenomena of

population aging over the world and in Korea are discussed. Then, in the same chapter, the

trends and related policies of residential electricity consumption in Korea are presented for

the residential electricity consumption is the scope to be analyzed in our research. The third

chapter comprises literature review of the related theoretical and empirical studies on the

research topic. The fourth chapter considers the explanations of the data and the empirical

methods used in the analysis. Empirical results are discussed next in chapter five. Lastly,

conclusions of the research along with policy implications and suggestions for further

research will be discussed in chapter six. References, appendices for supplementary figures,

and abstract in Korean are presented in chapter seven, the last part of the thesis.

24

Figure 1.2. Conceptual framework of the research

25

Chapter 2. Overview of population aging

and residential electricity consumption

2.1 Overview of population aging in the world and Korea

Population aging in the world

Population aging is defined as a shift in the distribution of the population of a region

or a country towards older ages. This situation is usually reflected by the declining fertility

rates, which reduces the proportion of the population composed of the children, and

increasing life expectancy, which increases the proportion of the population composed of

the elderly. The United Nations (UN) have reported that among the countries currently

classified as more developed countries with a total population of 1.2 billion in 2005, the

overall median age rose from 28 to 40 in 1950 and 2010, respectively. Further this value is

forecasted to rise to 44 by 2050. For the less developed countries, the median age is

predicted to rise from 26 to 35 in 2010 and 2050, respectively. For the entire countries, the

figures are forecasted to rise from 29 to 36 in 2010 and 2050, respectively.

26

Population aging in Korea

Considered as one of the fastest-aging countries in the world, Korea is expected

to have far more threatening and substantial effects of population aging on the society in

contrast to the other developed countries. As of 2017, Korea has officially entered the era

of an “aged society”1 by reaching 14 percent of the total population aged 65 years and

above (2017 Population and Housing Census, Statistics Korea). In specific, it is forecasted

to take only 27 years for Korea to become a “super-aged society” from an “aging society,”

whereas it is predicted to take 34 years for China, 89 years for the United States, and 157

years for France (US Bureau of Statistics).

The problem is on the fact that the speed at which population aging is progressing

in Korea is incomparably faster than almost every other country in the world. For instance,

Korea’s total fertility rate, which is defined as the number of children who would be born

to a woman in her life time and often is served as a proxy for the predicting the rate at

which population aging progresses in a society, is the lowest in the world with the value at

1.17 (World Population Prospects: 2017 Revision, UN). With these matters behind, many

professionals predict that the aging of population in Korea will have detrimental effects on

many areas in the society, either directly or indirectly, in multi-faceted ways. Namely, the

areas typically affected by population aging of a country include labor markets,

governments, pensions, health care and social security, education, economy, and energy use

1 According to the definitions from the United Nations, aged society is when the share of total

population aged 65 years and above is 14%. For further information, aging society is when the

share of total population aged 65 years and above is 7%, and super-aged society is when the share

of aged 65 years and above is 21%.

27

in the residential sector.

28

Figure 2.1. Comparison of the rate at which aged society becomes super-aged society in

major countries

Source: US Bureau of Statistics

27 yrs.32 yrs.

37 yrs.

89 yrs.100 yrs.

157 yrs.

Korea China Japan USA UK France

29

Figure 2.2. Comparisons between the population pyramids of the world and Korea in 1950

Source: Rearranged from populationpyramid.net (United Nations Population Division, Department of Economic and Social

Affairs. World Population Prospects: The 2015 Revision)

30




31




32

Population pyramids of the World and Korea

Figure 2.2, 2.3, and 2.4 present the comparisons of the population pyramids of the

world and Korea in the year 1950, 2000, and 2050. Population pyramid. Population

pyramid is a graphical method of illustrating the age and sex structures of a population and

often serves as a useful tool for seeing the trends of changes in age and sex structures and

making comparisons with one country to the other countries. Population pyramid has three

main stages: Expansive, stationary, and contractive.

First, populations at the stage of expansive population pyramids usually have the

following characteristics: concave sides for both male and female, high fertility rates, and

low life expectancies. These types of populations are typically characteristic of developing

nations. Figure 2.2 shows that the world and Korea are at the expansive stage, in which a

large percentage of the populations is in younger age groups, reflecting high fertility rates;

a small proportion of the population in older age groups, reflecting low life expectancies.

Also, population pyramid of the world in 2000 in Figure 2.3 has the characteristic of

expansive population pyramid even though some proportions at the younger age cohorts

such as 0-4 and 5-0 are tapered off.

Second, populations at the stage of stationary population pyramids have the

following characteristics in general: rectangular shapes, non-increasing populations, and

relatively equal distributions across the age cohorts but tapered-off toward the top. These

population pyramid shapes are recognized as the characteristic of developed countries,

where the birth rates are relatively low, and the overall quality of life is high. The population

pyramid of the world illustrated in 2050 in Figure 2.4 is representative of the typical shape

33

of a stationary population pyramid.

Lastly, populations at the stage of constrictive population pyramids are illustrated

to describe the populations with the following characteristics: large proportions weighted

at the elderly age groups, shrinking number of populations, smaller percentages of

populations in the younger age groups, very high levels of the dependency ratio of the

population, and patterns tapering in at the bottom. Countries with constrictive population

pyramids exhibit very higher levels of social and economic development, in which access

to health care and quality education is available to a large percentage of the population. The

population pyramids of Korea in 2000 and 2050 in Figure 2.3 and 2.4, respectively,

represent the typical shapes of constrictive population pyramids. The difference between

the two is that the proportions of the population of Korea in 2050 are more towards the

older age cohorts than those in 2000. This means that the socioeconomic and health

developments are progressing in Korea, and that Korea is forecasted to be at even higher

stage of constrictive population pyramids in 2050. Further, these changes in the age

structures with the major proportions of the populations at the elderly age cohorts in Korea

denote the significant impacts which will be imposed on Korea’s national economy and the

society in the future.

34

2.2 Overview of residential electricity demand in Korea

Brief background of electricity consumption in Korea

Over the past few decades since 1970s, the cheap electricity prices in Korea have

driven the industries to gain competitiveness in domestic and global markets, supporting

its national economy to flourish at a very fast speed compared to any other country in the

world. Moreover, the individuals in Korea have gained their quality of life owing to the

cheap electricity prices which allow the people to freely use residential electricity. In 2016,

electricity consumption in residential and commercial sectors accounted for approximately

38.95% of the total electricity consumption in all sectors; electricity consumption in

residential sector accounted for 13.31% of the total electricity consumption (Korea Energy

Agency, Energy Statistics Handbook 2018).

However, Korea’s cheap electricity prices exhibit some downsides since they

provide incentives for the industries and households to consume more energy than

necessary. Figure 2.5 presents the annual trend of final energy consumption structures by

energy sources in Korea from 1990 to 2016. In 1990, the share of electricity consumption

has reached 10%, and since then, its percentage has almost doubled to approximately 20%

in 2016. The point that is worth to take a note hereinafter is that the rate at which the

electricity consumption increases has slowed down significantly since 2000s. This means

that currently in which the changes in demand for electricity are frequently occurring, it is

of utmost importance for the scholars to closely investigate on the factors that affect the

35

electricity consumption and thus able to urge the policymakers to implement the suitable

electricity policies in the right way.

Explanations on Table 2.1 and Figures 2.5 through 2.9

Table 2.1 presents the estimated results of compound annual growth rate (CAGR)

of final energy consumption by energy source in Korea during the periods of 1990-2016,

1981-1990, 1991-2000, 2001-2010, and 2011-2016. The results show that during 1981-

1990, the total energy consumption rate had reached 6.73%, and during 1991-2000, the

energy consumption had reached 6.05%. These results of fast energy consumption rate are

mainly due to the continued growth of the national economy led by high energy-consuming

industries such as petrochemicals and steel. As shown in Figure 2.5, however, the energy

consumption rate which had been on a steady rise until the late 1990s has slowed down

significantly since the early 2000s. For instance, during 2010-2010, the compound annual

growth rate of final energy consumption had reached 2.44%, and during 2011-2016, the

compound annual growth rate had reached 1.51%. This sharp decline is mainly caused by

the changes in industrial structures from relatively high energy-consuming industries to

low energy-consuming industries such as telecommunications and service industries.

In addition, although some differences exist by energy type, the rate at which the

energy consumption increases overall has slowed down significantly since 2000s,

especially the rate of petroleum consumption. As presented in Figure 2.6, although the rate

at which electricity consumption decreases is relatively slower than the rate at which

36

petroleum consumption decreases, the former is slowing down over time. For instance,

during 1981-1990, the compound annual growth rate of final electricity consumption had

reached relatively high percentage of 10.29%, and during 1991-2000, the rate had reached

8.66%. However, since 2000s, the growth rate had reached 5.35% during 2001-2010, and

1.48 % during 2011-2016.

Figure 2.7 presents the trend of final electricity consumption in Korea during the

period of 1981-2016. Except for in 1997 and 1998 during which the Asian financial crisis

occurred, total electricity consumption in Korea had consistently increased. In 2009, during

which the global financial crisis occurred, the electricity consumption rate slowed down

but did not decline. However, experts predict that there may be some possibilities of decline

in the rate of electricity consumption due to the enhancement in the management of demand

following the cyclical blackout in September 2011 and the decline in heating and air-

conditioning decreases following the global climate change.

Figure 2.8 presents the annual trend of electricity consumption in residential

sector in Korea during the period of 1993-2016. It can be seen from the graph that except

for the period of 1997-1999 during which the Asian financial crisis had occurred and posed

severe burden on the global and national economies, the trend of residential electricity

consumption in Korea has increased continuously. This may be explained as a result of

increased number of household appliances supplied to households due to increased income

levels.

Figure 2.9 presents the monthly trend of electricity consumption in residential

sector in Korea during January 1999 and January 2015. Clear seasonal trend for electricity

consumption is shown in the graph. The demand for electricity consumption in the winter

37

has increased since the mid-2000s, and since the 2010s, the number of peak consumptions

occurring in winter has increased. This is believed by many that while oil and gas prices

have risen significantly due to the rapid increase in the international oil prices, increase in

electricity price has been curtailed due to the government regulations, leading to the

increase in demand for heating electricity.

38

Table 2.1. Compound annual growth rate (CAGR) of final energy consumption by energy source in Korea (1990-2016)

Source: Korea Energy Statistical Information System (KESIS)

Period Coal Petroleum Natural

Gas City Gas Electricity

Heat

Energy Renewable Total

1981-1990 2.99% 9.12% - 45.98% 10.29% - -10.77% 6.73%

1991-2000 0.11% 5.94% - 23.35% 8.66% 30.74% 13.17% 6.05%

2001-2010 3.31% 0.71% - 4.72% 5.35% 4.52% 8.09% 2.44%

2011-2016 -0.31% 1.86% -7.43% -0.51% 1.48% 1.67% 11.04% 1.51%

1981-2016 2.25% 5.12% - 21.03% 7.61% - 4.19% 4.99%

39

Units: 1000 TOE

Figure 2.5. Annual trends of final energy consumption by energy source in Korea,

1990-2016


0

20,000

40,000

60,000

80,000

100,000

120,000

Coal Petroleum Natural Gas City Gas

Electricity Heat Energy Renewable

40

Figure 2.6. Annual trend of final energy consumption structures by energy source in

Korea, 1990-2016


0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

199

0

199

1

199

2

199

3

199

4

199

5

199

6

199

7

199

8

199

9

200

0

200

1

200

2

200

3

200

4

200

5

200

6

200

7

200

8

200

9

201

0

201

1

201

2

201

3

201

4

201

5

201

6

Coal (%) Petroleum (%) Natural Gas (%) City Gas (%)

Electricity (%) Heat (%) Renewable (%)

41

Figure 2.7. Annual trend of final electricity consumption in Korea, 1981-2016


42

Figure 2.8. Annual trend of electricity consumption in residential sector in Korea, 1993-

2016

Source: Korea Power Statistics, Korea Electric Power Corporation (KEPCO)

43

Figure 2.9. Monthly trend of electricity consumption in residential sector in

Korea, 1999:01-2015:01

Source: Korea Power Statistics, Korea Electric Power Corporation (KEPCO)

44

2.3 Comparison with Germany and Sweden on the effects of

population aging on residential energy consumption

According to a comparison analysis conducted from Swedish Defence Research

Agency (FOI) in 2009, the amount of energy consumed differs by age cohorts in Germany

and Sweden. The report states that the youngest singles born after 1979 use energy the least,

and the oldest people born before 1949 use energy the most. The study has found that the

proportion of energy consumed in housing increases with age, and as shown in Figure 2.10

and 2.11, the disparities in energy consumption by age groups are more pronounced in

Sweden than in Germany. This can be explained by the fact that in Germany and Sweden,

the older people are relatively more economically stable than the younger people, and also

that the older people have similar patterns with younger people in using the home

appliances and the electronic devices. It is worth to take a note on the projections of some

professionals that, contrary to the current trend, the growing elderly population in Korea

may increase its residential energy consumption as Korea begins to possess the similar

characteristics of the advanced countries like Germany and Sweden in the future.

Also, Germany and Sweden are the two notable countries with high levels of

Energy equity index (12th and 17th, respectively), providing firm grounds for their

facilitated access to and affordability of energy among all groups of people; in contrast,

Korea ranks 26th in Energy equity index, respectively, which signals that energy is not

distributed equitably to all groups of people.

45

Units: Megajoule (MJ)

Figure 2.10. Energy consumption for housing by gender and age differences in

Germany

0

20000

40000

60000

80000

100000

120000

born before

1945

1959 1979 born after 1979

Men

Women

46

Units: Megajoule (MJ)

Figure 2.11. Energy consumption for housing by gender and age differences in

Sweden

0

10000

20000

30000

40000

50000

60000

70000

80000

born before

1945

1959 1979 born after 1979

Men

Women

47

Chapter 3. Literature Review

3.1 Theoretical studies

3.1.1 Houthakker (1951) and subsequent studies on the

demand for electricity

Houthakker (1951) has initiated the study on estimating the demand for electricity

in the UK using econometrics model. The study analyzes 42 British provincial cities for

the periods of 1938 and 1939, using cross-sectional data. The study applies generalized

least squares to address the issues of heteroscedasticity problem. The author assumes the

existence of a stable demand function and demonstrates that the demand for electricity is

responsive to changes in both price and income. The results of the study conclude that

electricity consumption is influenced by electricity price and household’s annual income.

Since this pioneering study of Houthakker (1951), vast number of studies have

attempted to estimate the residential electricity, incorporating the theoretical and empirical

concepts of Houthakker’s (1951) work. A few studies that estimate the residential

electricity demand following Houthakker (1951) are briefly summarized below.

Chern and Bouis (1982) two-stage least squares (2SLS) to estimate the structural

changes in demand for residential electricity in USA during 1955-1978. The study finds the

existence of significant structural changes associated with 1973 oil embargo or the reversal

48

of the declining trend of electricity prices during 1969-1972. Income and price elasticities

for residential electricity are found to be positive and negative, respectively. Silk and Joust

(1997) uses cointegration approach used by Bentzen and Engsted and error correction

model to estimate the short-run and the long-run elasticities of US residential electricity

during the period of 1949-1993. The empirical results of the study suggest that the short-

run elasticities for income and price exhibit the positive and negative signs, respectively;

the long-run income elasticity is found to be approximately 0.5. The evidence of a structural

shift in demand in the middle of the 1960s is found as well.

Holtedahl and Joutz (2004) use error correction model and Engle and Granger’s

cointegration methods to investigate the short-run and long-run elasticities of residential

electricity demand in Taiwan during 1955-1995, the time of rapid process of modernization

and economic development. The study in the empirical results find that both short-run

income and price effects are smaller than the long-run effects. For short-run and long-run

elasticities, coefficients for income are found to be 0.23 and 1.04 for income, respectively;

the price elasticity is found to be negative and inelastic with the coefficient of -0.15.

Narayan and Smyth (2005) uses autoregressive distributive lag (ARDL) method to

estimate the short-run and the long-run elasticities for the determinants of residential

electricity demand in Australia during the period of 1969-2000. The study finds that price

and income most significantly affect the demand for residential electricity; income

elasticity of demand is positive with 0.323~0.408, and price elasticity of demand in the

long run is negative with -.541.

Nakajima (2010) uses panel cointegration approach with dynamic ordinary least

squares (DOLS) estimators to examine the demand for residential electricity in Japan

49

excluding Okinawa-prefecture during the period of 1975-2005. The study concludes that

all variables, sales per household, electricity price, and income per household are appeared

to have a unit root and cointegrating relationships. The price effect is negative and elastic,

and the income effect is inelastic.

50

Table. 3.1. Summary of the literature studies for Section 3.1

Author Region Period Methodology

Relationship

with residential

electricity

consumption

Income Price

Houthakker (1951) UK 1938,

1939 GLS + -

Silk and Joust (1997) US 1949-

1993

ECM, Bentzen

and Engsted's

cointegration

+ -

Chern and Bouis

(1999) US

1955-

1978 2SLS + -

Holtedahl and Joutz

(2004) Taiwan

1955-

1995

ECM,

cointegration + -

Narayan and Smyth

(2005) Australia

1969-

2000 ARDL + -

Nakajima (2010) Japan 1975-

2005

panel

cointegration,

DOLS

+ -

51

3.1.2 Residential electricity demand model considering

socio-demographic characteristics

It has long been recognized that energy demand can change with demographic

characteristics, but of little interest to the scholars to analyze it empirically (Halvorsen,

1975; Flaig, 1990; Yamasaki and Tominaga, 1997). With this background, a stream of

research has started to apply the household production theory to estimate the residential

electricity demand, the demand derived from warming or cooling homes, cooking, using

home appliances, etc., considering demographic characteristics (Filippini, 1999).

The main idea of the household production theory is that the households are both

producers and consumers of goods and that the households try to maximize their utilities.

Applying this framework of household production theory, we can estimate the demand

for residential electricity. Filippini (1999) asserts that the households use electricity and

capital goods to produce and consume energy products; in other words, one’s utility is

not generated directly from the electricity itself but generated from consuming the

mixtures of energy-related services or products and capital goods.

52

Figure 3.1. The framework of application of household production theory on residential

electricity considering socio-demographic characteristics

Source: Reorganized from Filippini (1999)

53

First, the production function of the composite energy commodity S is written as

follows:

𝑆 = 𝑆(𝐸, 𝐶𝐺) (1)

where CG is the capital goods which consist of appliances and E is electricity consumption.

The output of the composite energy commodity is determined by the quantity of the capital

goods of appliances and the amount of electricity purchased. Further, the household is

assumed to have a utility function with normal properties of differentiability and curvature.

Then, the Equation (2) is set with the composite energy commodity S and a purchased

composite numeraire good X which directly yields a utility, and D and G which represent

demographic characteristics which determine the household’s preferences.

𝑈 = 𝑈(𝑆, 𝑋; 𝐷, 𝐺) (2)

where Equation (2) is satisfied with the following properties:

𝛿𝑆

𝛿𝐸≥ 0,

𝛿𝑆

𝛿𝐶𝐺≥ 0 ,

𝛿2𝑆

𝛿𝐸2 ≤ 0, 𝛿2𝑆

𝛿𝐶𝐺2 ≤ 0 (3)

Then, the household is assumed to maximize the utility of Equation (1) under the

54

budget constraint of the following equation:

𝑌 − 𝑃𝑆𝑆 − 1𝑋 = 0 (4)

where Y is income, 𝑃𝑆 is price of the composite energy commodity, and 𝑃𝑋 is the price

of composite numeraire good X, where the 𝑃𝑋 is assumed as 1 in Equation (4).

55

Figure 3.2. Steps to derive the demand for residential electricity consumption in

Deaton and Muellbauer (1980)

56

According to Deaton and Muellbauer (1980), decisions by individual households

to maximize their utilities in the household production model take place in two stages.

The first step is to minimize the cost of obtaining energy products and to derive the

demand for power as an input. The second step is to derive the demand for electricity by

maximizing the utility with a cost function given.

In the first stage of optimization process, the consumer behaving as a firm can

obtain the optimized levels of the demands for inputs for electricity consumption (E) and

capital goods (CG) by minimizing the cost of producing S as the following equation:

𝑀𝑖𝑛(𝑃𝐸𝐸 + 𝑃𝐶𝐺𝐶𝐺)

𝑠. 𝑡. 𝑆 = 𝑆̅(𝐸, 𝐶𝐺) (5)

where 𝑃𝐸 represents the electricity price and 𝑃𝐶𝑆 represents the price of a composite

good. With the cost minimization process under the budget constraints in Equation (5),

the following Equation (6) is achieved:

𝐶 = 𝐶(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑆) (6)

According to Varian (1992), Equation (6) has the following properties: (1) the

function is linear homogenous in S and in factor prices and (2) and increasing in S and

nondecreasing and concave in factor prices. With application of Shephard’s lemma, the

following input demand functions are derived:

57

𝐸 =𝛿𝐶(𝑃𝐸,𝑃𝐶𝐺,𝑆)

𝛿𝑃𝐸= 𝐸(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑆) (7)

𝐶𝐺 =𝛿𝐶(𝑃𝐸,𝑃𝐶𝐺,𝑆)

𝛿𝑃𝐶𝐺= 𝐶𝐺(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑆) (8)

In the second stage of optimization process, household maximizes its utility under

the budget constraint as follows:

𝑀𝑎𝑥 𝑈(𝑆, 𝑋; 𝐷, 𝐷)

𝑠. 𝑡. 𝐶(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑆) + 𝑋 = 𝑌 (9)

The corresponding Lagrangian function is:

𝐿 = 𝑈((𝑆, 𝑋; 𝐷, 𝐷) + 𝜆(𝑌 − 𝐶(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑆) − 𝑋)) (10)

The solution of Equation (8) leads to the following demand function for the energy

commodities S:

𝑆∗ = 𝑆∗(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑌, 𝐷) (11)

Then finally, by substituting Equation (11) into Equation (7), the demand functions for

residential electricity of a household can be found as follows:

58

𝐸 = 𝐸(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑆∗(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑌, 𝐷))

= 𝐸(𝑃𝐸 , 𝑃𝐶𝐺 , 𝑌, 𝐷) (12)

where Equation 12 reflects the demand for residential electricity of a household considering

the demographic characteristics (D) in the function. In this case, population aging and other

demographic factors may be included, representing the features of demographic

characteristics (D).

Usually in the empirical analysis, the price of capital goods (𝑃𝐶𝐺) is not included

for the following reasons: First, it is difficult to find the fitting data representing the price

of capital goods; second, the changes in the prices of capital goods are relatively small and

so excluding these prices from the model does not cause distortions of the remaining

estimated values (Halvorsen, 1975; Flaig, 1990; Filippini, 1999).

59

3.2 Population aging

Many studies have long been analyzing the factors that influence the electricity

demand from different viewpoints. In this section, the literature which mainly discuss

various factors that influence the electricity demand will be discussed in the following

section.

Studies on the impact of population aging on the residential electricity consumption in

Korea

This section primarily examines the literature that investigate the effects of

demographic factors, such as population aging, on residential electricity consumption. The

first part of the two following sections will discuss the literature that specifically investigate

the conditions in Korea, and the last part will discuss the literature that analyze the effects

on countries other than Korea. Studies that analyze the effects of population aging on

residential electricity consumption in Korea are as follows.

60

1) Shin et al. (2016)

Shin et al. (2016) analyzes whether population aging has varying effects on

residential electricity consumption according to different income levels. The study analyzes

the heterogeneity of household electricity consumption caused by income level using

threshold regression methods with cross-sectional data for urban electricity consumption

in Korea in 2013. The data used are residential electricity consumption (kWh), average and

marginal prices of electricity (KRW/kWh, monthly income (1000 KRW), HDD, CDD,

number of residents, size of the residential area, and a dummy variable for aged person

within a household. The empirical results confirm that the effects of aging on residential

electricity consumption differ according to monthly income levels: For low income levels,

the coefficient for population aging is -0.010; for middle income levels, the coefficients for

population aging is 0.008; for high-income levels, population aging does not have

significant effect on residential electricity consumption.

Also, the study suggests that the size of price elasticity, the kind of perception price,

and the effects of other factors differ according to household income level. For instance,

households with monthly income less than 1.5 million won respond to average price of

elasticity for electricity usage; households with monthly income more than 1.5 million won

respond to marginal price instead of the average price for electricity consumption. Further,

the study finds that the households with low income is relatively price inelastic compared

to those with high income. the income levels are below 15 million won, the relationship

between aging and residential electricity consumption are negative; if the income levels are

in the range between 15 million ~ 27 million won, the relationship is positive; if the income

61

levels are above 27 million won, the relationship is not statistically significant.

2) Shin (2018)

Shin (2018) investigates population aging index that best explains household

electricity consumption in Korea and further predicts the volatility of household electricity

consumption according to aging trends. To conduct an analysis, the study uses Mallow’s

moving-average model and fixed effects threshold panel regressions on 16 municipalities

in Korea during 2003-2015. The equation for the fixed effects threshold panel regression

model is as follows:

𝑒𝑖,𝑡 = (𝜇𝑖 + 𝛣1𝑦𝑖,𝑡 + 𝛣2𝑝𝑖,𝑡 + 𝛣3𝑚𝑖,𝑡)𝐼[𝑎𝑖,𝑡 ≤ 𝛾]

+ (𝜇𝑖′ + 𝛣1

′ 𝑦𝑖,𝑡 + 𝛣2′ 𝑝𝑖,𝑡 + 𝛣3

′ 𝑚𝑖,𝑡)𝐼[𝑎𝑖,𝑡 ≥ 𝛾] + 𝑒𝑖,𝑡

where 𝐼[∙] is an indicator function in which a random variable takes the value 1 if the

expression inside the brackets is true or takes the value 0 if the expression inside the bracket

is false. Hence, if 𝑎𝑖,𝑡 or population aging level is greater than 𝛾, the price and income

elasticities can become from 𝛣1, 𝛣2 to 𝛣1′ , 𝛣2

′ . The study concludes that as a society

progresses towards an aging society, income elasticity for residential electricity demand

becomes more inelastic, whereas price elasticity for residential electricity demand becomes

more elastic. The study further gives policy insights for stabilization of volatility in

62

electricity consumption, emphasizing the importance of implementations of both price

policies and efficiency improvement policies.

3) Lim et al. (2013)

Lim et al. (2013) consider both the aging variables and the temperature variables

such as cooling degree days (CDDs) and heating degree days (HDDs) to investigate these

impacts on the residential electricity consumption in Korea using autoregressive distributed

(ARDL) model and error correction model. Time series data from 1966 to 2011 are used

for analysis. The following equation represents the long-run equilibrium function for the

residential electricity demand:

𝑙𝑛𝑄𝑡 = 𝛼0 + 𝛼1𝑙𝑛𝑌𝑡 + 𝛼2𝑙𝑛𝑃𝑡 + 𝛼3𝑙𝑛𝐷𝑡 + 𝛼4𝑙𝑛𝐶𝐷𝐷 + 𝛼5𝑙𝑛𝐻𝐷𝐷 + 𝜖𝑡

which is derived from ARDL model. In the above equation, 𝛼𝑚 represents the long-run

elasticity parameters of the demand for residential electricity. Y, P, D, CDD, and HDD

represent the variables for real income per capita, real price for residential electricity, aging

index, cooling degree days, and heating degree days, respectively.

The study concludes that, in terms of population aging, 1% increase in the levels of

aging index decreases the residential electricity consumption by 0.48%. For the

temperature effects, the study finds that cooling degree day variables have statistically

significant impacts on Korea’s residential electricity consumption, whereas heating degree

63

day variables do not have any impacts on the residential electricity consumption. The study

is significant for it analyzes not only the aging effects but also the global warming effects

due to climate change on the electricity demand.

4) Noh and Lee (2013)

Noh and Lee (2013) analyzes the factors affecting the energy consumption of the

household sector including oil and city-gas consumption and electricity consumption using

panel data for 7 metropolitan areas and 9 provinces in Korea for the period of 2001-2010;

the study excludes the province of Jeju considering the specialties of the islands. For

estimation, the study applies feasible generalized least squares (FGLS) method. The panel

model for estimations is as follows:

𝐸𝑖𝑡 = 𝛼 + 𝐿𝑖𝑡𝛣1 + 𝑃𝑖𝑡𝛣2 + 𝜖𝑖𝑡

(𝜖𝑖𝑡 = 𝜇𝑖𝑡 + 𝜆𝑡 + 𝜐𝑖𝑡)

where i represents 15 metropolitan areas and provinces, t represents time, E represents

energy (electricity, oil and city gas) consumption in residential and transport sectors, L

represents household characteristics (income, the ratio of population aged over 65, average

household size, and female economic participation rate) and consumption styles, and P

represents consumption environment (built environment and energy price).

The empirical results find that the increase in the ratio of population aged over 65

64

by 1% increases residential electricity consumption by 0.29% but decreases the oil and

city-gas consumption by 1.09%. Overall, the total residential energy consumption

decreases by 0.87% as the ratio of population aged over 65 increases by 1%. In conclusion,

the study asserts that residential electricity consumption is not only affected by the

geographical and physical environments, but also by the socio-demographic and economic

factors including household characteristics and consumption styles and environment.

5) Won (2012)

Won (2012) analyzes how population aging affects residential electricity

consumption in Korea using the annual data times series data from 1965 to 2010. The study

uses error correction model and Johansen cointegration methods to estimate the short-run

dynamics and long-run equilibrium relationships for the residential electricity demand. The

error correction model equation to estimate the long-run equilibrium relationships is as

follows:

𝑑𝑙𝑛𝐸𝑡 = 𝛣0 + 𝐵1𝑑𝑙𝑛𝑃𝐸𝑡 + 𝐵2𝑑𝑙𝑛𝑌𝑡 + 𝐵3𝑑𝑙𝑛𝐷𝑡 + 𝐵4𝑢𝑡−1 + 𝜐𝑡

where d represents the difference operator, 𝜐𝑡, an error term, and 𝑢𝑡−1, an error correction

term. For the variables, E, Y, P, and D represent residential electricity consumption (kWh)

per capita, real income per capita (10000 KRW/person), residential electricity price

(KRW/kW), and aging index, respectively. The empirical results show that population

65

aging significantly reduces residential electricity demand in the short run, but the effect

gradually decreases in the long run even though the relationship between the two variables

is still negative.

6) Noh (2014)

The study investigates the effects of household demographics on household

electricity consumption in 2008. The paper focuses on the impacts of numerous household

demographics on electricity consumption after controlling for the consumptions of various

household appliances, durable goods, and traditional factors influencing electricity

consumption such as electricity price and income. The controlled variables for the usages

of household appliances include televisions, refrigerators, Kimchi refrigerators, washing

machine, dryers, air purifier, rice cooker, stoves, fans, heaters, laptops, and desktop

computers, etc. Also, the price uses city gas consumption variables for controlling. The

study uses ordinary least squares (OLS) method for estimation on the following model:

𝑙𝑛𝑌𝑖 = 𝛼 + 𝛣1𝑙𝑛𝑃𝑖 + 𝛣2𝑙𝑛𝑋𝑖 + 𝛣3𝑍𝑖 + 𝛣4𝛿𝑟 + 𝜖𝑖

where 𝑌, 𝑃, 𝑋, 𝑍, 𝛿 each represents residential electricity consumption, unit price of

electricity, dwelling types, household characteristics of a household i, and dummy variables

reflecting the regional characteristics of 16 municipalities.

The study finds that the presence of toddlers under age 5 or elders over age 65 in

66

a household has negative relationship with the residential electricity consumption.

However, these effects are found to disappear when the consumptions of household

appliances and city gas consumption are controlled.

7) Keum et al. (2018)

The purpose of this study is to estimate the determinants of the demand for

electricity by using First Differenced Generalized Method of Moments (FD GMM) method

with the unbalanced panel data of 16 municipalities in Korea during 1the period of 1996-

2013. The following two models represent ARDL model and partial adjustment model,

respectively:

𝛥𝑙𝑛𝐸𝑖,𝑡 = 𝛣0 + 𝛣1𝛥𝑙𝑛𝐸𝑖,𝑡−1 + 𝛣2𝛥𝑃𝑖,𝑡 + 𝛣3𝛥𝑙𝑛𝑃(𝑖,𝑡−1) + 𝛣4𝛥𝑙𝑛𝑌𝑖,𝑡 + 𝛣5𝛥𝑙𝑛𝐴𝑖,𝑡

+ 𝛣6𝛥𝑙𝑛𝐶𝐷𝐷𝑖,𝑡 + 𝛣7𝛥𝑙𝑛𝐻𝐷𝐷𝑖,𝑡 + 𝛣8𝛥𝑙𝑛𝑇𝐷𝑖,𝑡 + 𝛥𝑢𝑖,𝑡

𝛥𝑙𝑛𝐸𝑖,𝑡 = 𝛣0 + 𝛣1𝛥𝑙𝑛𝐸𝑖,𝑡−1 + 𝛣2𝛥𝑃𝑖,𝑡 + 𝛣3𝛥𝑙𝑛𝑌𝑖,𝑡 + 𝛣4𝛥𝑙𝑛𝐴𝑖,𝑡 + 𝛣5𝛥𝑙𝑛𝐶𝐷𝐷𝑖,𝑡

+ 𝛣6𝛥𝑙𝑛𝐻𝐷𝐷𝑖,𝑡 + 𝛣7𝛥𝑙𝑛𝑇𝐷𝑖,𝑡 + 𝛥𝑢𝑖,𝑡

where E, Y, P, A, CDD, and HDD represent residential electricity consumption, real income

per capita, average real electricity price, aging population ratio, cooling degree days

(CDDs), and heating degree days (HDDs), respectively on region i and time series t. The

empirical results of the study suggest the following: 1) Partial adjust model gives better

67

results than ARDL model does in terms of statistical significance; 2) the elasticities of price

and income are found to be negative (-) and positive (+), respectively, and the long-run

elasticities are more elastic than the short-run elasticities; 3) the elasticities of CDDs and

aging population ratio are positive (+) and (negative), respectively; 4) since increase of

income elasticity dominates the increase of price elasticity, the government is

recommended to consider this in implementing the electricity policies.

68

Studies on the impact of population aging on the residential electricity consumption in

overseas countries

In the following section, the studies that analyze the effects of population aging on

residential electricity consumption in overseas countries other than Korea are discussed.

1) Liddle (2011)

Liddle (2011) examines consumption-driven environmental impact and age

structure on the residential electricity consumption in 22 OECD countries during 1960-

2007. For empirical analysis, the study uses cointegration modeling, FMOLS approach in

the framework of Stochastic Impacts by Regression on Population, Affluence, and

Technology (STIRPAT) on several variables including the share of residential energy

consumption from electricity, the share of population between ages 20-34, 35-49, 50-69,

and 70 and older, and the total mid-year population. The equation to be estimated for

residential electricity consumption is as follows:

𝑙𝑛𝐼𝑖𝑡 = 𝑎𝑖 + 𝑣𝑙𝑛𝑃𝑇,𝑖𝑡 + 𝑤𝑙𝑛𝑆ℎ𝑃20−34,𝑖𝑡 + 𝑥𝑙𝑛𝑆ℎ𝑃35−49,𝑖𝑡 + 𝑦𝑙𝑛𝑆ℎ𝑃50−69,𝑖𝑡

+ 𝑧𝑙𝑛𝑆ℎ𝑃70+,𝑖𝑡 + 𝑐𝑙𝑛𝐴𝑖𝑡 + 𝑑𝑙𝑛𝑆ℎ𝐸𝑖𝑡 + 𝜖𝑖𝑡

where ShE, I, 𝑃𝑇, ShP, and A are the share of electricity in residential energy consumption,

the aggregate environmental impact or residential electricity consumption, population total,

share of population in the four age cohorts, and GDP per capita, respectively. 𝜖 and 𝛼 are

69

error terms and constants, respectively.

The empirical results find a U-shaped relationship between residential electricity

consumption and age structure: the youngest cohorts (age cohorts of 20-34) and the oldest

cohorts (age cohorts of 70 and above) have positive coefficients, while the middle ones

(age cohorts of 35-49 and 50-69) have negative coefficients.

2) Mikayilov et al. (2018)

Mikayilov et al. (2018) investigate the impacts of population age groups of 15-64

and 65 above on residential energy consumption in Kazakhstan covering the period of

1999-2012. The study uses the modelling framework of Stochastic Regression on

Population, Affluence, and Technology (STIRPAT) developed by Dietz and Rosa (1994)

and autoregressive distributed lag (ARDL) approach for estimation. The equation in the

STIRPAT framework to be estimated is as follows:

𝑙𝑛𝐼 = 𝑞 + 𝑏𝑙𝑛𝑃 + 𝑐𝑙𝑛𝐴 + 𝑑𝑙𝑛𝑇 + 𝑤

In STIRPAT framework, the variables I, P, A, and T represent environmental impact,

population, affluence, and technology, respectively. For conceptual modelling in the study,

I, P, A, and T serve as proxies for the residential electricity consumption, population age

groups of 15-64 and 65 and above, GDP per capita (2015=100), GDP, respectively. The

study applies time series cointegration and error correction methods for empirical analysis,

70

and the results indicate a significant positive impact of the age groups of 15-65 on the

residential electricity use in the long run. Further, the study finds that 65 and above has

positive short-run effects, whereas affluence has no effect.

3) Brounen et al. (2012)

Brounen et al. (2012) focus on gas and electricity consumption, investigating the

impact of the physical structure of home (i.e., the features of dwellings) on variations in

energy consumption. Then, the study compares the importance of these structural

characteristics with the demographic characteristics of households in the energy

consumption of Dutch households with the observations on 300,000 in the Netherlands

collected from January 2008 to December 2009. For analysis, the study conducts OLS

regressions using datasets including gas and electricity consumption, period of construction,

number of persons in household, age of the head of households and elderly households.,

and household income. The empirical results suggest that the presence of elder person in a

household decreases residential electricity consumption, with the coefficients ranging from

-0.039 to -0.020.

71

4) Deutsch and Timpe (2013)

According to Deutsch and Timpe (2013), residential energy demand is influenced

by several factors including socio-demographic attributes. Deutsch and Timpe (2013)

investigate the effect of aging on residential energy demand in Germany in 2010. The study

uses several variables for analysis: Residential energy consumption (heating, gas,

electricity), social-demographic factors at the household level (number of household

members, age). For control variables, building types, building age, building renovation

status, and climatic factors such as heating degree days are used. The methodological

approach used in the study is one that has been developed by Prognos and AGFW (the

German District Heating Association). The results of the study confirm that the effects of

aging on the residential energy consumption is positive.

5) Mikayilov and Hasanov (2017)

The objective of this study is to investigate the effects of population age groups of

0-14, 15-64, and 65 and above on residential electricity consumption in Azerbaijan during

the period of 2000-2015. The study considers the STIRPAT framework and employs error

correction (ECM) model to ease the possible spurious estimation results caused by using

the non-stationary data and autoregressive distributed lag (ARDL) approach which is

suitable method to use with small samples. The equation to be estimated is as follows:

72

𝛥𝑟𝑒𝑐𝑡 = 𝑐0′ + 𝜃1

′𝑟𝑒𝑐𝑡−1 + 𝜃2′ 𝑔𝑑𝑝𝑝𝑐𝑡−1 + 𝜃3

′ 𝑝𝑜𝑝65𝑡−1+ 𝛴𝑖=1

𝑛 𝜔𝑖′𝛥𝑟𝑒𝑐𝑡−𝑖

+ 𝛴𝑖=0𝑛 𝜑𝑖

′𝛥𝑔𝑑𝑝𝑝𝑐𝑡−𝑖 + 𝛴𝑖=0𝑛 𝜏𝑖

′𝛥𝑝𝑜𝑝_65𝑡−𝑖 + 𝑢𝑡′

where 𝜃𝑖′ is a long-run coefficient, ωi

′ , 𝜑𝑖′ , and 𝜏𝑖

′ are short-run coefficients, ci′ is

constant term, and 𝑢𝑡′ is an error term. Pop_65 represents the age groups of 65 and above

measured in persons. The study concludes that there exist significant effects of different

population age groups and affluence on residential electricity consumption. The elasticities

with respect to population age groups of 15-64 and 65 and above are found to be 10.46 and

2.33, respectively.

73

Table 3.2. Summary of literature on the relationship between population aging and

residential electricity consumption

Author Region(s) Period Methodology Relationship with residential

electricity consumption (REC)

Shin et al.

(2016) Korea 2013

threshold

regression

methods

- for low income households;

+ for middle income households;

not significant for high income

households

Shin (2018)

16

municipal

ities in

Korea

2003-

2015

threshold

panel

regressions

makes income elasticity of

residential electricity demand more

inelastic;

price elasticity of residential

electricity demand more elastic

Lim et al.

(2013) Korea

1966-

2011 ARDL -

Noh and

Lee (2013)

16

municipal

ities in

Korea

2001-

2010 FGLS -

Won

(2012) Korea

1965-

2010

ECM,

Johansen

cointegration

method

short-run: -;

long-run: -, but the effect gradually

decreases

Noh (2014) Korea 2008 OLS -

Keum et al.

(2018)

16

municipal

ities in

Korea

1996-

2013

FD GMM,

ARDL -

Liddle

(2011)

22 OECD

countries

1960-

2007

STIRPAT,

FMOLS,

panel

cointegration

age cohorts of 20-34, 70 and

above: +;

age cohorts of 35-49, 50-69: -

Mikayilov

et al.

(2018)

Kazakhst

an

1999-

2012

STIRPAT,

ARDL

age cohorts 15-65: +;

age cohorts 65 and above: +

Brounen et

al. (2012)

Netherlan

ds

Jan.

2008-

Dec.

2009

OLS -

74

Deutsch

and Timpe

(2013)

Germany 2010

Prognos and

AGFW

method

+

Mikayalov

and

Hasanov

(2017)

Azerbaija

n

2000-

2015

STIRPAT,

ECM, ARDL +

75

3.3 Household size

In this section, the studies analyzing the effects of household size on the residential

electricity consumption in Korea and in overseas countries are discussed.

Studies on the impact of household size on the residential electricity consumption in Korea



household sector (e.g., residential electricity consumption, oil and city gas consumption)

using panel data for 7 metropolitan areas and 9 provinces in Korea for the period of 2001-

2010. For estimation, the study excludes Jeju province considering its specialties of the

islands and applies both fixed effects and random effects models. The estimated model for

panel analysis is as follows:

𝐸𝑖𝑡 = 𝛼 + 𝐿𝑖𝑡𝐵1 + 𝑃𝑖𝑡𝐵2 + 𝜖

(𝜖𝑖𝑡 = 𝜇𝑖 + 𝜆𝑖 + 𝜐𝑖𝑡)

where 𝐸𝑖𝑡 represents energy consumption or residential electricity consumption at time t

in region i, 𝐿𝑖𝑡 represents variables for consumption styles, and 𝑃𝑖𝑡 represents variables

for consumption environment such as built environment and electricity price. The study

76

finds that household size and residential energy consumption exhibits negative relationship.

2) Hong et al. (2018)

The purpose of this study is to analyze whether single and elderly households have

different energy consumption patterns than the general household. In specific for energy

consumption, the study includes residential electricity consumption. The data are collected

from Household Energy Standing Survey conducted in 2011 on 16 municipalities, in which

2,520 household data are chosen in specific. For empirical analysis, the study applies

regional fixed effects model; the estimated panel model is as follows:

𝑙𝑛 𝐸𝑖𝑡 = 𝑎 + 𝐵𝑆𝑆𝑖𝑡 + 𝐵𝐻𝐻𝑖𝑡 + 𝐵1𝑙𝑛𝑀𝑃𝑖𝑡 + 𝐵2𝑙𝑛𝐷𝑃𝑖𝑡 + 𝐵𝑇𝑇𝑖𝑡 + 𝑎𝑅𝐷𝑅 + 𝜖𝑖𝑡

where t represents year, i represents household, 𝐸 represents energy consumption or

electricity consumption per capita, H represents variables that reflect household

characteristics, including household size, and gender, age, education level of head of

household. 𝑙𝑛𝑀𝑃 represents log-transformed variable for energy or electricity price, T

represents time dummy variables, and D represents variables that reflect regional

characteristics.

The results indicate that increase in household size or the number of people

residing per household decreases per capita electricity consumption by approximately 26%.

77

3.4 Education, gender, homeownership, working age

population, and dwelling type

1) Lee et al. (2011)

The purpose of the study is to investigate the characteristics of energy use behaviors

and energy saving consciousness of multi-family housing residents. The energy usages

referred to in this study are electric energy, heating energy and water usages. This study is

conducted from a survey carried out in Seoul and Gyeonggi-do on 227 samples. The survey

questions are categorized into four sections: 24 questions on electric energy use behaviors,

4 questions on heating energy use behavior, 6 questions on water use behavior, and 8

questions on energy saving consciousness. The study performs t-tests, one-way ANOVA,

and simple statistical analyses in the study. Results of the study conclude that gender

difference does not significantly affect the residential energy consumption and households

with low education levels and older people have higher level of energy consciousness than

the other households, thus using lesser amount of residential energy.

78

2) Brounen et al. (2012)

Brounen et al. (2012) focus on gas and electricity consumption, investigating the

impact of the physical structure of home (i.e., the features of dwellings) on variations in

energy consumption. Then, the study compares the importance of these structural

characteristics with the demographic characteristics of households in the energy

consumption of Dutch households with the observations on 300,000 in the Netherlands

collected from January 2008 to December 2009. The dwelling types analyzed are

categorized into apartment, row house, semi-detached, detached, and corner. For analysis,

the study conducts OLS regressions using datasets including gas and electricity

consumption, period of construction, number of persons in household, age of the head of

households and elderly households., and household income. It is found that apartment,

semi-detached, and detached variables exhibit positive relationships with electricity

consumption per capita, whereas row house variables has negative relationship with

electricity consumption per capita.

79

3.5 Heating degree days (HDDs) and cooling degree days

(CDDs)

Numerous studies have confirmed that heating degree days and cooling degree days

serve as decent proxies for estimating the demand for residential electricity consumption.

In this section, a number of literature which investigate the effects of heating degree days

(HDDs) and cooling degree days (CDDs) on residential electricity consumption in overseas

countries and in Korea are discussed.

Heating degree days (HDDs) and cooling degree days (CDDs) in overseas countries

1) Eskeland and Mideksa (2010)

Eskeland (2010) uses panel data of 31 countries in Europe with selected years from

1994 to 2005 to investigate the relationships among the climatic factors and other factors

on the residential electricity consumption. For conceptual framework, the study uses

functions of residential electricity consumption before estimating the elasticities of demand

in residential electricity consumption. For empirical analysis, Arellano-Bond method, a

developed model from Generalized Method of Moment (GMM) method, is applied.

Arellano-Bond method uses time series dependent variables as dummy variables and has

benefits in addressing the issues of omitted variable bias when using the regional panel

80

data. The climatic variables used in the study are cooling degree days (CDDs) and heating

degree days (HDDs), where the values of CDDs and HDDs are calculated as 𝐶𝐷𝐷𝑖𝑡 =

𝑇𝑖𝑡 − 22 and 𝐻𝐷𝐷𝑖𝑡 = 18 − 𝑇𝑖𝑡, respectively, where 𝑇𝑖𝑡 is the average daily temperature

(in Celsius) of country i at time t.

The economy’s total revenue from the value with tax is used as an instrument

variable to address the potential problems of endogeneity and the measurement errors that

may arise from using the traditional income data, Further, panel data techniques with time-

and country-fixed effects are introduced to control for the possible omitted variable bias

problems that can occur when estimating the parameters that may be correlated with the

unobserved variables. The empirical results suggest that weather has a statistically

significant effect on electricity demand, with effects that are of plausible magnitude as in

the previous literature. In specific, a unit increase in HDD or CDD will raise electricity

consumption by 0.01% and 0.04%, respectively. In a simulation of climate change for the

next 100 years, holding other factors constant, the study finds that the demand for heating

will decrease in Northern Europe, while the demand for cooling will increase in Southern

Europe. In countries such as Cyprus, Greece, Italy, Malta, Spain, and Turkey, the net effects

of increased cooling outweigh the decreased heating consumption, whereas the opposite

holds in most of the European countries. Estimated elasticities with respect to income and

price are 0.8 and -0.2 respectively.

81

2) Lim et al. (2013)

Lim et al. (2013) consider not only the aging variables but also the temperature

variables such as cooling degree days (CDDs) and heating degree days (HDDs) to

investigate these impacts on the residential electricity consumption in Korea using

autoregressive distributed (ARDL) model and error correction model. Time series data

from 1966 to 2011 are used for analysis. The following equation represents the long-run

equilibrium function for the residential electricity demand:

𝑙𝑛𝑄𝑡 = 𝛼0 + 𝛼1𝑙𝑛𝑌𝑡 + 𝛼2𝑙𝑛𝑃𝑡 + 𝛼3𝑙𝑛𝐷𝑡 + 𝛼4𝑙𝑛𝐶𝐷𝐷 + 𝛼5𝑙𝑛𝐻𝐷𝐷 + 𝜖𝑡

which is derived from ARDL model. In the above equation, 𝛼𝑚 represents the long-run

elasticity parameters of the demand for residential electricity. Y, P, D, CDD, and HDD

represent the variables for real income per capita, real price for residential electricity, aging

index, cooling degree days, and heating degree days, respectively.

The study concludes from the estimation results that cooling degree days (CDDs)

significantly influence the residential electricity demand, but heating degree days (HDDs)

do not significantly influence the residential electricity demand. In specific, CDDs increase

by 1%, the residential electricity consumption will increase by 0.35% in the long-run and

0.11% in the short-run. The study finds that the effect of CDD on the residential electricity

demand in the long-run is greater than in the short-run, giving insights that may enhance

the policies for stable electricity supply. The study is significant for it analyzes both the

82

aging and the global warming effects due to climate change on the electricity demand.

3) Fullerton et al. (2012)

This study examines the residential electricity consumption in Seattle,

Washington, which is the largest metropolitan economy in the northwestern region of the

United States. The study uses time series data from Seattle City Light (SCL) annual reports

during the period of 1960 and 2007. For the variables used in the analysis, consumption

data or kilowatt hours per customer, which is calculated with total residential consumption

in megawatt hours (MHW) and the number of customers, are used. Also, real income per

capita, heating degree days (HDD), real average electricity price, population of the SCL

service area, and non-agricultural wage and salary employment in the SCL service aria are

used. A unit root test conducted with the residuals confirms that a cointegrating

relationships exists among the variables. Further, to test for consistency of the long-run

multipliers, CSUMQ and CUSUM tests are conducted. As results from both these tests,

coefficient stability for the outcomes are achieved. The study makes the following

conclusions from the empirical results: 1) Heating degree days (HDD) and residential

electricity consumption has a positive relationship, in which 1 percent increase in annual

heating degree days (HDD) increases residential electricity consumption by 0.303% in the

long-run and 0.338% in the short-run.

83

4) Na and Ryu (2000)

This study performs both qualitative and quantitative analyses to estimate the

demand for energy consumption by energy types and sectors. The study applies

autoregressive distributed lag (ARDL) model using monthly and quarterly data in Korea

from the first quarter in 1983 to the second quarter in 1999. For estimation, lagged

independent variables and lagged dependent variables are included to handle the problems

of autocorrelation and endogeneity. Also, quarterly dummy variables are included to reduce

the issues of seasonality. The estimated ARDL model is as follows:

𝑙𝑛𝐸𝑡 = 𝛼0 + ∑ 𝑎1𝑖ln 𝐸𝑡−𝑖𝑛𝑖=1 + ∑ 𝛼2𝑖𝑙𝑛𝑌𝑡−𝑖 + ∑ 𝛼3𝑖𝑙𝑛𝑃𝑡−𝑖

𝑛𝑖=0

𝑛𝑖=0 + α4𝐻𝐷𝐷 +

α5𝐶𝐷𝐷 + 𝐴𝑋

where Q, Y, P, HDD, CDD, and X represent the residential electricity demand, quarterly

gross domestic product (GDP), quarterly electricity price by sectors, heating degree days

(HDD), and cooling degree days (CDD), and other variables, respectively. The results

suggest that income elasticity and price elasticity for residential demand are found to be

positive with the coefficients of 0.17 and negative with the coefficient of -.042, respectively.

Also, HDD and CDD are found to exhibit positive relationships with residential electricity

demand with the coefficients of .0001 and .0002, respectively.

.

84

5) Keum et al. (2018)

The purpose of this study is to estimate the determinants of the demand for

electricity by using First Differenced Generalized Method of Moments (FD GMM) method

with the unbalanced panel data of 16 municipalities in Korea during the period of 1996-

2013. The following two models represent ARDL model and partial adjustment model,

respectively:

𝛥𝑙𝑛𝐸𝑖,𝑡 = 𝛣0 + 𝛣1𝛥𝑙𝑛𝐸𝑖,𝑡−1 + 𝛣2𝛥𝑃𝑖,𝑡 + 𝛣3𝛥𝑙𝑛𝑃(𝑖,𝑡−1) + 𝛣4𝛥𝑙𝑛𝑌𝑖,𝑡 + 𝛣5𝛥𝑙𝑛𝐴𝑖,𝑡

+ 𝛣6𝛥𝑙𝑛𝐶𝐷𝐷𝑖,𝑡 + 𝛣7𝛥𝑙𝑛𝐻𝐷𝐷𝑖,𝑡 + 𝛣8𝛥𝑙𝑛𝑇𝐷𝑖,𝑡 + 𝛥𝑢𝑖,𝑡

𝛥𝑙𝑛𝐸𝑖,𝑡 = 𝛣0 + 𝛣1𝛥𝑙𝑛𝐸𝑖,𝑡−1 + 𝛣2𝛥𝑃𝑖,𝑡 + 𝛣3𝛥𝑙𝑛𝑌𝑖,𝑡 + 𝛣4𝛥𝑙𝑛𝐴𝑖,𝑡 + 𝛣5𝛥𝑙𝑛𝐶𝐷𝐷𝑖,𝑡

+ 𝛣6𝛥𝑙𝑛𝐻𝐷𝐷𝑖,𝑡 + 𝛣7𝛥𝑙𝑛𝑇𝐷𝑖,𝑡 + 𝛥𝑢𝑖,𝑡

where E, Y, P, A, CDD, and HDD represent residential electricity consumption, real income

per capita, average real electricity price, aging population ratio, cooling degree days

(CDDs), and heating degree days (HDDs), respectively on region i and time series t. The

empirical results of the study suggest the following: 1) Partial adjust model vies better

results than ARDL model does in terms of statistical significance; 2) the elasticities of price

and income are found to be negative (-) and positive (+), respectively, and the long-run

elasticities are more elastic than the short-run elasticities; 3) the elasticities of CDDs and

aging population ratio are positive (+) and (negative), respectively

85

6) Kamerschen and Porter (2004)

This study estimates the total, residential, and industrial electricity demand using

partial adjustment approach and simultaneous equation approach in US during 1973-1998

and suggests that simultaneous equation approach is suitable in estimating the electricity

demand since they provide negative estimates for the price elasticity. In specific for the

estimating models, the study uses three-stage least squares (3SLS) model on the following

equations:

𝑙𝑛𝑄 = 𝛼1 + 𝛼2𝑙𝑛𝑃 + 𝛼3𝑙𝑛𝑋 + 𝛼4𝑙𝑛𝐺 + 𝛼6𝑙𝑛𝐶 + 𝜖

𝑙𝑛𝑃 = 𝐵1 + 𝐵2𝑙𝑛𝑄 + 𝐵3𝑙𝑛𝐿 + 𝐵4𝑙𝑛𝐹 + 𝐵5𝑙𝑛𝐾 + 𝑢

where Q, P, X, G, D, C, L, F, and K are average annual electricity sales per customer, real

marginal price of residential electricity, real annual GDP, real price for residential natural

gas, HDD, CDD, cost of labor, composite fuel cost, and cost of capital, respectively. Of the

parameter estimates on HDD and CDD, only CDD estimates are found to be statistically

significant with the coefficients of 0.145 and 0.137 for equation with CDD only and with

both CDD and HDD, respectively.

86

Table 3.3. Summary of literature on the relationship of heating degree days (HDDs) and

cooling degree days (CDDs) on residential electricity consumption

Author Region(s) Period Methodology Relationship with residential

electricity consumption (REC)

Eskelan

d and

Mideks

a

(2010)

31

countries in

Europe

1994-

2004

Arellano-

Bond method

HDD: positive, with coefficient

of 0.01

CDD: positive, with coefficient

of 0.04

Lim et

al.

(2013)

Korea 1966-

2011 ARDL

CDD: positive, with coefficients

of 0.35 in the long-run and 0.11

in the short-run

HDD has no significant effect

Fullerto

n et al.

(2012)

Seattle,

Washingto

n, US

1960-

2007

Cointegration

test

HDD: positive, with coefficients

of 0.303 in the long-run and

0.338 in the short-run

Na and

Ryu

(2000)

Korea 1983-

1999 ARDL

HDD: positive, with coefficient

of 0.001


of 0.002

Keum

et al.

(2018)

16

municipalit

ies in

Korea

1996-

2013 ARDL


of 0.018

HDD has no significant effect

Kamers

chen

and

Porter

(2004)

US 1973-

1998 3SLS

CDD: positive, with the

coefficients of 0.145 and 0.137,

respectively

HDD has no significant effects

87

3.6 Studies using panel estimation methods

1) Shin (2017)

Shin (2017) uses panel data for 16 regions in Korea to investigate the effects of

population aging on the residential electricity consumption. The study argues that using

regional panel data has some benefits over using time-series data for analysis.

First, using panel data for analysis can effectively solve the issues of multi-

collinearity of time series data. If time-series data is used for analyzing the effects of

population aging in a study, income and population aging variables may suffer from the

problems of multi-collinearity because economic growth leads population aging, and the

two conditions are usually happening at the same time. This multi-collinearity problem can

lead to inaccurate parameter estimations and cause severe loss of statistical power. Hence,

addressing this issue is important to avoid misleading interpretations for the analysis.

Second, using panel data can effectively investigate the changes in volatility of

electricity consumption without the usage of microeconomic data. If micro data are used

for analysis, multi-collinearity problem may be somewhat addressed since time series

correlation between income and aging is weak. However, it may be difficult to capture the

changes in income or price elasticities which cause changes in volatility of electricity

consumption over time.

88



household sector (i.e., residential electricity consumption) using panel data for 7

metropolitan areas and 9 provinces in Korea for the period of 2001-2010. For estimation,

the study excludes the province of Jeju considering the specialties of the islands. Several

results are presented in the study: First, household size, the ratio of the old-aged, female

economic participation rate, net population ratio, and the ratio of the apartment influence

household energy consumption. Second, electric energy consumption is affected by energy

price and the level of income. Third, electricity price, average household size, and the ratio

of expenditures on restaurants/hotels and the residential electricity consumption exhibit

negative relationships; the ratio of population aged over 65, the ratio of apartment house,

and income and the residential electricity consumption exhibit positive relationships.

Before estimating the coefficients of the panel data, assumptions of homogeneity

and no serial correlations must hold. Estimating the coefficients by fixed and random

effects models without verifying the former assumptions may result in biased and

inefficient estimates. Hence, Wooldridge test for time-series autocorrelation and Wald test

for heteroscedasticity are conducted. The null hypotheses of Wooldridge test and Wald test

are no autocorrelations among the panel data and homoscedastic characteristics among the

panel data sets, respectively. In the study, the test statistics reject the null hypotheses of

both Woolridge and Wald tests. In this case, panel Generalized Least Square (GLS) is a

better method for estimation. Hence, the study panel uses panel GLS under the assumptions

89

of heteroscedasticity and autocorrelations and Feasible Generalized Least Square (FGLS)

which is used when the structures of variance and covariance of the error terms are

unknown.

The study states that estimations via panel models have several benefits: First, with

panel models, regional differences and time-variant effects can be controlled to estimate

the effects of independent variables on the dependent variables. Second, although the cross-

sectional samples (in this case, the municipalities of Korea) are not enough in number, the

data can be integrated with the corresponding time series data, thus increasing the

significance levels and efficiency of the estimated parameters

90

Chapter 4. Data and research methods

4.1 Data

Dependent variable

1) Residential electricity consumption per capita (MWh/person)

The dependent variable used in this analysis is residential electricity consumption

per capita (MWh/person). The value is calculated by dividing the residential electricity

sales by the number of populations by region at a certain point of time. The equation is as

follows:

𝑅𝑒𝑠𝑖𝑑𝑒𝑛𝑡𝑖𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎

=𝑟𝑒𝑠𝑖𝑑𝑒𝑛𝑡𝑖𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑡𝑦 𝑠𝑎𝑙𝑒𝑠 𝑏𝑦 𝑟𝑒𝑔𝑖𝑜𝑛

𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑏𝑦 𝑟𝑒𝑔𝑖𝑜𝑛

Residential electricity data for 16 municipalities excluding Sejong City during

2000-2016 are collected from Electric Power Statistics Information System (EPSIS), and

population data are collected from Statics Korea (KOSTAT). Figure 4.1 shows that per

capita residential electricity consumption in all 16 municipalities in Korea has increased

sharply since 2000 until it reaches the peak in 2013, which is followed by the sudden drop

in 2014. The declined electricity consumption is rebounded back in 2015.

91

Figure 4.1. Trends of residential electricity consumption per capita (MWh/person) of 16 municipalities in Korea, 2000-2016

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.62000

2005

2010

2015

2003

2008

2013

2001

2006

2011

2016

2004

2009

2014

2002

2007

2012

2000

2005

2010

2015

2003

2008

2013

2001

2006

2011

2016

2004

2009

2014

2002

2007

2012

2000

2005

2010

2015

2003

2008

2013

2001

2006

2011

2016

2004

2009

2014

2002

2007

2012

2000

2005

2010

2015

2003

2008

2013

Busan ChungbukChungnam Daegu Daejeon entire Gangwon GwangjuGyeongbukGyeonggiGyeongnam Incheon Jeju Jeonbuk Jeonnam Seoul Ulsan

92

Independent variables

2) Residential electricity price (KRW/kWh)

Residential electricity price variable (KRW/kWh) is included as an independent

variable. The data is in real average retail price of residential electricity, which is attained

by using the data of average price of electricity considering the regional consumer price

index (CPI) with base year 2015=100. The study uses the average retail price for residential

electricity as the proxy for residential electricity price following Jo and Jang (2015), Ito

(2014), and Shin (2017). These studies have used the average retail price for residential

electricity to represent the residential electricity price under the current progressive billing

system in Korea. In specific, the average retail price is calculated as follows:

𝑅𝑒𝑠𝑖𝑑𝑒𝑛𝑡𝑖𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝑝𝑟𝑖𝑐𝑒

=𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑟𝑒𝑡𝑎𝑖𝑙 𝑝𝑟𝑖𝑐𝑒 𝑜𝑓 𝑟𝑒𝑠𝑖𝑑𝑒𝑛𝑡𝑖𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦

𝑟𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝑐𝑜𝑛𝑠𝑢𝑚𝑒𝑟 𝑝𝑟𝑖𝑐𝑒 𝑖𝑛𝑑𝑒𝑥 (𝐶𝑃𝐼) × 100

where

𝑟𝑒𝑡𝑎𝑖𝑙 𝑝𝑟𝑖𝑐𝑒 𝑓𝑜𝑟 𝑟𝑒𝑠𝑖𝑑𝑒𝑛𝑡𝑖𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦

=𝑟𝑒𝑡𝑎𝑖𝑙 𝑝𝑟𝑖𝑐𝑒 𝑓𝑜𝑟 𝑟𝑒𝑠𝑖𝑑𝑛𝑒𝑖𝑡𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝑏𝑦 𝑟𝑒𝑔𝑖𝑜𝑛

𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝑠𝑎𝑙𝑒𝑠 𝑣𝑜𝑙𝑢𝑚𝑒 𝑏𝑦 𝑟𝑒𝑔𝑖𝑜𝑛

93

The data for the average retail price of residential electricity and regional CPI are

attained from Electric Power Statistics Information System (EPSIS) and Korean Statistical

Information Service (KOSIS), respectively. The data cover 16 municipalities excluding

Sejong City during the period of 2000-2016. Figure 4.2 shows that the real price of

residential electricity in Korea has declined from 2000 to 2016 from the values at

approximately 160 KRW/kWh to 120 KRW/kWh, respectively.

94

Figure 4.2. Trends of residential electricity price (KRW/kWh) of 16 municipalities in Korea, 2000-2016

0

20

40

60

80

100

120

140

160

1802000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

Busan ChungbukChungnam Daegu Daejeon entire GangwonGwangjuGyeongbukGyeonggiGyeongnamIncheon Jeju Jeonbuk Jeonnam Seoul Ulsan

95

3) Personal income per capita (1000 KRW)

For income variable, the data for real personal income per capita in 1000 South

Korean Won (KRW) are used. Real personal income refers to disposable income or the total

annual disposable earnings of a citizen from all income sources including salaries and

wages, dividends and investment interest, and other ventures. Personal income well reflects

the socio-demographic characteristic for income since it captures the ability of the people

to spend money in their pockets to meet the consumption needs and to participate in the

society in which they live. Real personal income per capita is calculated from the nominal

personal income per capita with using the regional CPI with the base year 2015=100 as

follows:

𝑅𝑒𝑎𝑙 𝑝𝑒𝑟𝑠𝑜𝑛𝑎𝑙 𝑖𝑛𝑐𝑜𝑚𝑒 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎 =𝑛𝑜𝑚𝑖𝑛𝑎𝑙 𝑝𝑒𝑟𝑠𝑜𝑛𝑎𝑙 𝑖𝑛𝑐𝑜𝑚𝑒 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎

𝑟𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝐶𝑃𝐼

The data for nominal personal income per capita and regional CPI, which are used

to calculate the real personal income per capita, are obtained from Korean Statistical

Information Service (KOSIS). The annual panel data consist of the 16 municipalities

excluding Sejong City from 2000 to 2016. Figure 4.3 presents personal income capita by

region. All municipalities in Korea have experienced gradual increase in personal income

per capita since 2000; Seoul and Ulsan are shown to be the cities with first and second

highest income per capita among all municipalities in Korea.

96

Figure 4.3. Trends of personal income per capita (1000 KRW) of 16 municipalities in Korea, 2000-2016

0

5000

10000

15000

20000

25000

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

Busan ChungbukChungnamDaegu Daejeon entire GangwonGwangjuGyeongbukGyeonggiGyeongnamIncheon Jeju Jeonbuk Jeonnam Seoul Ulsan

97

4) Share of population aged 65 and above (%)

The share of population aged 65 and above is used as an independent variable

which represents population aging, a key socio-demographic factor analyzed in this

research. Numerous studies have used the share of different age groups to explain how the

people in certain age groups have varying influences on the residential electricity

consumption; in specific, they have used the share of population aged 65 and above as a

proxy variable for population aging (Pessanha and Leon, 2015; Kelly, 2011; Zhou, 2016;

Noh and Lee, 2013). Collected from Korean Statistical Information Service (KOSIS), the

data is the annual panel data of 16 municipalities excluding Sejong City during the period

of 2000-2016. Figure 4.4 shows that Jeonnam has the highest share of population aged 65

and above, and Ulsan, the lowest. This may be explained with the regional characteristics

of Jeonnam and Ulsan, in which the former has large portions with pastoral areas and the

latter has huge industrial hub operating with many plants, factories, facilities, et cetera. In

general, the elder populations tend to live in the rural areas (e.g., Jeonnam), and the younger

populations tend to live in the industrial (e.g., Ulsan) and metropolitan (e.g., Seoul) cities.

98

Figure 4.4. Trends of the share of population aged 65 and above (%) of 16 municipalities in Korea, 2000-2016

0

5

10

15

20

252000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


99

5) Working age population (%)

The independent variable, the share of population aged in the range of 15-64 is

used to represent the working age population in Korea during 2000-2016. OECD defines

working age population as those who are aged 15 to 64 and considers it as a typical indicator

for potential labor supply. The data is the annual panel data of 16 municipalities excluding

Sejong City during the period of 2000-2016, collected from Korean Statistical Information

Service (KOSIS). Figure 4.5 shows that currently, the trends of working age populations

are decreasing in almost all municipalities in Korea starting from 2010. Also, it is shown

that Jeonnam, Seoul, and Ulsan have the high shares of working populations compared to

other cities in Korea, and Jeonbuk, the lowest.

100

Figure 4.5. Trends of working age population (%) of 16 municipalities in Korea, 2000-2016

0

10

20

30

40

50

60

70

80

902000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


101

6) Household size

Household size is defined as the number of persons residing in a household in a

region. This often the time represents the number of persons that are financially responsible

within a family. The values are calculated by dividing the number of populations by the

number of households in a region at a specific time as the following equation:

𝐴𝑣𝑒𝑟𝑎𝑔𝑒 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑠𝑖𝑧𝑒 =𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑠 𝑏𝑦 𝑟𝑒𝑔𝑖𝑜𝑛

𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠 𝑏𝑦 𝑟𝑒𝑔𝑖𝑜𝑛

The data, collected from Korean Statistical Information Service (KOSIS), is the

annual panel data of 16 municipalities excluding Sejong City during the period of 2000-

2016. Figure 4.6 represents the trends of average household size in Korea from 2000 to

2016. It is clearly shown in the graph that the household size has been rapidly decreasing

with the values of approximately 3~3.3 in 2000 to 2.25~2.5 in 2016. Gangwon, Chungbuk,

Chungnam, Gyeongbuk, Jeonnam, Jeonbuk, and Seoul have the low household size

relatively to other municipalities such as Deagu, Daejeon, Gwangju, Gyeonggi, Incheon,

and Ulsan.

102

Figure 4.6. Trends of household size of 16 municipalities in Korea, 2000-2016

0

0.5

1

1.5

2

2.5

3

3.52000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


103

7) Share of population with high education level (%)

Share of populations with high education level consists of the share of populations

with degrees in college or university level or with higher degrees (such as Master’s or

PhD’s degrees). Variable related to education level is used as one of the socio-demographic

factors focused on this study. The data are obtained from Korean Statistical Information

Service (KOSIS), Population and Housing Census. KOSIS conducts surveys and provides

the data once every five years (i.e., 2000, 2005, 2010, and 2015); in the study, the data used

in the analysis are approximated through interpolation methods. Figure 4.7 shows that the

shares of population with higher education level are increasing for all the municipalities in

Korea. In specific, Seoul has the highest percentage of population with high education level,

and Jeonnam, the lowest.

104

Figure 4.7. Trends of the share of population with degrees in higher education (%) of 16 municipalities in Korea, 2000-2016

0

0.2

0.4

0.6

0.8

1

1.22000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


105

8) Heating degree days (HDDs)

Heating degree days (HDDs) are measurements designed to quantify the demand

for energy needed to heat a building. In specific, heating degree days are measured relative

to balance temperature, the outside temperature above which a building needs no heating.

Usually, the balance temperature differs by countries; in Korea, the balance temperature of

heating degree days is 18℃. Numerous ways exist in which heating degree days can be

calculated. One popular method for approximating heating degree days is to take the

average temperature on any given day and subtract it from the given balance temperature

(Lim et al. 2013). Conceptually, the higher the value of heating degree days means the

colder the climate and the higher the fuel cost to heat a building.

𝐻𝑒𝑎𝑡𝑖𝑛𝑔 𝐷𝑒𝑔𝑟𝑒𝑒 𝐷𝑎𝑦𝑠 = 𝛴[(18℃) − (𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒 𝑜𝑓 𝑎 𝑑𝑎𝑦)]

In this study, the estimated values for heating degree days are obtained by region

during the months on a seasonal basis from October to April. The balance temperature for

heating degree days is set as 18℃ following Lim et al. (2013). The data are obtained from

National Climate Data Service System (NCDSS) provided by Korean Meteorological

Administration (KMA). It is shown in Figure 4.8 that overall trends of HDDs in all

municipalities are shown to be gradually declining with large fluctuations. Also, it is

presented that Busan, Daegu, Ulsan, and Gyeongnam have relatively low HDDs, and

Chungbuk and Gyeonggi have relatively high HDDs than other municipalities.

106

Figure 4.8. Trends of heating degree days (HDDs) of 16 municipalities in Korea, 2000-2016

0

20

40

60

80

100

1202000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


107

9) Cooling degree days (CDDs)

Cooling degree days (CDDs) are measurements designed to quantify the demand

for energy needed to cool a building. In specific, cooling degree days are measured relative

to “balance temperature,” the outside temperature above which a building needs no cooling

an average daily-mean temperature threshold for human thermal comfort. Usually, the

balance temperature differs by countries and is commonly set to a value between 18~24℃

(Glickman, 2000; MRCC, 2007); in Korea, the “balance temperature” of HDDs is in the

range of 18~24℃. Numerous ways exist in which CDDs can be calculated. One popular

method for approximating CDDs is to subtract the balance temperature from the average

temperature on any specific day (Lim et al. 2013). In concepts, the higher the value of

cooling degree days means the hotter the climate and the higher the fuel costs to cool a

building.

𝐶𝑜𝑜𝑙𝑖𝑛𝑔 𝐷𝑒𝑔𝑟𝑒𝑒 𝐷𝑎𝑦𝑠 = 𝛴[(18℃) − (𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒 𝑜𝑓 𝑎 𝑑𝑎𝑦)]

In this study, the approximated values for cooling degree days are obtained during

the months on a seasonal basis from May to September. Following the values used in Lim

et al. (2013), the balance temperature is set as 18℃. The data are obtained from National

Climate Data Service System (NCDSS) provided by Korean Meteorological

Administration (KMA).

108

Figure 4.9. Trends of cooling degree days (CDDs) of 16 municipalities in Korea, 2000-2016

0

5

10

15

20

25

30

35

40

45

502000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


109

10) Attached house to apartment ratio (%)

Attached house to apartment ratio refers to the ratio of multi-family residences and

townhouses to apartments in Korea. This variable is representative of one of the factors of

dwelling types used in this study. The data are obtained from Korean Statistical Information

Service (KOSIS) and consist of the unbalanced panel data of 16 municipalities excluding

Sejong City in Korea in the years 2006, 2008, 2010, 2012, 2014, and 2016. It is illustrated

in Figure 4.10 that the shares of attached houses to apartments are decreasing throughout

the years for all 16 municipalities in Korea. This can be interpreted that the people

nowadays tend to reside more in the apartments than the attached houses including

townhouse and multi-family residences.

110

Figure 4.10. Trends of attached house to apartment ratio (%) of 16 municipalities in Korea, 2000-2016

0

0.5

1

1.5

2

2.5

3

3.5

42000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


111

11) Detached house to apartment ratio

Detached house to apartment ratio refers to the ratio single-detached to apartments

in Korea. This variable represents one of the factors of dwelling types used in the study.

The data are obtained from Korean Statistical Information Service (KOSIS) and consist of

the unbalanced panel data of 16 municipalities excluding Sejong City in Korea in the years

2006, 2008, 2010, 2012, 2014, and 2016. As shown in Figure 4.11, the shares of detached

houses to apartments are decreasing throughout the years for all 16 municipalities in Korea.

This can be interpreted that the people nowadays tend to reside more in the apartments than

in the detached houses.

112

Figure 4.11. Trends of detached house to apartment ratio (%) of 16 municipalities in Korea, 2000-2016

0

0.5

1

1.5

2

2.5

3

3.5

4

4.52000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


113

12) Female to male ratio

Female to male ratio is one of the socio-demographic factors considered in the

study. The data are obtained from Korean Statistical Information Service (KOSIS) and

Ministry of the Interior and Safety using Resident Registry Population Census. Figure 4.12

presents the trends of the female to male ratio in 16 municipalities excluding Sejong City

in Korea during 2000-2016.

The 6 municipalities with female to male ratio greater than 100 are Busan, Daegu,

Gwangju, Jeonbuk, Jeonnam, and Seoul; this means that these cities have more female

citizens than male citizens residing. The other 10 municipalities, Chungbuk, Chungnam,

Daejeon, Gangwon, Gyeongbuk, Gyeonggi, Gyeongnam, Incheon, Jeju, and Ulsan, have

female to male ratio smaller than 100; this means that the populations consist of more male

citizens than female citizens in these municipalities. The directions to which the trends of

female and male ratio are increasing or decreasing vary by region: the 8 municipalities (i.e.,

Busan, Daegu, Daejeon, Gwangju, Gyeonggi, Incheon, Jeonbuk, and Seoul) have

increasing trends and the 8 municipalities (i.e., Chungbuk, Chungnam, Gangwon,

Gyeongbuk, Gyeongnam, Jeju, Jeonnam, and Ulsan) have decreasing trends.

114

Figure 4.12. Trends of female to male ratio (%) of 16 municipalities in Korea, 2000-2016

88

90

92

94

96

98

100

102

104

1062000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


115

13) Homeownership rate (%)

The data for homeownership rate for 16 municipalities in Korea during the period

of 2000 to 2016 are attained from Korea Housing Surveys published by Ministry of Land,

Infrastructure, and Transport (MOLIT). Homeownership rate is defined as the percentage

of homes that are owned by their occupants, and this is one of the socio-demographic

factors considered in the analysis. Figure 4.13 presents the trends of homeownership ratio

of 16 municipalities in Korea from 2000 to 2016. It is shown that Seoul and Jeonnam have

the lowest and highest homeownership rates, respectively. The low homeownership rate in

Seoul can be dedicated to its exceptionally high housing prices compared to the housing

prices in other cities. This situation can make people who do not have the financial ability

to purchase a house to live on rent instead. Also, the low homeownership rate in Korea can

be explained by the investors owning numerous houses but not residing for the speculative

and pecuniary purposes.

116

Figure 4.13. Trends of homeownership ratio (%) of 16 municipalities in Korea, 2000-2016

0

10

20

30

40

50

60

70

802000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016

2005

2011

2000

2006

2012

2001

2007

2013

2002

2008

2014

2003

2009

2015

2004

2010

2016


117

Table 4.1. Definition of the variables (16 municipalities in Korea excluding Sejong City, 2000-2016)

Variable Data Unit Reference Period Regions Classifications

Dependent 𝑦𝑒𝑙𝑒𝑐 Residential electricity

consumption per capita MWh/person

KOSIS,

EPSIS 2000-2016

16

municipalities

Independent

𝑥𝑝𝑟𝑖𝑐𝑒 Residential electricity

price KRW/kWh

KOSIS,

EPSIS 2000-2016

16

municipalities Others

𝑥𝑖𝑛𝑐𝑜𝑚𝑒 Personal income per

capita 1000 KRW KOSIS 2000-2016

16

municipalities Socio-demographic

𝑥𝑓𝑒𝑚𝑎𝑙𝑒 Female to male ratio % KOSIS 2000-2016 16


𝑥𝑤𝑜𝑟𝑘 Share of population aged

15~64 % KOSIS 2000-2016

16


𝑥𝑜𝑙𝑑 Share of population aged

65 years and above % KOSIS 2000-2016

16


𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒 Household size Total population /

# of household KOSIS 2000-2016

16


𝑥𝑒𝑑𝑢𝑐 Share of population with

high education level % KOSIS 2000-2015

16


𝑥𝑐𝑑𝑑 Cooling degree days

(CDDs) °C day NCDSS 2000-2016

16

municipalities Climatic

𝑥ℎ𝑑𝑑 Heating degree days

(HDDs) °C day NCDSS 2000-2016

16

municipalities Climatic

𝑥𝑎𝑡𝑡𝑎𝑐ℎ𝑒𝑑 Attached house to

apartment ratio % KOSIS

2006, 2008,

2010, 2012,

2014, 2016

16


𝑥𝑑𝑒𝑡𝑎𝑐ℎ𝑒𝑑 Detached house to

apartment ratio % KOSIS

2006, 2008,

2010, 2012,

2014, 2016

16


𝑥ℎ𝑜𝑚𝑒𝑜𝑤𝑛𝑒𝑟𝑠ℎ𝑖𝑝 % of homes owned by

their occupants % MOLIT

2006, 2008,

2010, 2012,

2014, 2016

16


118

Table 4.2. Descriptive statistics of the data

Variable # of observations Mean Standard

deviation Minimum Maximum

Dependent 𝑦𝑒𝑙𝑒𝑐 272 1.082 0.159 0.763 1.351

Independent

𝑥𝑝𝑟𝑖𝑐𝑒 272 137.187 12.242 119.972 163.949

𝑥𝑖𝑛𝑐𝑜𝑚𝑒 272 14353.94 2032.037 10490.780 20266.850

𝑥𝑓𝑒𝑚𝑎𝑙𝑒 272 99.387 1.722 94.008 103.630

𝑥𝑤𝑜𝑟𝑘 272 71.308 3.149 63.800 77.000

𝑥𝑜𝑙𝑑 272 11.069 3.783 4.000 20.900

𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒 272 2.679 0.228 2.240 3.222

𝑥𝑒𝑑𝑢𝑐 256 0.628 0.143 0.310 0.992

𝑥𝑐𝑑𝑑 272 31.183 3.379 22.625 43.500

𝑥ℎ𝑑𝑑 272 78.188 12.757 46.300 102.800

𝑥𝑎𝑡𝑡𝑎𝑐ℎ𝑒𝑑 96 2.779 0.365 2.223 3.666

𝑥𝑑𝑒𝑡𝑎𝑐ℎ𝑒𝑑 96 1.071 0.568 0.371 3.843

𝑥ℎ𝑜𝑚𝑒𝑜𝑤𝑛𝑒𝑟𝑠ℎ𝑖𝑝 96 58.979 6.891 40.200 73.400

119

Table 4.3. Results from Pearson’s correlation analysis

𝒚𝒆𝒍𝒆𝒄 𝒙𝒊𝒏𝒄𝒐𝒎𝒆 𝒙𝒑𝒓𝒊𝒄𝒆 𝒙𝒇𝒆𝒎𝒂𝒍𝒆 𝒙𝒘𝒐𝒓𝒌 𝒙𝒐𝒍𝒅 𝒙𝒉𝒐𝒖𝒔𝒆𝒉𝒐𝒍𝒅𝒔𝒊𝒛𝒆 𝒙𝒅𝒆𝒕𝒂𝒄𝒉𝒆𝒅 𝒙𝒂𝒕𝒕𝒂𝒄𝒉𝒆𝒅 𝒙𝒄𝒅𝒅 𝒙𝒉𝒅𝒅 𝒙𝒆𝒅𝒖𝒄 𝒙𝒉𝒐𝒎𝒆𝒐𝒘𝒏

𝒚𝒆𝒍𝒆𝒄 1.000

𝒙𝒊𝒏𝒄𝒐𝒎𝒆 0.740* 1.000

𝒙𝒑𝒓𝒊𝒄𝒆 -0.911* -0.621* 1.000

𝒙𝒇𝒆𝒎𝒂𝒍𝒆 0.027 -0.176* -0.013 1.000

𝒙𝒘𝒐𝒓𝒌 0.384* 0.622* -

0.162* -0.049 1.000

𝒙𝒐𝒍𝒅 0.344* -0.066 -

0.503* 0.217*

-

0.638* 1.000

𝒙𝒉𝒐𝒖𝒔𝒆𝒉𝒐𝒍𝒅 -0.819* -0.468* 0.876* -0.110* 0.080 -

0.741* 1.000

𝒙𝒅𝒆𝒕𝒂𝒄𝒉𝒆𝒅 -0.542* -0.393* 0.242* -0.068 -

0.669* 0.417* -0.162 1.000

𝒙𝒂𝒕𝒕𝒂𝒄𝒉𝒆𝒅 -0.717* -0.341* 0.925* -0.089 0.050 -

0.579* 0.875* 0.096 1.000

𝒙𝒄𝒅𝒅 0.278* 0.293* -

0.189* 0.202* 0.189* -0.029 -0.083 -0.107 -0.196 1.000

𝒙𝒉𝒅𝒅 0.106 -0.116 0.001 -0.111 -0.018 0.080 -0.146* -0.334* -0.135 -0.106 1.000

𝒙𝒆𝒅𝒖𝒄 0.663* 0.718* -

0.552* 0.253* 0.714*

-

0.261* -0.332* -0.406* -0.151 0.366*

-

0.057 1.000

𝒙𝒉𝒐𝒎𝒆𝒐𝒘𝒏 -0.236* -0.474* -0.115 -0.231* -

0.722* 0.674* -0.274* 0.264* -0.168 -0.150

-

0.137 -0.819* 1.000

120

Table 4.4. Compounded annual growth rate (CAGR) of the variables from 2000 to 2016 by regions

Region REC per capita Personal income

per capita

Real electricity

price

Working age

population Aged population Household size HDD CDD

Entire 2.95% 1.97% -1.75% 0.15% 3.86% -1.37% 0.55% -0.53%

Busan 2.97% 2.11% -1.86% -0.13% 5.74% -1.69% 0.54% -0.42%

Chungbuk 2.62% 2.08% -1.67% 0.21% 2.70% -1.64% 0.53% -0.56%

Chungnam 2.66% 2.00% -1.74% 0.20% 1.83% -1.57% 0.49% -0.69%

Daegu 2.42% 2.00% -1.74% 0.06% 5.06% -1.40% 0.23% -0.16%

Daejeon 2.58% 2.47% -1.69% 0.22% 4.49% -1.45% 0.96% -0.65%

Gangwon 2.83% 2.12% -1.68% 0.03% 3.50% -1.77% 0.07% -0.52%

Gwangju 2.75% 2.04% -1.76% 0.21% 4.60% -1.49% 0.80% -0.88%

Gyeongbuk 2.98% 1.86% -1.61% 0.09% 2.79% -1.61% 0.49% -0.48%

Gyeonggi 2.24% 1.74% -1.82% 0.38% 3.95% -1.07% 0.83% -0.72%

Gyeongnam 2.89% 2.02% -1.78% 0.24% 2.83% -1.49% 0.55% -0.46%

Incheon 2.77% 2.37% -1.75% 0.35% 4.37% -1.25% 0.02% -0.21%

Jeju 2.80% 1.73% -1.67% 0.14% 3.35% -1.49% 0.87% -0.89%

Jeonbuk 2.93% 2.16% -1.73% 0.03% 3.07% -1.68% 0.67% -0.63%

Jeonnam 3.07% 2.19% -1.66% -0.04% 2.82% -1.64% 0.70% -0.43%

Seoul 2.21% 1.71% -1.93% -0.02% 5.56% -1.28% 0.94% -0.44%

Ulsan 3.18% 1.70% -1.87% 0.46% 5.20% -1.39% -0.02% -0.31%

121

4.2 Methodology

Panel estimation methods are used for empirical analysis in this research since

using panel data has some benefits over other estimation methods. First, using panel data

for analysis can effectively solve the multi-collinearity problems that may occur when

using time series data. The multi-collinearity problem can lead to inaccurate parameter

estimations and cause severe loss of statistical power. Hence, addressing this issue is

important to avoid misleading interpretations and conclusions in the analysis. Second,

using panel data can effectively investigate the changes in volatility of electricity

consumption without using the microeconomic data. If microeconomic data are used for

analysis, multi-collinearity problem may be somewhat addressed by easing the problems

of time series correlations among the explanatory variables. However, it may be difficult

to capture the changes in income or price elasticities which can cause the changes in

volatility of electricity consumption over time. With these advantages, the study applies

panel cointegration methods and panel estimations for the empirical analysis.

122

4.2.1 Panel cointegration approach

4.2.1.1 Panel unit root test: Testing the stationarity of the panel series

Famous studies on panel unit roots test (Maddala and Wu, 1999; Im et al., 2003;

Levin et al., 2002) have suggested that the tests are not robust in the presence of cross-

sectional dependence among the variables. In order to diagnose this problem, cross-

sectional dependence test for panel data proposed by Pesaran (2004) is performed.

Pesaran’s (2004) cross-sectional test is based on the average of the correlation coefficients

of the ordinary-least-squares (OLS) residuals, obtained from augmented Dickey-Fuller

(ADF) regressions. The null hypothesis is cross-sectional independence and is asymptotic

distribution as a two-tailed standard normal distribution (Boubtane et al., 2012). If the p-

values are small enough to reject the null hypothesis of cross-sectional independence

among the panel series, panel unit root test must be employed to handle the problem of

cross-sectional dependence.

After proving the existence of cross-section dependence among the panel series,

the next step to take is determining whether the variables are integrated of the same order.

If a variable is integrated of order k: I(k), the variable is said to be integrated of order k and

thus is non-stationary. The stationary variable can be defined as I(0), and a non-stationary

variable, as I(1); an I(1) variable must be first-differenced to achieve stationarity (Harris

and Sollis, 2013). When conducting panel cointegration test, addressing the issues of non-

stationary of panel time-series data is crucial because using standard regression methods,

123

such as OLS, on non-stationary time-series with a unit root can produce the problem of

spurious regression, making the results statistically insignificant.

Hence in this study, panel unit root test developed by Pesaran (2007) is conducted.

Pesaran (2007) proposes a simple procedure for testing the unit roots in dynamic panel

series addressing the issues of cross-sectional dependence and serial correlations with error

terms. Let 𝑦𝑖𝑡 be the observations on the ith cross-section unit at time t that is generated

according to a simple dynamic panel data model:

𝑦𝑖𝑡 = (1 − 𝜌𝑖)𝜇𝑖 + 𝜌𝑖𝑦𝑖𝑡−1 + 𝑢𝑖𝑡 (1)

where yi0 has a given density function with a finite mean and variance, and uit is an error

term with the single-factor structure

𝑢𝑖𝑡 = 𝛾𝑖𝑓𝑡 + 𝜖𝑖𝑡 (2)

where 𝑓𝑡 is the unobserved common effect and 𝜖𝑖𝑡 is the idiosyncratic error.

124

Consider the following assumptions necessary for attaining the consistency of the

panel unit root tests: First, the idiosyncratic shocks, 𝜖𝑖𝑡 are independently distributed both

across i and t with mean zero and variance 𝜎𝑖2 and finite fourth order moment; second, 𝑓𝑡

is serially uncorrelated with mean zero and a constant variance σ𝑓2 which is set equal to

one without loss of generality and finite fourth order moment; third, 𝑒𝑖𝑡, 𝑓𝑡, and 𝛾𝑖 are

independently distributed for all i. More conveniently, eq. (1) can be written as the

following:

𝛥𝑦𝑖𝑡 = (1 − 𝜌𝑖)𝜇𝑖 − (1 − 𝜌𝑖)𝑦𝑖𝑡−1 + 𝜆𝑖𝑡𝑓𝑡 + 𝜖𝑖𝑡 (3)

where 𝛥𝑦𝑖𝑡 = 𝑦𝑖𝑡 − 𝑦𝑖𝑡−1. Pesaran’s (2007) unit root tests the following null hypothesis:

𝐻0: 𝜌𝑖 = 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖

against the alternative hypotheses:

Ha: {

𝜌𝑖 < 1 𝑓𝑜𝑟 𝑖 = 1, … , 𝑁1

𝜌𝑖 = 1 𝑓𝑜𝑟 𝑖 = 𝑁1, … , 𝑁

Such that N1

𝑁→ 𝑘 as 𝑁 → ∞ with 0 < 𝑘 ≤ 1.

125

4.2.1.2 Panel cointegration test: Testing the convergence of panel series to long-

run equilibrium

If the variables are found to be non-stationary with unit roots, panel cointegration

test can be applied to investigate whether the series converge to the long-run equilibrium.

In the analysis, panel cointegration tests developed by Pedroni (1999, 2004) and Kao (1999)

are performed.

As an extension to the panel framework of Engle-Granger, Pedroni’s (1999, 2000)

heterogenous panel cointegration test considers regressing the variables along with cross-

section specific intercepts and tests the null hypothesis of no cointegrating relationships.

The two sets of test statistics for alternative hypotheses are proposed: 1) a test based on the

within-dimension approach, with which the four panel cointegration statistics are

calculated (i.e., panel v-statistic, panel rho-statistic, panel PP-statistics, and panel ADF-

statistic); 2) a test based on between-dimension approach, with which the four group mean

panel cointegration statistics are calculated. A consensus of these statistics is often

interpreted as a sign in favor of cointegration, but having majority of these statistics

rejecting the hypothesis is sufficient to confirm the presence of cointegration relationships

(Liddle, 2011).

126

Kao is an Engle-Granger based test which follows the same basic approach as

Pedroni cointegration test, but specifies the homogenous coefficients and the cross-section

specific intercepts on the firs-stage regressors. Under the null hypothesis of no

cointegration, the augmented version of test statistic

𝐴𝐷𝐹 =𝑡�̂�+

√6𝑁�̂�𝜐2�̂�0𝜐

√�̂�0𝜐2 /(2�̂�𝜐

2)+3�̂�𝜐2/(10�̂�0𝜐

2 )

(4)

converge to N(0,1) asymptotically.

127

4.2.1.3 Fully Modified Ordinary Least Squares (FMOLS) and Dynamic

Ordinary Least Squares (DOLS): Estimating long-run cointegrating

vectors

Once the cointegrating relationships of the variables in panel data sets are found,

we can estimate the long-run cointegrating vectors and find the signs or the values of

coefficients on the variables. The long-run parameters are estimated using fully modified

OLS (FMOLS) and dynamic OLS (DOLS) estimators developed by Pedroni (2001).

FMOLS is a non-parametric approach, while DOLS is a parametric approach, where lagged

first-differenced terms are explicitly estimated (Harris and Sollis, 2003). Both FMOLS and

DOLS address the issues of endogeneity and serial correlation. There are conflicting views

as to whether FMOLS or DOLS is preferred, with one side that views FMOLS to be more

biased than DOLS and the other side that regards DOLS has a smaller distortion in size

than FMOLS (Kao and Chiang, 2000; Pedroni, 2001). Hence in this study, both FMOLS

and DOLS are applied to test for statistical significance of the coefficients. The regression

model to be estimated with FMOLS and DOLS is as follows:

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = 𝛽0 + 𝛽1 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 + 𝛽2 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)𝑖𝑡

+ 𝛽3 𝑥𝑜𝑙𝑑𝑖𝑡+

𝛽4 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 𝛽5 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡 + 𝛽6 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 𝑒𝑖𝑡 (5)

where i (i = 1, …, 16) indicates 16 municipalities in Korea, t (t = 1, …, 16) indicates annual

period from 2000 to 2016, 𝑒𝑖𝑡 is an idiosyncratic error term. The variables 𝑙𝑛(𝑦𝑒𝑙𝑒𝑐) ,

128

𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) , 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) , 𝑥𝑜𝑙𝑑 , 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒) , 𝑙𝑛(𝑥𝑐𝑑𝑑) , and 𝑙𝑛(𝑥ℎ𝑑𝑑) represent

logged term of residential electricity consumption per capita, logged term of personal

income per capita, percentage of population aged 65 and above, logged term of household

size, logged term of cooling degree days (CDDs), and logged term of heating degree days

(HDDs).

The FMOLS and DOLS estimators are given as the following:

�̂�𝐹𝑀𝑂𝐿𝑆 =1

𝑁𝛴𝑖=1

𝑁 (𝛴𝑡=1𝑇 (𝑥𝑖𝑡 − �̅�𝑖)2)−1(𝛴𝑡=1

𝑇 (𝑥𝑖𝑡 − �̅�𝑖)𝑦𝑖𝑡∗ − 𝑇𝛾𝑖) (6)

�̂�𝐷𝑂𝐿𝑆 =1

𝑛𝛴𝑖=1

𝑛 (𝛴𝑡=1𝑇 𝑧𝑖𝑡𝑧𝑖𝑡

′ )−1(𝛴𝑡=1𝑇 𝑧𝑖𝑡𝑦𝑖𝑡) (7)

129

4.2.2 Tests for fixed effects and random effects models

4.2.2.1 Breusch-Pagan Lagrange Multiplier (LM) test for random effects

and Wald test for fixed effects

The model for panel data estimations can be chosen based on the results of testing

the assumptions about the error terms 𝑢𝑖 in a panel linear regression model. Assume the

following linear panel regression model which considers the heterogeneity properties of

the panel data series, i:

𝑦𝑖𝑡 = 𝛼 + 𝛽𝑥𝑖𝑡 + 𝑢𝑖 + 𝑒𝑖𝑡 (1)

where the model includes time-invariant error term 𝑢𝑖, an error term which does not reflect

the changes over time but reflects the changes in regional characteristics, and time-variant

error term 𝑒𝑖𝑡, an error term which reflects both the changes over time and the changes in

regional characteristics. Fixed effects model considers the error term 𝑢𝑖 as the parameter

to be estimated, and random effects model assumes that the expected value of error term

𝑢𝑖 equals zero with consistent variance, 𝑢𝑖~𝑁(0, 𝜎𝑒2). In contrast, pooled ordinary least

squares (OLS) model assumes that there are no error terms that reflect heterogeneity in

regions, ignoring the characteristics of panel data, and estimates the model with the typical

OLS method.

The first step to choosing the suitable model for panel estimations is by comparing

both fixed effects model and random effects model to pooled OLS with Wald test and

130

Breusch-Pagan LM test.

The Wald test tests the significance of the fixed effect model. To determine

whether is suitable to select fixed effects model for estimation, Wald test conducts

hypothesis testing on 𝐻0: 𝑎𝑙𝑙 𝑢𝑖 = 0. Failure to reject the null hypothesis means that the

constant terms in the model becomes 𝛼 , and thus is acceptable to use pooled OLS

estimation which does not consider the heterogeneity across the regions. Rejection of the

null hypothesis means that estimating with fixed effects model is a suitable method.

Breusch-Pagan LM test tests the significance of the random effects model. To

confirm whether random effects model is suitable for estimating the panel data, the test

conducts hypothesis testing on 𝐻0: 𝑣𝑎𝑟(𝑢𝑖) = 𝜎𝑢2 = 0. Failure to reject the null hypothesis

means that pooled OLS is suitable method to estimate the panel data since the variance of

the error terms as a whole simply becomes (σu2 + 𝜎𝑒

2) = 𝜎𝑒2 ; rejection of the null

hypothesis implies the presence of heteroscedastic variances of the error terms and means

that the estimation with random effects model which considers the regional characteristics

of a panel data is an appropriate method.

131

4.2.2.2 Hausman test

Panel data estimation method is chosen depending on whether the error term 𝑢𝑖

is presented as fixed effects or random effects. Consider the following linear panel

regression model:

𝑦𝑖𝑡 = (𝑎 + 𝑢𝑖) + 𝛽𝑥𝑖𝑡 + 𝑒𝑖𝑡 (2)

From eq. (2), fixed effects model considers (𝑎 + 𝑢𝑖) as a fixed parameter for

each panel data object, and random effects model considers (𝑎 + 𝑢𝑖) as a random variable

following random distribution with (𝑎 + 𝑢𝑖)~𝑁(𝛼, 𝜎𝑢2). It is possible to select the fitting

model for panel estimations between fixed and random effects models by conducting

Hausman test. The null and alternative hypotheses for Hausman test are as follows:

𝐻0: 𝑐𝑜𝑣(𝑥𝑖𝑡 , 𝑢𝑖) = 0 (3)

𝐻1: 𝑐𝑜𝑣(𝑥𝑖𝑡 , 𝑢𝑖) ≠ 0 (4)

Failure to reject the null hypothesis means that random effects model is suitable

for estimating the panel data; rejection of the null hypothesis means that fixed effects model

is more efficient model to obtain consistent estimates in panel series.

132

4.2.2.3 Modified Wald test for heteroscedasticity and Wooldridge test for auto-

correlations within panel units

Since panel data has the characteristics of both cross-sectional and time-series

data, the problems of heteroscedasticity among panel data sets and autocorrelations among

the error terms may occur. Estimating without addressing these issues can produce

inconsistent and inefficient estimates. Hence, it is of utmost importance to detect and

address these issues prior to conducting panel data estimations (Wooldridge, 2010).

First, modified Wald test tests for the panel-level heteroscedasticity among the

variances of all cross-sectional unit i. The null hypothesis for the test is 𝐻0: 𝜎𝑖2 = 𝜎𝑢

2 ,

which presents the homoscedastic properties of the panel data. To account for the presence

of heteroscedasticity among the variances, standard errors robust to heteroscedasticity must

be computed and applied. Second, Wooldridge test tests for autocorrelation among the error

term in panel groups against the null hypothesis of no first-order autocorrelation, 𝐻0 =

𝑐𝑜𝑣(𝑒𝑖𝑡 , 𝑒𝑖𝑠) = 0. If the error term 𝑒𝑖𝑡 exhibits first-order autocorrelation, it is denoted as

AR(1) process, 𝑒𝑖𝑡 = 𝜌𝑒𝑖𝑡−1 + 𝑣𝑖𝑡. Consider the following linear one-way model:

𝑦𝑖𝑡 = 𝑎 + 𝑋𝑖𝑡𝐵1 + 𝑍𝑖𝐵2 + 𝜇𝑖 + 𝜖𝑖𝑡 (5)

133

Wooldridge test uses the residuals from the regression with first-differences:

𝑦𝑖𝑡 − 𝑦𝑖𝑡−1 = (𝑋𝑖𝑡 − 𝑋𝑖𝑡−1)𝐵1 + 𝜖𝑖𝑡 (6)

𝛥𝑦𝑖𝑡 = 𝛥𝑋𝑖𝑡𝐵1 + 𝛥𝑒𝑖𝑡 (7)

where 𝛥 is the first-difference operator.

The procedure of Wooldridge starts by estimating the parameter 𝐵1 by regressing

𝛥𝑦𝑖𝑡 on 𝛥𝑋𝑖𝑡 and attaining the residuals 𝜖�̂�𝑡 . If 𝜖𝑖𝑡 is not serially correlated, then

𝑐𝑜𝑟𝑟(𝛥𝜖𝑖𝑡 , 𝛥𝜖𝑖𝑡−1) = −0.5. To adjust for the panel correlations in the regression of 𝜖�̂�𝑡

on 𝜖�̂�𝑡−1, robust standard errors or clustering at the panel level must be applied, which is

also robust to heteroscedasticity issues.

134

4.2.2.4 Equation to be estimated with fixed effects and random effects models

The following equation considers the panel regression model with seven statistically

significant variables to be estimated and further interpreted in the results:

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = 𝛽0 + 𝛽1 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 + 𝛽2 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)𝑖𝑡

+ 𝛽3 𝑥𝑜𝑙𝑑𝑖𝑡

+ 𝛽4 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 𝛽5 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡 + 𝛽6 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 𝑢𝑖 + 𝑒𝑖𝑡


period from 2000 to 2016, 𝑢𝑖 is an individual-specific effect error term, and 𝑒𝑖𝑡 is an

idiosyncratic error term. The variables 𝑙𝑛(𝑦𝑒𝑙𝑒𝑐) , 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) , 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) , 𝑥𝑜𝑙𝑑 ,

𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒) , 𝑙𝑛(𝑥𝑐𝑑𝑑) , and 𝑙𝑛(𝑥ℎ𝑑𝑑) represent logged term of residential

electricity consumption per capita, logged term of personal income per capita, percentage

of population aged 65 and above, logged term of household size, logged term of cooling

degree days (CDDs), and logged term of heating degree days (HDDs). In fixed effects

model, 𝑢𝑖 is considered as a parameter to be estimated, whereas in random effects model,

𝑢𝑖 is considered as a random variable following the distribution 𝑢𝑖~𝑁(0, 𝜎𝑢2) along with

the distribution of an error term 𝑒𝑖𝑡~𝑁(0, 𝜎𝑒2).

135

Chapter 5. Empirical results

5.1 Panel cointegration results

First, the presence of cross-sectional dependence among the variables is checked in

order to confirm the robustness of the test. Table 5.1 reports the results of cross-sectional

dependence test proposed by Pesaran (2004, 2015). All the test statistics of the results

strongly reject the null hypothesis of cross-sectional independence among the series at the

1% significance levels, indicating that cross-sectional dependence exists in the panel units.

Table 5.2 summarizes the results of panel unit root tests and show that all the variables

appear to be non-stationary integrated of order one I(1) variables. This holds true because

the test statistics of the variables at level fail to reject the null hypothesis of non-stationarity

of the variables but reject the null hypothesis with the variables at first-difference. Table

5.3 reports the results of Pedroni’s (1999, 2004) and Kao’s (1999) panel cointegration tests.

The results for Pedroni panel cointegration test indicate that two out of four within-

dimension statistics and two out of three between-dimension statistics strongly reject the

null hypothesis of no cointegration.

136

Table 5.1. Results of Pesaran’s (2004, 2015) cross-sectional dependence tests

Variable Test statistic p-value

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐𝑡) 44.770 0.000

𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 43.782 0.000

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) 44.454 0.000

𝑥𝑜𝑙𝑑 44.694 0.000

𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒) 45.080 0.000

𝑙𝑛(𝑥ℎ𝑑𝑑) 41.299 0.000

𝑙𝑛(𝑥𝑐𝑑𝑑) 34.687 0.000

Note: ***, ** and * denote rejection of the null hypothesis of cross-sectional independence among the series in the panel at 1%, 5% and 10%

significance levels, respectively.

137

Table 5.2. Results of Pesaran’s (2007) panel unit root tests

Note: ***, ** and * denote rejection of null hypothesis of non-stationarity of the variables at 1%, 5% and 10% significance levels, respectively.

Figures in the parentheses are p-values.

Variable

Test statistic at level Test statistic at first difference

Without trend With trend Without trend With trend

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐𝑡) -0.784

(0.217)

1.167

(0.878)

-7.563***

(0.000)

-5.791***

(0.000)

𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 0.044

(0.518)

-1.243

(0.107)

-10.565***

(0.000)

-9.316***

(0.000)

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) 2.416

(0.992)

2.562

(0.995)

-3.212***

(0.001)

-1.500*

(0.067)

𝑥𝑜𝑙𝑑 3.389

(1.000)

3.510

(1.000)

-0.050

(0.480)

-6.957***

(0.000)

𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒) -0.758

(0.224)

0.769

(0.779)

-3.217***

(0.001)

-1.556*

(0.060)

𝑙𝑛(𝑥ℎ𝑑𝑑) 0.927

(0.823)

1.572

(0.942)

-9.952***

(0.000)

-7.592***

(0.000)

𝑙𝑛(𝑥𝑐𝑑𝑑) 2.134

(0.984)

4.391

(1.000)

-11.237***

(0.000)

-10.031***

(0.000)

138

Table 5.3. Pedroni’s (1999, 2004) and Kao’s (1999) panel cointegration tests for individual models without the deterministic trends

Note: ***, ** and * denote rejection of the null hypothesis of no cointegration at 1%, 5% and 10% significance levels, respectively. Figures in

the parentheses are p-values. Lag length are determined using Akaike Information Criterion (AIC).

Pedroni’s (1999, 2004) residual cointegratrion test

Within-dimension test statistics Between-dimension test statistics

Panel v-statistic -0.999 (0.841) Group rho-statistic 4.66 (1.000)

Panel rho-statistic 3.352 (0.999) Group-PP statistic -3.036*** (0.001)

Panel PP-statistic -1.976*** (0.024) Group ADF statistic -2.913*** (0.001)

Panel ADF-statistic -2.111*** (0.017)

Kao’s (1999) residual cointegration test

ADF -4.751*** (0.000)

139

5.2 Fully modified ordinary least squares (FMOLS) and

dynamic least squares (DOLS) results

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = 𝛽0 + 0.191 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 − 0.600 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)𝑖𝑡

− 0.018𝑥𝑜𝑙𝑑𝑖𝑡−

1. 406 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 0.193 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 0.088 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡 + 𝑒𝑖𝑡 (1)

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = 𝛽0 + 0.213 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 − 0.565 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)𝑖𝑡

− 0.017𝑥𝑜𝑙𝑑𝑖𝑡−

1.410 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 0.199 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 0.077 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡 + 𝑒𝑖𝑡 (2)

Having determined the cointegrating relationship in the panel series, the study

conducts FMOLS and DOLS methods proposed by Pedroni (2001) to find the long-run

parameters. In Table 5.4 and Table 5.5, the results of FMOLS and DOLS estimates are

reported, respectively. Eq. 1 and Eq. 2 show the estimated equations for FMOLS and DOLS

estimators, respectively. The coefficients on logged personal income per capita or price

elasticity of the demand for residential electricity are found to be 0.191 and 0.213 at 1%

and 5% significance levels for FMOLS and DOLS, respectively. This means that 1%

increase in disposable or personal income per capita increases the residential electricity

consumption per capita by 0.191 or 0.213% in the long run. This is consistent with the

previous studies which confirm the positive relationship between income and residential

electricity consumption. The coefficients on logged residential electricity price is estimated

to be -0.600 and -0.565 for FMOLS and DOLS, respectively, and are statistically

significant at 1% level. This can be interpreted that 1% increase in the prices of residential

140

electricity decreases residential electricity consumption per capita by 0.600% or 0.565% in

the long run. Also, price elasticity for residential electricity demand is found to be negative

and inelastic, consistent with the results of the previous studies.

The share of population aged 65 and above is shown to exhibit negative relationship

with residential electricity consumption, with the coefficients of -0.018 and -0.017 at 1%

significance level for FMOLS and DOLS, respectively. This means that 1% increase in the

share of population aged 65 and above will reduce the residential electricity consumption

per capita by 1.8% or 1.7% in the long run. This result is consistent with the previous

studies that have confirmed the negative aging effects on residential electricity

consumption in Korea.

The coefficients on the logged variable of household size is estimated to be -1.406

and -1.410 at 1% significance level for FMOLS and DOLS, respectively. This implies that

1% increase in household size reduces residential electricity consumption by 1.406% or

1.410% in the long run. This is consistent with the results of previous studies that state that

the increase in the number of people in a household reduces residential electricity

consumption since the energy efficiency per household improves as the size of a household

gets larger in general.

Logged variables for heating degree days (HDDs) have the coefficients of 0.193 and

0.199 at 1% significance level for FMOLS and DOLS, respectively. This means that 1%

increase of HDDs increases residential electricity consumption per capita by 0.193% or

0.199% in the long run. Logged variables for cooling degree days (CDDs) have the

coefficients of 0.088 and 0.077 at 1% significance level for FMOLS and DOLS,

141

respectively. This can be interpreted that as 1% of cooling degree days increase, 0.088% or

0.077% of residential electricity consumption rises in the long run.

142

Table 5.4. Results from the estimations with Fully Modified OLS (FMOLS)

Variable Coefficient Standard error Test statistic p-value Direction

𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 0.191*** 0.039 4.821 0.000 +

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) -0.600*** 0.053 -11.243 0.000 -

𝑥𝑜𝑙𝑑 -0.018*** 0.002 -8.169 0.000 -

𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒) -1.406*** 0.078 -17.798 0.000 -

𝑙𝑛(𝑥ℎ𝑑𝑑) 0.193*** 0.017 10.838 0.000 +

𝑙𝑛(𝑥𝑐𝑑𝑑) 0.088*** 0.013 6.318 0.000 +

R-squared 0.960

Adjusted R-squared 0.957

Note: *** denotes rejecting the hypothesis at 1% significance level.

143

Table 5.5. Results from the estimations with Dynamic Least Squares (DOLS)


𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 0.213** 0.093 2.280 0.023 +

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) -0.565*** 0.127 -4.441 0.000 -

𝑥𝑜𝑙𝑑 -0.017*** 0.005 -3.198 0.001 -


𝑙𝑛(𝑥ℎ𝑑𝑑) 0.199*** 0.043 4.561 0.000 +

𝑙𝑛(𝑥𝑐𝑑𝑑) 0.077** 0.033 2.307 0.021 +

R-squared 0.966


Note: *** and ** denote rejecting the hypothesis at 1% and 5% significance levels, respectively.

144

5.3 Results for tests of hypotheses with panel data

The following section briefly describes the results for tests of hypotheses with panel

data which are considered prior to conducting panel estimations with fixed effects and

random effects models. These tests are important for they confirm the suitability of the

models used for estimations.

145

Table 5.6. Results of the Breusch-Pagan Lagrange Multiplier (LM) test for random

effects

Chi-squared statistics p-values

χ2(1) = 599.48 Prob>χ2 = 0.000

𝑯𝟎: 𝒗𝒂𝒓(𝒖𝒊) = 𝝈𝒖𝟐 = 𝟎

Table 5.6 presents the results from the Breusch-Pagan Lagrange Multiplier (LM) test

for random effects. The null hypothesis is rejected with the p-value of 0.000 at the

significance level of 1%. This result implies the presence of heteroscedastic variances of

the error terms and means that the estimation with random effects model which considers

the regional characteristics of a panel data is an appropriate method.

146

Table 5.7. Results of Wald test for fixed effects


χ2(6) = 5906.52 Prob.>χ2 = 0.000

𝑯𝟎: 𝒂𝒍𝒍 𝒖𝒊 = 𝟎.

Wald test is conducted to choose whether fixed effects model or pooled OLS

estimation is a better tool for estimation. In specific for the testing of the null hypothesis,

failure to reject the null hypothesis means that the constant terms in the model becomes 𝛼,

and thus is acceptable to use pooled OLS estimation which does not consider the

heterogeneity across the regions; rejection of the null hypothesis means that estimating with

fixed effects model is a suitable method. Since the null hypothesis is rejected at the

significance level of 1%, using fixed effects model which considers heterogeneity across

the regions is a better method.

147

Table 5.8. Results of Hausman test for fixed effects and random effects models


χ2(6) = 42.84 Prob>χ2 = 0.000

𝑯𝟎: 𝒄𝒐𝒗(𝒙𝒊𝒕, 𝒖𝒊) = 𝟎

Failure to reject the null hypothesis means that random effects model is suitable for

estimating the panel data; rejection of the null hypothesis means that fixed effects model is

more efficient model to obtain consistent estimates in panel series.

148

Table 5.9. Results of modified Wald test for heteroscedasticity


χ2(16) = 127.18 Prob>χ2 = 0.000

𝑯𝟎: 𝝈𝒊𝟐 = 𝝈𝟐 𝒇𝒐𝒓 𝒂𝒍𝒍 𝒊

Table 5.9 presents that the null hypothesis of groupwise homoscedasticity is

strongly rejected at 1% significance level. Rejection of the null hypothesis for modified

Wald test means that cross-sectional units in across the panel data exhibit heteroscedasticity.

To relax the presence of heteroscedasticity, the final model is estimated using robust

standard errors which are robust to heteroscedasticity issues.

149

Table 5.10. Results of Wooldridge test for autocorrelation

F-statistics p-values

𝐹(1, 15) = 31.225 Prob>F = 0.000

𝑯𝟎: 𝒄𝒐𝒓𝒓(𝜟𝝐𝒊𝒕, 𝜟𝝐𝒊𝒕−𝟏) = −𝟎. 𝟓

Table 5.10 reports the results for Wooldridge test for autocorrelation. The F-

statistics is 152.97, which strongly reject the null hypothesis of no first order correlation in

panel data at 1% significance level. This means that the panel data are under the issues of

autocorrelation. To ease these issues, the final model is estimated considering the robust

standard errors or clustering at panel level.

150

5.4 Fixed effects and random effects results

Through the process of eliminating the statistically insignificant variables one by one,

the equation with the variables which are statistically significant for every single one is

achieved. This leads to the interpretation on the estimated coefficients more robust. The

following equations present the estimated coefficients with fixed effects and random effects

models, respectively, for seven statistically significant variables that are chosen ultimately

for empirical analysis in the study:

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = 1.251 + 0.212 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 − 0.564 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)𝑖𝑡

− 0.017𝑥𝑜𝑙𝑑𝑖𝑡

− 1.412 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 0.200 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 0.077 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡

+ 𝑢𝑖 + 𝑒𝑖𝑡

(3)

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = 2.365 + 0.178 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 − 0.734 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)𝑖𝑡

− 0.014𝑥𝑜𝑙𝑑𝑖𝑡

− 1.165 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 0.123 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 0.108 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡

+ 𝑢𝑖 + 𝑒𝑖𝑡

(4)


period from 2000 to 2016, 𝑢𝑖 is an individual-specific effect error term, and 𝑒𝑖𝑡 is an

151

idiosyncratic error term. The variables 𝑙𝑛(𝑦𝑒𝑙𝑒𝑐) , 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) , 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) , 𝑥𝑜𝑙𝑑 ,

𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒) , 𝑙𝑛(𝑥𝑐𝑑𝑑) , and 𝑙𝑛(𝑥ℎ𝑑𝑑) represent logged term of residential

electricity consumption per capita, logged term of personal income per capita, percentage

of population aged 65 and above, logged term of household size, logged term of cooling

degree days (CDDs), and logged term of heating degree days (HDDs). In fixed effects

model, 𝑢𝑖 is considered as a parameter to be estimated, whereas in random effects model,

𝑢𝑖 is considered as a random variable following the distribution 𝑢𝑖~𝑁(0, 𝜎𝑢2) along with

the distribution of an error term 𝑒𝑖𝑡~𝑁(0, 𝜎𝑒2).

Table 5.11. and Table 5.12. present the estimated results with fixed effects and

random effects models. The coefficients on logged personal income per capita or price

elasticity of the demand for residential electricity are found to be 0.212 and 0.178 at 1%

significance level for fixed effects and random effects, respectively. This means that 1%

increase in disposable or personal income per capita increases the residential electricity

consumption per capita by 0.212 or 0.178%. This is consistent with the previous studies

which confirm the positive relationship between income and residential electricity

consumption. The coefficients on logged residential electricity price is calculated to be -

0.564 and -0.734 for fixed effects and random effects, respectively, and are statistically

significant at 1% level. This can be interpreted that 1% increase in the prices of residential

electricity decreases residential electricity consumption per capita by 0.564% or 0.734%.

Also, price elasticity for residential electricity demand is found to be negative and inelastic,

consistent with the results of the previous studies.

152

The share of population aged 65 and above is shown to exhibit negative relationship

with residential electricity consumption, with the coefficients of -0.017 and -0.014 at 1%

significance level for fixed effects and random effects models, respectively. This means

that 1% increase in the share of population aged 65 and above will reduce the residential

electricity consumption per capita by 1.7% or 1.4%. This can be explained in part by the

general propensities of older people towards energy savings and preference of analogue

styles to digital styles; with these, older people use much less home appliances and

electricity than younger generations. The result is consistent with the previous studies that

have confirmed the negative aging effects on residential electricity consumption in Korea.

The coefficients on the logged variable of household size is estimated to be -1.412

and -1.165 at 1% significance level for fixed effects and random effects models,

respectively. This implies that 1% increase in household size reduces residential electricity

consumption by 1.412% or 1.165%, which is consistent with the previous studies that state

that the increase in the number of people in a household reduces residential electricity

consumption since the energy efficiency per household improves as the size of a household

gets larger in general.

Logged variables for heating degree days (HDDs) have the coefficients of 0.200 and

0.123 at 1% significance level for fixed effects and random effects models, respectively.

This means that 1% increase of HDDs increases residential electricity consumption per

capita by 0.200% or 0.123%. Logged variables for cooling degree days (CDDs) have the

coefficients of 0.077 and 0.108 at 1% significance level for fixed effects and random effects

153

models, respectively. This can be interpreted that as 1% of cooling degree days increases,

0.077% or 0.108% of residential electricity consumption rises.

Table 5.13 presents the results from the panel estimations with Least Square Dummy

Variable (LSDV) model considering the characteristics that are inherent in a certain region.

With the reference group being Seoul city, the equation for LSDV model with the estimated

coefficients is as follows:

𝑙𝑛(𝑦𝑒𝑙𝑒𝑐)𝑖𝑡 = ∑ 𝐴𝑖15𝑖=1 + 1.185 + 0.212 𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒)𝑖𝑡 − 0.564 𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒)

𝑖𝑡−

0.017𝑥𝑜𝑙𝑑𝑖𝑡− 1.412 𝑙𝑛(𝑥ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑𝑠𝑖𝑧𝑒)𝑖𝑡 + 0.200 𝑙𝑛(𝑥ℎ𝑑𝑑)𝑖𝑡 + 0.077 𝑙𝑛(𝑥𝑐𝑑𝑑)𝑖𝑡 + 𝑢𝑖 +

𝑒𝑖𝑡

(5)

where 𝐴𝑖 represents the constant term for each regional dummy which demonstrates the

inherent regional characteristics affecting the residential electricity consumption per capita

in Korea, in comparison to the reference group, Seoul. It is shown in Table 5.13 that the six

major metropolitan cities in Korea consisting of Busan, Incheon, Daegu, Daejeon, and

Gwangju exhibit statistically significant coefficients with the values 0.165, 0.060, 0.124,

0.041, and 0.085, respectively. This means that due to their inherent regional characteristics,

these six metropolitan cities consume more residential electricity per capita than Seoul does

by about 0.165%, 0.060%, 0.124%, 0.041%, and 0.085% more for Busan, Incheon, Daegu,

Daejeon, and Gwangju, respectively. This may be explained from the fact that these

154

metropolitan cities consist of people more than one million and possess somewhat lesser

population densities and thus larger residential area per person than Seoul does; the larger

the residential area per capita, the greater the residential electricity consumption per capita

may be. On the other hand, it is shown that the provinces which mainly consist of the rural

cities with less than 500,000 people and low population densities such as Chungbuk,

Chungnam, Gangwon, Gyeongbuk, and Jeju-do do not have significantly different effects

than Seoul does. However, further investigations must be conducted for more concrete

interpretations.

155

Table 5.11. Results from the estimations with fixed effects model


𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 0.212*** 0.068 3.100 0.002 +

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) -0.564*** 0.093 -6.070 0.000 -

𝑥𝑜𝑙𝑑 -0.017*** 0.003 -4.370 0.000 -


𝑙𝑛(𝑥ℎ𝑑𝑑) 0.200*** 0.032 6.250 0.000 +

𝑙𝑛(𝑥𝑐𝑑𝑑) 0.077*** 0.024 3.150 0.002 +

constants 1.251 0.879 1.420 0.156 +

R-squared (within) 0.963

R-squared (between) 0.409

R-squared (overall) 0.878

Note: *** denotes rejecting the hypothesis at 1% significance level.

156

Table 5.12. Results from the estimations with random effects model


𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 0.178*** 0.061 2.900 0.004 +

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) -0.734*** 0.098 -7.490 0.000 -

𝑥𝑜𝑙𝑑 -0.014*** 0.003 -4.200 0.000 -


𝑙𝑛(𝑥ℎ𝑑𝑑) 0.123*** 0.042 2.910 0.004 +

𝑙𝑛(𝑥𝑐𝑑𝑑) 0.108*** 0.025 4.230 0.000 +

constants 2.365** 0.930 2.540 0.011 +

R-squared (within) 0.962

R-squared (between) 0.397

R-squared (overall) 0.905

Note: *** and ** denote rejecting the hypothesis at 1% and 5% significance levels, respectively.

157

Table 5.13. Results from the estimations with LSDV (Reference group: Seoul)

Variable Coefficient Standard

error

Test

statistic p-value Direction

𝑙𝑛(𝑥𝑖𝑛𝑐𝑜𝑚𝑒) 0.212*** 0.068 3.100 0.002 +

𝑙𝑛(𝑥𝑝𝑟𝑖𝑐𝑒) -0.564*** 0.093 -6.070 0.000 -

𝑥𝑜𝑙𝑑 -0.017*** 0.005 -6.480 0.000 -


𝑙𝑛(𝑥ℎ𝑑𝑑) 0.200*** 0.032 6.250 0.000 +

𝑙𝑛(𝑥𝑐𝑑𝑑) 0.077*** 0.024 3.150 0.002 +

constants 1.185 0.895 1.320 0.187

Chungbuk 0.020 0.030 0.680 0.495

Chungnam 0.010 0.034 0.310 0.755

Daegu 0.124** 0.023 5.220 0.000

Daejeon 0.041** 0.019 2.180 0.030

Gangwon 0.042 0.035 1.200 0.233

Gwangju 0.085*** 0.022 3.800 0.000

Gyeongbuk 0.007 0.037 0.190 0.851

Gyeonggi 0.037** 0.015 2.380 0.018

Gyeongnam 0.108** 0.027 3.910 0.000

Incheon 0.060*** 0.019 3.020 0.003

Jeju 0.045 0.030 1.480 0.140

Jeonbuk 0.116*** 0.038 3.030 0.003

Jeonnam 0.120** 0.050 2.350 0.019

Busan 0.165*** 0.024 6.660 0.000

Ulsan 0.074*** 0.019 3.740 0.000

R-squared 0.9663


Note: *, ** and *** denote rejecting the hypothesis at 1%, 5%, and 1% significance levels,

respectively

158

Table 5.14. Confirmation of the research hypotheses with the estimated results

Research

hypotheses

Estimated

results from

FMOLS

Estimated

results from

DOLS

Estimated

results from

fixed effects

model

Estimated

results from

random

effects model

Hypothesis I:

Population aging

and residential

electricity

consumption exhibit

a negative

relationship

Yes, with the

coefficient of

-0.018

Yes, with the

coefficient of

-0.017

Yes, with the

coefficient of

-0.017

Yes, with the

coefficient of

-0.014

Hypothesis II:

Increase in

household size

reduces residential

electricity

consumption per

capita

Yes, with the

coefficient of

-1.406

Yes, with the

coefficient of

-1.410

Yes, with the

coefficient of

-1.412

Yes, with the

coefficient of

-1.165

Hypothesis III:

Education level,

gender difference,

working population,

and homeownership

do not have

significant effects on

residential electricity

consumption

Yes. Hypothesis III is confirmed via the process of eliminating

the statistically significant variables. Education level, female

to male ratio, working population, and homeownership rate are

found to have no effects on residential electricity consumption,

at least for the empirical analysis.

Hypothesis IV:

Heating degree days

(HDDs) and cooling

degree days (CDDs)

have positive

relationships with

residential electricity

consumption

Yes, with the

coefficient of

0.193 for

HDDs and

0.088 for

CDDs

Yes, with the

coefficient of

0.199 for

HDDs and

0.077 for

CDDs

Yes, with the

coefficient of

0.200 for

HDDs and

0.077 for

CDDs

Yes, with the

coefficient of

0.123 for

HDDs and

0.108 for

CDDs

159

Chapter 6. Conclusions

6.1 Conclusions

The study has conducted empirical analyses on how population aging and other

socio-demographic factors affect residential electricity consumption per capita in 16

municipalities, excluding Sejong City, in Korea during the period of 2000-2016. For

empirical methods, the study has chosen Fully Modified OLS (FMOLS) and Dynamic OLS

(DOLS) with panel cointegration approach and fixed effects and random effects models for

panel estimations. Then, the study has applied least squares dummy variables (LSDV)

models considering each regional characteristic that may differently affect the residential

electricity consumption per capita in Korea.

By classifying several socio-demographic factors, the study has proposed the

following four research hypotheses for in-depth analysis: 1) Population aging and

residential electricity consumption exhibit a negative relationship; 2) increase in household

size reduced residential electricity consumption per capita; 3) Education level, gender

difference, working age population, and homeownership do not have significant effects on

residential electricity consumption; and 4) Heating degree days (HDDs) and cooling degree

days (CDDs) have positive relationships with residential electricity consumption.

From the empirical results, the study draws conclusions that all four research

hypotheses presented above are valid. For the first hypothesis, population aging has a

negative impact on the residential electricity consumption in the long-run with the

160

parameter estimates ranging from -0.018 to -0.014. This means that population aging

reduces the residential electricity consumption per capita in Korea, but the aging society is

not a desirable condition nationwide. From the energy-related perspective, rapid population

aging in Korea suggests the widening gap in energy equity in Korea; growing number of

elderly generations may lead to elderly households that lack accessibility to or affordability

of energy at homes. As clearly exemplified in the World Energy Council’s Energy

Trilemma Index, Korea is ranked 26th in Energy equity, an index which measures

accessibility to or affordability of energy; it is worth to take note that the other advanced

countries such as Netherlands, Switzerland, Austria, and Denmark are ranked 6th, 7th, 8th,

and 12th, respectively. The government must devise effective plans to make sure that the

elderly and the vulnerable groups in Korea are under the access to energy safely and freely.

Further, since Korea's population aging is progressing at an unprecedented rate, this

suggests that its impact on residential electricity consumption will be much different than

now in the future. It is recommended that the government set up plans for electricity supply

and demand in preparation for the upcoming changes in population structure and their

effects on the volatility of residential electricity consumption.

For the second hypothesis, as the size of households grows, the residential electricity

consumption per capita decreases. This result parallels that of Ota et al. (2018) which states

that energy efficiency per household improves as the size of a household grows, in general,

considering that the increase in the number of people in a household reduces the needs to

duplicate usage of shared appliances or electricity. Policy-wise, it is important to introduce

R&D policies to encourage the development of technologies for large home appliances to

improve their energy efficiency and product qualities.

161

For the third hypothesis, heating degree days (HDDs) and cooling degree days

(CDDs) positively affect residential electricity consumption in Korea. This can be

interpreted that for heating, more people are using their means to heat a house by using

electricity than city gas. For cooling, more people today than in the past have air-

conditioners installed at their homes, and these generations have propensity to freely use

air-conditioners and other cooling devices which use electricity. Hence, it is recommended

for the government to devise energy plans considering the demand for electricity for

heating and cooling in accordance with the rise and decline in temperatures due to climate

change.

For the fourth hypothesis, through the process of eliminating the statistically

insignificant variables, education level, gender difference, working age population, and

homeownership are found have no significant effects on residential electricity consumption.

The differences in education level and gender do not pose significant effects on residential

electricity consumption because. Homeownership rate as well does not significantly affect

residential electricity consumption since a person, regardless of owning or renting a house,

must pay its bills for using electricity at homes.

Price elasticities and income elasticities of demand for residential electricity

consumption are found to be negative and positive, respectively. This means that as

electricity price increases, residential electricity consumption per capita decreases; as per

capita income increases, residential electricity consumption increases. This result gives

further insights for policy designs that while the plans that tackle electricity price rises may

help reduce the residential electricity consumption in Korea, solely considering the price

of electricity may pose some constraints as per capita personal income is expected to

162

increase.

In this study, brief overviews of Germany and Sweden regarding the effects of

population aging on the residential energy consumption are discussed; however, more

detailed and in-depth analyses consisting of cross-country comparisons considering multi-

faceted indicators are recommended to provide for Korea more cutting-edge and effective

academic and policy insights. Also, due to insufficiency of data, the study has not

considered other socio-demographic factors that may well affect residential electricity

consumption. More dynamic results are expected if the variables for the consumer behavior

factors are included in the analysis. Also, conducting thorough research by carrying out

surveys to obtain more precise data on socio-demographic and consumer-behavior factors

is recommended.

163

Bibliography

Bedir, M., Hasselaar, E., & Itard, L. (2013). Determinants of electricity consumption

in Dutch dwellings. Energy and buildings, 58, 194-207.

Boubtane, E., Coulibaly, D., & Rault, C. (2012). Immigration, Growth and

Unemployment: Panel VAR Evidence from OECD Countries (IZA Discussion Paper

No: 6966).

Brounen, D., Kok, N., & Quigley, J. M. (2012). Residential energy use and

conservation: Economics and demographics. European Economic Review, 56(5),

931-945.

Chern, W. S., & Bouis, H. E. (1982). Investigation of structural changes in

residential electricity demand (No. UCRL-87391; CONF-820878-2). Lawrence

Livermore National Lab., CA (USA).

Cowgill, D.O. (1970). The demographic of ageing. In A. M. Hoffman (Ed.). The

daily needs and interests of older people (pp. 27-69). Springfield: Charles C.

Thomas.

Cramer, J. C., Miller, N., Craig, P., Hackett, B. M., Dietz, T. M., Vine, E. L., ... &

Kowalczyk, D. J. (1985). Social and engineering determinants and their equity

implications in residential electricity use. Energy, 10(12), 1283-1291.

164

Deaton, A., & Muellbauer, J. (1980). Economics and consumer behavior. Cambridge

university press.

Deutsch, M., & Timpe, P. (2013). The effect of age on residential energy

demand. ECEEE Summer Study Proceedings pages, 2177-2188.

Donghyun Shin, Hahyun Jo, Min-Woo Jang. (2015). An Analysis on the

Heterogeneity of Residential Electricity Consumption Depending on Income Level :

Evidence from Uran Household in South Korea. Korean Energy Economic Review,

14(3), 27-81.

Druckman, A., & Jackson, T. (2008). Household energy consumption in the UK: A

highly geographically and socio-economically disaggregated model. Energy

Policy, 36(8), 3177-3192.

Electric Power Statistics Information System (EPSIS). 2016.

http://epsis.kpx.or.kr/epsisnew/selectEkccIntroEn.do?menuId=110100.

Eskeland, G. S., & Mideksa, T. K. (2010). Electricity demand in a changing

climate. Mitigation and Adaptation Strategies for Global Change, 15(8), 877-897.

Filippini, M., & Pachauri, S. (2004). Elasticities of electricity demand in urban

Indian households. Energy policy, 32(3), 429-436.

Flaig, G. (1990). Household production and the short-and long-run demand for

electricity. Energy economics, 12(2), 116-121.

165

Frederiks, E., Stenner, K., & Hobman, E. (2015). The socio-demographic and

psychological predictors of residential energy consumption: A comprehensive

review. Energies, 8(1), 573-609.

Fullerton Jr, T. M., Juarez, D. A., & Walke, A. G. (2012). Residential electricity

consumption in Seattle. Energy Economics, 34(5), 1693-1699.

Glickman, T., Ed., 2000: Glossary of Meteorology. 2d ed. Amer. Meteor. Soc., 855.

Halvorsen, R. (1975). Residential demand for electric energy. The review of

Economics and Statistics, 12-18.

Harris, R., & Sollis, R. (2003). Applied time series modelling and forecasting. Wiley.

Hasanov, F. J., & Mikayilov, J. I. (2017). The impact of age groups on consumption

of residential electricity in Azerbaijan. Communist and Post-Communist

Studies, 50(4), 339-351.

Holtedahl, P., & Joutz, F. L. (2004). Residential electricity demand in Taiwan. Energy

economics, 26(2), 201-224.

Houthakker, H. S. (1951). Some calculations on electricity consumption in Great

Britain. Journal of the Royal Statistical Society. Series A (General), 114(3), 359-371.

HyunJin Lim, Sukwan Jung, DooHwan Won. (2013). An Analysis of the Impact of

Global Warming on Residential Energy Consumption: Focused on the Case of

Electricity Consumption. Korean Energy Economic Review, 12(2), 33-58.

166

Im, K. S., Pesaran, M. H., & Shin, Y. (2003). Testing for unit roots in heterogeneous

panels. Journal of econometrics, 115(1), 53-74.

Ito, K. (2014). Do consumers respond to marginal or average price? Evidence from

nonlinear electricity pricing. American Economic Review, 104(2), 537-63.

Judt. Tony (2010). Ill Fares the Land. New York: Penguin.

Kamerschen, D. R., & Porter, D. V. (2004). The demand for residential, industrial

and total electricity, 1973–1998. Energy Economics, 26(1), 87-100.

Kao, C., & Chiang, M. H. (2001). On the estimation and inference of a cointegrated

regression in panel data. In Nonstationary panels, panel cointegration, and dynamic

panels (pp. 179-222). Emerald Group Publishing Limited.

Kao, C., & Chiang, M. H. (2001). On the estimation and inference of a cointegrated

regression in panel data. In Nonstationary panels, panel cointegration, and dynamic

panels (pp. 179-222). Emerald Group Publishing Limited.

Kelly, S. (2011). Do homes that are more energy efficient consume less energy?: A

structural equation model of the English residential sector. Energy, 36(9), 5610-5620.

Keum, Y., Kwon O., Goo, Y. (2018). Estimation of residential electricity demand in

Korea - by using dynamic panel GMM model -. Korean Energy Economic Review,

17(1), 37-65.

167

Kim, Y., Kim, S., Baek, J., & Heo, E. (2018). The linkages between democracy and

the environment: Evidence from developed and developing countries. Energy &

Environment, 0958305X18813637.

Kirkwood, T. (2001) Lecture 1: The brave old world, the end of age. Reith lectures

2001. London: BBC. http://www.bbc.co.uk/radio4/reith2001/lecture1.shtml.

Korean Statistical Information Service (KOSIS). 2016. http://kosis.kr/index/index.do.

Levin, A., Lin, C. F., & Chu, C. S. J. (2002). Unit root tests in panel data: asymptotic

and finite-sample properties. Journal of econometrics, 108(1), 1-24.

Liddle, B. (2004). Demographic dynamics and per capita environmental impact:

Using panel regressions and household decompositions to examine population and

transport. Population and Environment, 26(1), 23-39.

Liddle, B. (2011). Consumption-driven environmental impact and age structure

change in OECD countries: A cointegration-STIRPAT analysis. Demographic

Research, 24, 749-770.

Maddala, G. S., & Wu, S. (1999). A comparative study of unit root tests with panel

data and a new simple test. Oxford Bulletin of Economics and statistics, 61(S1), 631-

652.Martı́nez-Zarzoso, Inmaculada, and Aurelia Bengochea-Morancho. "Pooled

mean group estimation of an environmental Kuznets curve for CO 2." Economics

Letters 82.1 (2004): 121-126.

168

Mikayilov, J., Hasanov, F., & Yusifov, S. (2018). Residential electricity use effects of

population in Kazakhstan. International Journal of Energy Technology and

Policy, 14(1), 114-132.

MRCC, cited 2007: Weather terminology. Midwestern Regional Climate Center.

[Available online at http://mrcc.sws.uiuc.edu/resources_links/wxfaq5.htm].

Nakajima, T. (2010). The residential demand for electricity in Japan: an examination

using empirical panel analysis techniques. Journal of Asian Economics, 21(4), 412-

420.

Narayan, P. K., & Smyth, R. (2005). The residential demand for electricity in

Australia: an application of the bounds testing approach to cointegration. Energy

policy, 33(4), 467-474.

National Climate Data Service System (NCDSS). 2016

Noh, Joung Yeo Angela. (2014). Effects of Demographics and Usage of Appliances

on Household Electricity Demand in Korea. Journal of Korean Economics Studies,

32(2), 177-202.

Noh, Seung Chul, Lee, Hee Yeon. (2013). An Analysis of the Factors Affecting the

Energy Consumption of the Household in Korea. Journal of Korea Planning

Association, 48(2), 295-312.

Notestein, F. W. (1954). Some demographic aspects of aging. Proceedings of the

American Philosophical Society, 98(1), 38-45.

169

O'neill, B. C., & Chen, B. S. (2002). Demographic determinants of household energy

use in the United States. Population and development review, 28, 53-88.

O'neill, B. C., Dalton, M., Fuchs, R., Jiang, L., Pachauri, S., & Zigova, K. (2010).

Global demographic trends and future carbon emissions. Proceedings of the National

Academy of Sciences, 201004581.

Ota, T., Kakinaka, M., & Kotani, K. (2018). Demographic effects on residential

electricity and city gas consumption in the aging society of Japan. Energy

Policy, 115, 503-513.

Park, H. C., & Heo, E. (2007). The direct and indirect household energy

requirements in the Republic of Korea from 1980 to 2000—An input–output

analysis. Energy Policy, 35(5), 2839-2851.

Pedroni, P. (1999). Critical values for cointegration tests in heterogeneous panels

with multiple regressors. Oxford Bulletin of Economics and statistics, 61(S1), 653-

670.

Pedroni, P. (2001). Fully modified OLS for heterogeneous cointegrated panels.

In Nonstationary panels, panel cointegration, and dynamic panels (pp. 93-130).

Emerald Group Publishing Limited.

Pedroni, P. (2004). Panel cointegration: asymptotic and finite sample properties of

pooled time series tests with an application to the PPP hypothesis. Econometric

theory, 20(3), 597-625.

170

Pesaran, M. H. (2004). General diagnostic tests for cross section dependence in

panels.

Pesaran, M. H. (2007). A simple panel unit root test in the presence of cross‐section

dependence. Journal of applied econometrics, 22(2), 265-312.

Pessanha, J. F. M., & Leon, N. (2015). Forecasting long-term electricity demand in

the residential sector. Procedia Computer Science, 55, 529-538.

Prskawetz, A., Leiwen, J., & O'Neill, B. C. (2004). Demographic composition and

projections of car use in Austria. Vienna yearbook of population research, 175-201.

Shin. (2018). An Analysis on the Relationship between Aging and Residential

Electricity Consumption in Korea. Korea Energy Economic Review, 17(1), 95-129.

Silk, J. I., & Joutz, F. L. (1997). Short and long-run elasticities in US residential

electricity demand: a co-integration approach. Energy economics, 19(4), 493-513.

Statistics Korea (KOSTAT). 2016. http://kostat.go.kr/portal/korea/index.action.

Thompson, W. (1929). Population. American Journal of Sociology, 34, 959-975.

Varian, H. R. (1992). Microeconomic analysis.

Walker, A. (1990). The economic ‘burden’ of ageing and the prospect of

intergenerational conflict. Ageing & Society, 10(4), 377-396.

Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data.

MIT press.

171

Yamasaki, Y., Kawamori, R., Wasada, T., Sato, A., Omori, Y., Eguchi, H., & Ishida,

Y. (1997). Pioglitazone (AD-4833) ameliorates insulin resistance in patients with

NIDDM. The Tohoku journal of experimental medicine, 183(3), 173-183.

Yohanis, Y. G., Mondol, J. D., Wright, A., & Norton, B. (2008). Real-life energy use

in the UK: How occupancy and dwelling characteristics affect domestic electricity

use. Energy and Buildings, 40(6), 1053-1059.

Zhou, T. (2016). The Impact of Demographics on Residential Electricity

Usage (Doctoral dissertation).

나인강, 류지철, 김영덕, 김태헌. (2000). 에너지수요 분석 및 전망.

에너지경제연구원 연구보고서, 1-279.

신동현. (2017). 고령화가 국내 전력 소비 변동성에 미치는 영향.

에너지경제연구원 에너지포커스 2017 가을호, 14(3) 65, 99-112.

원두환. (2012). 고령화가 가정부문 에너지 소비량에 미치는 영향 분석:

전력수요를 중심으로. Environmental and Resource Economics Review,

21(2), 341-370.

이윤재, 이현수, & 박소윤. (2011). 공동주택 거주자의 에너지 사용행태 및

에너지 절약의식 분석. 한국주거학회논문집, 22(6), 31-42.

임기추, & 강윤영. (2004). 생활 양식 이 가정 부문 에너지 소비 에

미치는 영향 분석. 에너지경제연구원.

172

임현진, 정수관, & 원두환. (2013). 지구온난화가 가정부문 에너지

소비량에 미치는 영향 분석. 에너지경제연구, 12(2), 33-58.

조하현, & 장민우. (2015). 구간별 가격체계를 고려한 우리나라 주택용

전력수요의 가격탄력성과 전력누진요금제 조정방안.

자원·환경경제연구, 24(2), 365-410.

한국에너지공단 (2017) 에너지통계 핸드북.

홍종호, 오형나, 이성재. (2018). 가구 패널자료를 이용한 가계부문 에너지

소비행태 분석-1 인 가구 및 고령가구를 중심으로. 자원· 환경경제연구

27(3): 463-493.

173

Appendix

Trends of all data for 16 municipalities in Korea during 2000-2016 presented in Figure

A.1~Figure A.34.

Figure A.1. Trends of all variables excluding population aging for the entire municipalities,

2000-2016 ....................................................................................................................... 176

Figure A.2. Trends of the population aging and residential electrity consumption variables

for the entire municipalities, 2000-2016 ......................................................................... 177

Figure A.3. Trends of all variables excluding population aging for Busan, 2000-2016 . 178


for Busan, 2000-2016 ...................................................................................................... 179

Figure A.5. Trends of all variables excluding population aging for Chungbuk, 2000-2016

......................................................................................................................................... 180


for Chungbuk, 2000-2016 ............................................................................................... 181

Figure A.7. Trends of all variables excluding population aging for Chungnam,

2000-2016 ....................................................................................................................... 182


for Chungnam, 2000-2016 .............................................................................................. 183

Figure A.9. Trends of all variables excluding population aging for Daegu,

2000-2016 ....................................................................................................................... 184


for Daegu, 2000-2016 ..................................................................................................... 185

Figure A.11. Trends of all variables excluding population aging for Daejeon,

2000-2016 ....................................................................................................................... 186


for Daejeon, 2000-2016 .................................................................................................. 187

Figure A.13. Trends of all variables excluding population aging for Gangwon,

2000-2016 ....................................................................................................................... 188

174


for Gangwon, 2000-2016 ................................................................................................ 189

Figure A.15. Trends of all variables excluding population aging for Gwangju, 2000-2016

......................................................................................................................................... 190


for Gwangju, 2000-2016 ................................................................................................. 191

Figure A.17. Trends of all variables excluding population aging for Gyeongbuk,

2000-2016 ....................................................................................................................... 192


for Gyeongbuk, 2000-2016 ............................................................................................. 193

Figure A.19. Trends of all variables excluding population aging for Gyeonggi, 2000-2016

......................................................................................................................................... 194


for Gyeonggi, 2000-2016 ................................................................................................ 195

Figure A.21. Trends of all variables excluding population aging for Gyeongnam,

2000-2016 ....................................................................................................................... 196


for Gyeongnam, 2000-2016 ............................................................................................ 197

Figure A.23. Trends of all variables excluding population aging for Incheon, 2000-2016

......................................................................................................................................... 198


for Incheon, 2000-2016 ................................................................................................... 199

Figure A.25. Trends of all variables excluding population aging for Jeju, 2000-2016 ..... 20


for Jeju, 2000-2016 ......................................................................................................... 201

Figure A.27. Trends of all variables excluding population aging for Jeonbuk, 2000-2016

......................................................................................................................................... 202


for Jeonbuk, 2000-2016 .................................................................................................. 203

Figure A.29. Trends of all variables excluding population aging for Jeonnam,

175

2000-2016 ....................................................................................................................... 204


for Jeonnam, 2000-2016 ................................................................................................. 205

Figure A.31. Trends of all variables excluding population aging for Seoul, 2000-2016 206


for Seoul, 2000-2016 ....................................................................................................... 207

Figure A.33. Trends of all variables excluding population aging for Ulsan, 2000-2016 208


for Ulsan, 2000-2016 ...................................................................................................... 209

176

Figure A.1. Trends of all variables excluding population aging for the entire

municipalities, 2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5

lx_price lx_householdsize lx_cdd

lx_hdd ly_elec lx_income

177

Figure A.2. Trends of the population aging and residential electrity consumption

variables for the entire municipalities, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

178

Figure A.3. Trends of all variables excluding population aging for Busan,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



179


variables for Busan, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

180

Figure A.5. Trends of all variables excluding population aging for Chungbuk, 2000-

2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



181


variables for Chungbuk, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

182

Figure A.7. Trends of all variables excluding population aging for Chungnam,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



183


variables for Chungnam, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

18.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

184

Figure A.9. Trends of all variables excluding population aging for Daegu,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



185


variables for Daegu, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

186

Figure A.11. Trends of all variables excluding population aging for Daejeon,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



187


variables for Daejeon, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

188

Figure A.13. Trends of all variables excluding population aging for Gangwon,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



189


variables for Gangwon, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

18.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

190

Figure A.15. Trends of all variables excluding population aging for Gwangju,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



191


variables for Gwangju, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

192

Figure A.17. Trends of all variables excluding population aging for Gyeongbuk,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



193


variables for Gyeongbuk, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

18.000

20.000

5.75

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

194

Figure A.19. Trends of all variables excluding population aging for Gyeonggi, 2000-

2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5

20002001200220032004200520062007200820092010201120122013201420152016



195


variables for Gyeonggi, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

196

Figure A.21. Trends of all variables excluding population aging for Gyeongnam,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



197


variables for Gyeongnam, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

198

Figure A.23. Trends of all variables excluding population aging for Incheon, 2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



199


variables for Incheon, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

200

Figure A.25. Trends of all variables excluding population aging for Jeju, 2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



201


variables for Jeju, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

5.75

5.8

5.85

5.9

5.95

6

6.05

6.1

ly_elec x_old

202

Figure A.27. Trends of all variables excluding population aging for Jeonbuk, 2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



203


variables for Jeonbuk, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

16.000

18.000

20.000

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

204

Figure A.29. Trends of all variables excluding population aging for Jeonnam,

2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



205


variables for Jeonnam, 2000-2016

0.000

5.000

10.000

15.000

20.000

25.000

5.75

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

206

Figure A.31. Trends of all variables excluding population aging for Seoul, 2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



207


variables for Seoul, 2000-2016

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

208

Figure A.33. Trends of all variables excluding population aging for Ulsan, 2000-2016

0

1

2

3

4

5

6

7

0

0.5

1

1.5

2

2.5



209


variables for Ulsan, 2000-2016

0.000

1.000

2.000

3.000

4.000

5.000

6.000

7.000

8.000

9.000

10.000

5.75

5.8

5.85

5.9

5.95

6

6.05

6.1

6.15

ly_elec x_old

210

Note: For Figure A.1 through Figure A.17, the variables are logged first to see the trends in

the fitted range. In specific, before logging the values of residential electricity consumption

𝑦𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 variables, they are converted from Megawatts hour per person to Watt hours

per person by multiplying by 106 to make them fit into the corresponding range.

211

Abstract (Korean)

한국의 고령화 속도는 주요 선진국들의 고령화 속도와 비교하여 매우

빠른 편으로, 미국 통계국은 한국이 고령사회에서 초고령사회로 진입하는데

세계에서 가장 빠른 수준인 27년이 걸린다고 전망하였다. 인구고령화 현상은

성장, 고용, 복지 등 경제 및 사회 전반에 영향을 미칠 뿐만 아니라 에너지

소비 행태에 구조변화를 일으킬 가능성이 높기 때문에 최근 학계에서는

인구고령화를 포함한 사회·인구학적 요인들이 에너지 소비 변동에 미치는

영향에 대한 분석을 활발히 진행하고 있다. 특히 대부분의 관련 연구들은

주로 수송부문과 가정부문에 집중되어 있는데, 이는 사회·인구학적 현상이

산업 및 발전 부문보다 수송 및 가정부문에 더 직접적인 영향을 끼치기

때문이다. 고령화가 수송부문과 가정부문의 에너지소비에 미치는 영향에 대한

연구가 급증하고 있는 가운데, 아직까지 사회·인구학적 요인들을 세분화하여

지역적 특성을 고려한 패널분석을 실시한 연구는 미미한 실정이다.

본 연구의 목적은 고령화를 포함한 사회·인구학적 요인들이 국내

1인당 주거용 전력수요에 미치는 영향을 실증분석 하는것이다. 사회·인구학적

요인을 크게 고령화, 소득, 생산가능인구, 교육수준, 가구원크기, 성별,

자가점유율, 그리고 주택형태로 세분화하여 분석을 진행한다. 이 외에

전력소비에 유의미한 영향을 미칠 것으로 판단되는 전력가격, 냉방도일,

그리고 난방도일 변수를 분석에 포함한다. 분석범위는 세종시를 제외한 16개

광역자치단체이며, 분석기간은 2000년부터 2016년까지인 지역별 패널자료를

사용한다. 지역별 패널자료를 사용할 경우 변수들간의 다중공선성(multi-

collinearity) 문제의 해결과 시간에 따른 전력소비의 변동성을 포착할 수

있다는 이점이 있다.

실증분석을 실시하기에 앞서 관련 선행연구의 조사와 정리를 통해

이론적 근거 및 관련 연구들을 소개한다. 그 뒤, 다음과 같은 네 가지 연구

212

가설을 설정하여 검정한다. 첫째, 인구 고령화가 증가하면 1인당 주거용 전력

소비량이 감소한다. 둘째, 가구원의 크기가 증가하면 1인당 주거용 전력

소비량이 감소한다. 셋째, 교육수준, 성별차이, 생산가능인구, 자가점유율은

주거용 전력 소비량에 유의한 영향을 미치지 않는다. 넷째, 난방도일(HDDs:

Heating Degree Days)과 냉방도일(CDDs: Cooling Degree Days)이 증가하면

주거용 전력 소비량이 증가한다.

통계적으로 유의한 소득, 가격, 고령화, 가구원크기, 냉방도일,

난방도일 변수들을 채택하여 분석을 실시하였다. 패널 공적분 검정 결과,

변수들 간에 장기균형관계가 성립함을 확인하였으며, FMOLS와 DOLS

분석기법을 사용하여 변수들의 장기균형 벡터를 추정하였다. FMOLS와 DOLS

분석결과 고령화, 전력가격, 가구원크기가 증가하면 주거용 전력소비가

장기적으로 감소하고, 개인소득, 난방도일, 냉방도일이 증가하면 주거용

전력소비가 장기적으로 증가하는 것을 확인하였다. 패널 고정효과 및

확률효과 모형 추정 결과 고령화, 전력가격, 가구원크기가 증가하면 주거용

전력소비가 감소하고, 개인소득, 난방도일, 냉방도일이 증가하면 주거용

전력소비가 증가하는 것을 확인하였다.

본 연구는 국내 고령화 현상이 주거용 전력소비 감소의 요인임을

확인하였으나, 고령화 사회는 국가적으로 보면 매우 큰 부담이다. 한국의

인구고령화가 전례 없는 속도로 진행되고 있기 때문에 이에 따른 인구구조

변화가 향후 주거용 전력 소비에 미치는 영향은 크게 변동할 것으로 판단된다.

또한 가구원크기가 증가할수록 주거용 전력 소비가 줄어드는데, 이는 세대당

인구수가 증가할수록 한 가정에서 가전기기를 공유하여 사용하는 빈도가

늘어나기 때문이다. 가전기기를 대상으로 한 에너지 효율적인 설계와 보급을

장려하는 정책의 도입이 바람직해 보인다.

난방도일이 증가할수록 주거용 전력소비가 증가한다는 것은, 최근

전력으로 난방 설비를 가동하는 건물이 많아지면서 가스난방의 대체로 겨울철

전기난방이 사용되거나, 값싼 전기요금으로 인해 전기를 사용하는 보조 난방

213

기구의 사용이 증가하는 것에서 설명될 수 있다. 냉방도일이 증가할수록

주거용 전력소비가 증가하는 현상은 과거에 비해 에어컨과 냉방기기를

보유하고 사용하는 가정이 늘어난 데 기인한다. 소득이 증가할수록 주거용

전력소비가 증가한다는 것은, 가처분소득의 증가로 인해 개인의 구매력이

증가하면서, 이들이 전기를 사용하는 가전제품을 더 많이 구매하고 사용하는

현상에서 비롯된다.

최근 한국은 공식적으로 고령사회에 진입하였으며, 국제사회는 한국의

고령화가 훨씬 더 빠르게 진행될 것으로 전망하고 있다. 고령사회에서 정책은

훨씬 더 신중하게 수립되어야 한다는 점을 비추어 보았을때, 향후 주거용

전력소비를 줄이기 위하여 가격정책만을 고수하는 것은 한계가 있어 보인다.

안정적인 주거용 전력 수급을 위하여 고령화를 포함한 여러 요인들을

고려하는 것이 바람직할 것이다.

주요어 : 사회·인구학적 요인, 고령화, 주거용 전력 소비, 냉방도일, 난방도일,

패널공적분, 패널분석

학 번 : 2017-22870

Documents

Effects of socio-demographic factors on residential electricity … · 2019. 11. 14. · i Effects of socio-demographic factors on residential electricity consumption in Korea Yoori