Regional determinants of energy intensity in Japan: the impact of population density

Article
  • 99 Downloads

Abstract

The Japanese economy must contend with environmental restrictions; hence, both controlling greenhouse gas emissions by improving energy intensity and boosting national and regional economic growth are important policy goals. Given the potential conflicts between these goals, this study investigates the current energy consumption levels in the Japanese regional economy to determine the factors contributing to improvements in energy intensity. We conduct an empirical analysis using econometric methods to examine whether population density, which is considered a driving force of productivity improvements, contributes to improved energy intensity. The analysis results reveal that population density influences energy intensity improvements. However, the impact differs across regions. In large metropolitan areas, population agglomeration has improved energy intensity, whereas in rural areas, population dispersion has worsened it. The policy implication from this study is that population agglomeration should be encouraged in each region to improve energy intensity, which could protect the environment along with future economic growth.

Keywords

Energy intensity Population density Agglomeration Japanese regions 

JEL Classification

Q40 Q50 R10 

1 Introduction

In Japan, energy demand grew throughout the 1990s and 2000s. It increased at an average annual rate of 0.39% from 1990 to 2010, which was brought about by moderate economic growth (0.78% per annum). On the other hand, energy intensity, which is defined as units of energy per unit of GDP, has decreased by −0.39% per annum on average.1 In particular, energy intensity was greatly reduced in the 2000s compared to the 1990 s. However, there were regional differences in the reduction rate. While energy intensity significantly declined in large metropolitan areas, such as Tokyo, Kansai, and Chubu, the rate of decrease in rural areas was weak; conversely, energy intensity increased in some rural areas, such as Tohoku and Okinawa (Otsuka 2017a).

Boosting regional economic growth, while also curbing greenhouse gas (GHG) emissions via improving energy efficiency, is an important policy goal for Japan because most GHG emissions arise from energy use in modern society. The government has investigated the adoption of several policies for reducing GHGs, mainly centered on green innovation and renewable energy expansion. However, the hot-button issue of climate change has recently become too large to handle with only individual technologies. Japan must also pursue the United Nations Global Compact Cities initiative of urban and regional development, which is designed to reduce carbon emissions by adopting urban and regional policies, including new system design, system changes, and regulation or deregulation.

Population density is the key to decreasing energy intensity. This is because it is known to suppress energy intensity, as shown in a number of previous studies (Newman and Kenworthy 1989; Mindali et al. 2004; Bento and Cropper 2005; Brownstine and Golob 2009; Karathodorou et al. 2010; Su 2011). In Japan, population density has been known to reduce the energy intensity of the manufacturing and commercial sectors (Morikawa 2012; Otsuka et al. 2014). Boyd and Pang (2000) and Otsuka et al. (2014) empirically showed that productivity improvements themselves cause energy-intensity improvements. In other words, the authors claim that energy intensity serves as an indicator of productivity improvements. However, population density might also affect the energy intensity of the residential and transportation sectors as well as the manufacturing and commercial sectors. To accurately predict the future total energy demand of Japanese regions, it is necessary to analyze the energy intensity of the overall sectors. For this purpose, we focus on the energy intensity of all sectors, rather than individual sectors, in Japanese regions. In particular, to contribute to a review of the country’s Basic Energy Plan, it is essential to quantitatively evaluate the impact of population density on energy intensity.

This study aims to provide policy suggestions for the Basic Energy Plan and National Land Planning designed to reduce GHG emissions via improved energy intensity. Using Japanese regional data, we attempt to analyze the effect that population density (the driving force behind sustainable regional economic growth) has on the energy intensity in each regional economy. This study contributes to the existing literature in the following three ways.

First, we clarify the effect of population density on energy intensity from both static and dynamic perspectives. Previous studies that examine the relationship between energy intensity and population density used static models of cross-sectional or panel data (Morikawa 2012; Otsuka et al. 2014). However, population density not only affects the regional disparities of energy intensity, but also its temporal dynamic changes. Considering the interregional migration of the population, the influence of population density on energy intensity is thought to have not only an immediate (static) effect, but also a long-term (dynamic) effect. A static panel model does not allow us to conduct an adequate analysis of dynamic effects, whereas a dynamic panel model enables us to analyze both the short-run effects (short-run elasticity) and long-run effects (long-run elasticity) of population density on energy intensity.

Second, we clarify the difference of the impact of population density on the energy intensity among regions. Populations and firms are agglomerated in a narrow space in Japan’s large metropolitan areas, and the commercial sectors are relatively concentrated in these areas. Japanese commercial sectors have caused energy efficiency to deteriorate to low levels (Otsuka 2017a); therefore, these areas might have less impact on the energy intensity of population density. On the other hand, rural areas might be efficient compared to large metropolitan areas because the manufacturing sectors are concentrated. Manufacturing sectors, which account for a large proportion of the industrial sector, have a high level of energy efficiency, and this level has improved significantly (Otsuka 2017a). Otsuka et al. (2014) showed that the effect of industrial agglomeration on the improvement of energy intensity was significant in the manufacturing sectors. In rural areas, therefore, the impact of population density on energy intensity might be strong, owing to differences in the impact of industrial agglomerations, when compared to large metropolitan areas. We verify this hypothesis.

Third, we focus on the Japanese regional economy with a declining population, which is different from the situations discussed in previous studies. By performing the analysis on Japan, we can provide useful policy implications for energy policymakers in other developed countries that face similar challenges of a potential declining population or falling birth rate.

The remainder of this paper is structured as follows: Section 2 describes the methods and data for empirical analysis. Section 3 summarizes the results of the empirical analysis. Finally, Sect. 4 presents our conclusions and policy suggestions.

2 Methods

2.1 Determinants of energy intensity

In this study, energy intensity (ENERGY) is defined as energy consumption per unit of production value. The focus of this study was to evaluate the effectiveness of population density as a factor affecting energy intensity. Previous studies (e.g., Morikawa 2012; Otsuka et al. 2014) that analyze the relationship between population density and energy consumption efficiency for individual applications (e.g., the manufacturing or commercial sector) show a positive correlation between these factors. This finding suggests that the urban and regional structures represented by population density might improve a sector’s energy intensity, although the effect is not clear in terms of a region’s energy intensity.

Similar to previous studies, this study uses the gross population density (DENS), which is the total population divided by the habitable land area measured in km2. The habitable land area is calculated by deducting the forest, wildlife, and lake areas from the total land area. In addition, several socioeconomic variables that explain variations in energy intensity are included in our model. The selection of the socioeconomic variables is described as follows, and is based on variables used by several previous studies (e.g., Metcalf 2008; Wu 2012; Otsuka et al. 2014; Otsuka and Goto 2015a):

First, we consider energy price (P). According to economic theory, energy consumption decreases when energy prices increase, as long as price elasticity is not zero. Moreover, there is another effect caused by an increase in energy prices: the cost of production increases and producers might respond to it by improving energy intensity. Thus, if the energy market is functioning adequately, higher energy prices are expected to decrease energy intensity through more efficient or reduced energy use. Because of this relationship, the coefficient of P is expected to be negative.

Second, we consider per capita income (Y). This variable is included to reflect the level of economic development in the regions. According to economic theory, rising incomes create greater demand for energy, but also increase people’s ability to adopt more energy-efficient residential lifestyles. Thus, rising incomes would be expected to improve energy intensity. That is, the coefficient of Y is expected to be negative.

Third, we incorporate the capital–labor ratio (KL) to account for the effect of capital intensity on energy intensity. Thompson and Taylor (1995) and Metcalf (2008) showed that capital and energy have a substitution relationship over both the short and long run, whereas Antweiler et al. (2001) held that capital and pollution have a complementary relationship. Here, the capital–labor ratio is employed as a proxy for the level of technology involved. Thus, the KL variable might be negatively related to energy intensity. That is, energy intensity is expected to decline as production technology improves.

Fourth, we indirectly consider the effect of the capital stock’s vintage, which might reflect, to some extent, the speed of old machines and structures being replaced. New capital might be endowed with energy-saving technology and, thus, be more energy efficient. A particular regional industry’s low investment level in upgrading capital stock suggests that the industry’s energy intensity might also be high. Similarly, regional industries that are quick to invest in upgrading capital stock might replace it with more energy-efficient capital, thereby improving the industries’ energy intensity. To measure this capital vintage effect, we consider the investment–capital ratio (IK) (private-sector corporate capital investment divided by capital stock) for each year. The coefficient of this variable is expected to be negative.

Fifth, we introduce temperature data to account for the effects of temperature change on production activities. Specifically, we use cooling degree days (COOL) and heating degree days (HEAT). The annual number of cooling degree days is the cumulative difference of temperatures between 22 °C and the average temperature on each day in a year whose average temperature exceeds 24 °C, while the annual number of heating degree days is the cumulative difference of temperatures between 14 °C and the average temperature on each day in an annual period whose average temperature is below 14 °C. In the energy economic analysis, these indexes are usually used as variables representing cooling and heating, respectively (Metcalf 2008). These indexes are assumed to be related to energy consumption, and their use in this manner has precedent applications. For example, Metcalf and Hassett (1999) and Reiss and White (2008) used cooling and heating degree days to analyze energy consumption.

Finally, a time trend (Time) is also included in the model to capture the general trend of technology change over time, and is expected to have a negative coefficient.

2.2 Empirical models

This study uses the variables described in Sect. 2.1 above to empirically analyze the effect of population density on energy intensity. The first model, which is a revised version of the model used by Morikawa (2012) and Otsuka et al. (2014), is as follows:
$$\begin{aligned} \ln \left( {ENERGY_{jt} } \right) &= \beta_{1} \ln \left( {P_{t} } \right) + \beta_{2} \ln \left( {Y_{jt} } \right) + \beta_{3} \ln \left( {DENS_{jt} } \right) \\ \quad &+ \beta_{4} \ln \left( {KL_{jt} } \right) + \beta_{5} \ln \left( {KL_{jt} } \right)^{2} + \beta_{6} \ln \left( {IK_{jt} } \right) + \beta_{7} \ln \left( {IK_{jt} } \right)^{2} \\ \quad &+ \beta_{8} \ln \left( {COOL_{jt} } \right) + \beta_{9} \ln \left( {HEAT_{jt} } \right) + \beta_{10} {Time} + \alpha_{j} + u_{jt} \\ \end{aligned}$$
(1)

Note that all of the variables in Eq. (1) are expressed in natural logarithms. The subscript j represents the region (j = 1, …, J) and t represents the time (t = 1, …, T). ENERGY is the energy intensity (final energy consumption per unit of production value), P is the energy price, Y is per capita income, DENS is population density, KL is the capital–labor ratio (capital stock divided by the number of workers), and IK is the investment–capital ratio (annual private-sector corporate capital investment divided by capital stock). To consider the possibility of the non-linearity of the influence of the independent variables, the squared terms of those variables are added to the model.2 Furthermore, as previously mentioned, COOL is the cooling degree days, HEAT is the heating degree days, Time is the time trend, and u is an error term.

Both α and β are the estimated coefficients. Because panel data are used, α represents an individual effect. Energy price increases improve energy intensity and, thus, β1 is expected to have a negative sign. Income increases also improve energy intensity and, thus, β2 is also expected to have a negative sign. If population density improves energy intensity, then the sign of β3 will be negative, whereas if population density worsens energy intensity, its sign will be positive. Moreover, the signs of β4 and β5 will be negative if capital and energy have a substitution relationship, while those of β6 and β7 will be negative if upgrading capital improves energy intensity. Further, β10 is expected to have a negative sign because of technological development.

The problem with Eq. (1) is its assumption that energy intensity immediately reflects changes in economic variables. A more realistic assumption is that energy intensity is affected by changes in economic variables, such as energy price, after a certain time lag. Accordingly, this study considers the factors that affect energy intensity after a time lag by using a partial adjustment model as the second model.3

2.3 Partial adjustment model

Let \(y_{jt}^{ * }\) represent a desirable level of energy intensity in region j at time t. We assume that this desirable level is a function of an independent variable represented by vector \({\mathbf{x}}_{jt}\). The relationship is given by the following equation:
$$y_{jt}^{ * } = {\mathbf{x}}_{jt}^{{\prime }} {\varvec{\upbeta}} + \varepsilon_{jt} ,$$
(2)
where the error term is structured by \(\varepsilon_{jt} = \alpha_{j} + u_{jt}\) for j = 1,…,J and i = 1,…,I, which includes the region’s individual effect α. The adjustment process follows a partial adjustment model that relates the actual energy intensity \(y_{jt}\) to the desirable level of energy intensity \(y_{jt}^{*}\):
$$y_{jt} - y_{jt - 1} = \left( {1 - \lambda } \right)\left( {y_{jt}^{ * } - y_{jt - 1} } \right),$$
(3)
where λ is a measurement of the adjustment from the actual to desirable level of energy intensity.The following equation is obtained by combining Eqs. (2) and (3):
$$y_{jt} = {\mathbf{x^{\prime}}}_{jt} {\hat{\varvec{\beta }}} + \lambda y_{jt - 1} + \hat{\varepsilon }_{jt} ,$$
(4)
where \({\hat{\varvec{\beta }}} = \left( {1 - \lambda } \right){\varvec{\upbeta}}\) is the effect (short-run elasticity) that a short-run change in x has on y, and \({{{\hat{\varvec{\beta }}}} \mathord{\left/ {\vphantom {{{\hat{\mathbf{\beta }}}} {\left( {1 - \lambda } \right)}}} \right. \kern-0pt} {\left( {1 - \lambda } \right)}}\) is the effect (long-run elasticity) that a long-run change in x has on y. The error term is \(\hat{\varepsilon } = \left( {1 - \lambda } \right)\varepsilon\). In recursive Eq. (4), the lagged form of the explained variable is included as an explanatory variable. Therefore, a correlation exists between the explanatory variable and error term, creating the problem of inconsistency in the standard ordinary least squares estimates. To solve this problem, Arellano and Bond (1991) proposed a generalized method of moments for dynamic panel estimates that have consistency under these conditions. We use this dynamic panel estimate method to estimate the parameters of Eq. (4).

2.4 Data

Before describing the data source of this study, this section briefly describes the regional distribution of economic activity in 2010. In Japan, the geographical distribution of industries is unique. The population of the Greater Tokyo area (i.e., Saitama Prefecture, Chiba Prefecture, Tokyo, and Kanagawa Prefecture) accounts for 27.36% of the total national population, and production within this area is 32.30% of the total national production. However, the Greater Tokyo area accounts for only 7.34% of the total national livable land. In particular, the population of the Greater Tokyo area is significantly higher than that of the second most populous region, Kansai (16.25%; this region includes Shiga Prefecture, Kyoto, Osaka, Hyogo Prefecture, Nara Prefecture, and Wakayama Prefecture), and of the third most populous region, Chubu (13.46%; this region covers Gifu, Shizuoka, Aichi, and Mie Prefectures). In fact, it is roughly equivalent to the total population of the latter two regions. Similarly, production in the Greater Tokyo area is roughly equivalent to the total production in both Kansai (15.70%) and Chubu (14.14%). Both population and overall economic production show similar distributions, since both are highly concentrated in metropolitan areas.

The data analyzed in this study are the final energy consumption of Japan’s regions, which include 47 prefectures, from 1990 to 2010. These are, therefore, panel data by prefecture and year. All of the data are extracted from official publications of the Japanese government or prominent research institutions. The source of the prefectural final energy consumption data is the Energy Consumption Statistics by Prefecture (Agency for Natural Resources and Energy, Ministry of Economy, Trade and Industry 2015). The production-value data, which are used as the denominator in the calculation of energy intensity, are the actual total prefectural production, extracted from the Annual Report on Prefectural Accounts (Cabinet Office 2013). Energy price data are taken from those published by the International Energy Agency. Income figures are obtained from the Annual Report on Prefectural Accounts (Cabinet Office 2013) and converted to real figures based on the total prefectural expenditure deflator. Population and habitable surface area data are extracted from the Basic Resident Register Population and Society/Population Statistical Survey, respectively (Statistics Bureau, Ministry of Internal Affairs and Communications 2015a, 2015b). Other socioeconomic variables data are drawn from the Central Research Institute of Electric Power Industry’s regional economic database. Finally, the data on heating and cooling degree days are obtained from prefectural capitals and meteorological observation points.

Table 1 shows the summary statistics for the variables used in the empirical analysis. The average energy efficiency for the entire sample is 28.619 (GJ per million yen). Energy efficiency remained relatively constant during the 1990s and then improved significantly during the 2000s. The improvement rate during the entire observation period is 8.7%.
Table 1

Summary statistics

  

Energy intensity (GJ per million yen)

Energy price (2010 = 100)

Per-capita income (million yen)

Population density (people per km2)

Capital–labor ratio (million yen per

person)

Capital-investment ratio

Cooling degree days (degree days)

Heating degree days (degree days)

ENERGY

P

Y

DENS

KL

IK

COOL

HEAT

1990

 Avg.

29.520

94

2.518

1336

10.726

0.099

413

1048

 SD

19.796

0

0.469

1574

1.801

0.009

163

411

 Max.

105.394

94

4.542

8456

14.701

0.118

864

2239

 Min.

8.483

94

1.818

259

7.274

0.071

45

5

2000

 Avg.

29.280

85

2.750

1350

15.951

0.057

412

1140

 SD

17.638

0

0.446

1585

2.383

0.006

140

533

 Max.

87.916

85

4.953

8413

20.896

0.078

840

2769

 Min.

8.937

85

1.977

260

11.137

0.044

66

3

2010

 Avg.

26.953

100

2.879

1360

19.491

0.045

492

1267

 SD

15.745

0

0.442

1679

3.023

0.004

137

467

 Max.

83.860

100

4.854

9066

26.189

0.055

909

2591

 Min.

8.288

100

2.180

249

12.975

0.035

124

122

Total sample, 1990–2010

 Avg.

28.619

92

2.715

1355

15.481

0.061

367

1106

 SD

16.948

7

0.469

1592

3.506

0.016

176

471

 Max.

105.394

111

5.478

9066

26.189

0.118

1186

2769

 Min.

8.288

85

1.818

249

7.274

0.035

0

0

The average population density during the observation period is 1355 people per km2. The population density increased from the 1990s through the 2000s. This finding indicates that population agglomeration increased during the observation period, which suggests a high likelihood that the observed energy intensity improvements are caused by increased population agglomeration.

Among the socioeconomic variables other than population density, only energy price is a nationwide measure. Energy prices declined in the 1990s before increasing markedly in the 2000s. The average per capita income during the observation period was 2.715 million yen; the per capita income rose steadily during the 1990s and 2000s, and it is highly likely that higher incomes contributed to improved energy intensity. The average capital–labor ratio during the observation period was 15.481. The capital intensity increased from the 1990s through the 2000s; therefore, it is highly likely that product processes were increasingly mechanized. The average investment–capital ratio (0.061) during the period was low, declining from the 1990s through the 2000s. This suggests that little upgrading of production equipment occurred. The Appendix shows the correlation matrix of these explanatory variables.

Figure 1 illustrates the relationship between energy intensity (vertical axis) and population density (horizontal axis). The figure shows the values for each prefecture in the years 1990 and 2010. As indicated by the downward-sloping scatter plots from northwest to southeast, energy intensity and population density have a negative correlation. In other words, the higher a region’s population density, the lower its energy intensity. This negative relationship between energy intensity and population density is stable and applicable to the other years between 1990 and 2010.
Fig. 1

Energy intensity and population density

Figure 2 illustrates the relationship between the change in energy intensity (vertical axis) and that in population density (horizontal axis) between 1990 and 2010. Similar to energy intensity and population density, these change variables might also be negatively correlated, indicating that the increasing population density might have contributed to improvements in energy intensity.4 For example, in the Greater Tokyo area prefectures consisting of Tokyo, Saitama, Kanagawa, and Chiba, population density increased, while energy intensity decreased. Section 3 quantitatively examines the degree to which the changes in population density affect those in energy intensity.
Fig. 2

Changes in energy intensity and population density

All explanatory variables have been standardized and, thus, it is possible to compare each estimated coefficient and interpret the magnitude of the effect of different variables measured in different scales.

3 Results and discussion

Table 2 shows the results of the estimation of Eq. (1). Model A’s results are for the standard fixed-effects model, while Model B’s results are for the fixed-effects model estimated using the instrumental variable method. First, we performed an F-test to check whether fixed effects exist, and then we rejected the null hypothesis of the non-existence of individual effects at the 1% significance level. Furthermore, we performed a Hausman test in terms of the hypothesis that the observed individual effects were random effects; we rejected this null hypothesis at the 1% significance level. Therefore, the results shown in Table 2 are for fixed-effects models (Models A and B).
Table 2

Estimation results for Eq. (1) (static panel data models)

Variable

Coefficient

FE

FE-IV

Model A

Model B

ln(P)

β1

−0.019**

(−3.09)

−0.038*

(−2.05)

ln(Y)

β2

−0.156**

(−12.06)

−0.159**

(−6.74)

ln(DENS)

β3

−0.747**

(−4.91)

−0.915**

(−3.04)

ln(KL)

β4

−0.081**

(−2.80)

−0.219**

(−3.09)

ln(KL)2

β5

−0.033**

(−10.53)

−0.033**

(−4.25)

ln(IK)

β6

−0.023*

(−2.21)

−0.058*

(−2.06)

ln(IK)2

β7

−0.025**

(−6.69)

−0.030**

(−3.84)

ln(COOL)

β8

0.021**

(3.48)

−0.055

(−1.14)

ln(HEAT)

β9

−0.002

(−0.15)

−0.220

(−1.11)

Time

β10

0.002

(0.41)

0.018

(1.68)

Number of observations

 

987

940

F test

 

1791.2**

1732.5**

Hausman test

 

39.78**

  

FE fixed-effects model, FE-IV fixed-effects model using the instrumental variable method

** Significant at 1% level; * significant at 5% level

Values in parentheses are t-statistics

Explanatory variables lagged by one period are used as instrumental variables

The results in Table 2 show that population density is associated with low energy intensity. As both the explanatory and explained variables are logarithmic values, coefficients β1 through β10 represent degrees of elasticities. Accordingly, the larger the estimated coefficients (elasticities), the greater the effect of the corresponding explanatory variable on the explained variable. Therefore, as shown in Table 2, the effects of population density significantly exceed those of other explanatory variables. The coefficients of energy price and per capita income have negative signs and, thus, the sign condition mentioned previously is satisfied. In other words, increases in both energy prices and incomes decrease energy intensity. The coefficient of the capital–labor ratio is negative and, therefore, capital and energy consumption have a substitution relationship, meaning that a higher capital–labor ratio translates into lower energy intensity. In addition, the coefficient of the investment–capital ratio is negative, indicating that upgrading capital stock improves energy intensity. Although statistically significant, the estimated parameters are relatively much smaller than are those of income and density. As such, it can be said that the effect of the investment-capital ratio is negligible.

It is important to avoid an endogeneity problem in order to make the fixed-effects model estimates meaningful. In this study, we might need to consider the endogeneity problem between DENS and ENERGY because of the possible influence from some omitted variables in Eq. (1).5 Considering this problem, this study used the instrumental variable method to show the results (Model B) for estimating Eq. (1). The coefficient of population density in Model B is similar to that in Model A. Hence, even if the problem of endogeneity exists in the equation, it is reasonable to assume that the influence would be negligible.

We extend the fixed-effects model and estimated it following Eq. (5), by considering a regional dummy (dum) for population density.
$$\begin{aligned} \ln \left( {ENERGY_{jt} } \right) & = \beta_{1} \ln \left( {P_{t} } \right) + \beta_{2} \ln \left( {Y_{jt} } \right) + \left( {\beta_{3} + \delta dum} \right) \cdot \ln \left( {DENS_{jt} } \right) \\ \quad + \beta_{4} \ln \left( {KL_{jt} } \right) + \beta_{5} \ln \left( {KL_{jt} } \right)^{2} + \beta_{6} \ln \left( {IK_{jt} } \right) + \beta_{7} \ln \left( {IK_{jt} } \right)^{2} \\ \quad + \beta_{8} \ln \left( {COOL_{jt} } \right) + \beta_{9} \ln \left( {HEAT_{jt} } \right) + \beta_{10} {Time} + \alpha_{j} + u_{jt} \\ \end{aligned}$$
(5)
Figure 3 shows the regional distribution of population density. The population is concentrated in large metropolitan areas, such as Tokyo, Osaka, and Aichi Prefecture. In particular, there is a large disparity in changes in population density between these large metropolitan areas and rural areas (Fig. 4). This strongly suggests that different effects of changes in population density might exist on energy intensity among regions.
Fig. 3

Population density (Y2010, logarithmic value)

Fig. 4

Changes in population density (%, annual growth rate)

Table 3 shows the estimation results for the dummy variable model (static panel data models). A regional dummy variable takes a value of 1 in the case of an urban (large metropolitan) area, and 0 in the case of a rural area. The urban area is defined as large metropolitan areas, which include the Greater Tokyo area, Chubu, and Kansai, while the rural area is defined as regions that include areas other than these large metropolitan areas. In Model C, the regression coefficient of population density for large metropolitan areas is −0.2012 (= −1.2147 + 1.0135), and that for rural areas is −1.2147. This implies that the effect of population density on energy intensity is relatively higher in rural areas. The result supports the hypothesis that we presented in Sect. 1, that is, population agglomeration in rural areas significantly improves energy intensity. This reflects that manufacturing sectors, which have a high impact of industrial agglomeration on energy intensity, are concentrated in the rural areas.
Table 3

Estimation results for dummy variable model (static panel data models)

Variable

Coefficient

FE

FE-IV

Model C

Model D

ln(P)

β1

−0.021**

(−3.41)

−0.041*

(−2.20)

ln(Y)

β2

−0.158**

(−12.26)

−0.159**

(−6.89)

ln(DENS)

β3

−1.215**

(−6.19)

−1.237**

(−2.50)

dum

δ

1.014**

(3.73)

0.634

(1.11)

ln(KL)

β4

−0.083**

(−2.87)

−0.220**

(−3.21)

ln(KL)2

β5

−0.033**

(−10.68)

−0.032**

(−4.11)

ln(IK)

β6

−0.022*

(−2.15)

−0.055*

(−1.91)

ln(IK)2

β7

−0.025**

(−6.83)

−0.031**

(−3.90)

ln(COOL)

β8

0.021**

(3.42)

−0.056

(−1.19)

ln(HEAT)

β9

0.002

(0.13)

−0.200

(−0.96)

Time

β10

0.001

(0.28)

0.018

(1.74)

Number of observations

 

987

940

F-test

 

1784.5**

1711.3**

Hausman test

 

49.54**

 

FE fixed-effects model, FE-IV fixed-effects model using the instrumental variable method

** Significant at 1% level; * significant at 5% level

Values in parentheses are t-statistics

Explanatory variables lagged by one period are used as instrumental variables

Next, Table 4 shows the results of estimating Eqs. (4) and (5) in which dynamic changes in energy intensity are considered. According to the standard for dynamic panel estimates, Eqs. (4) and (5) were estimated using the level of the lagged explained variable (period t − 2) as the instrumental variable. When conducting the dynamic panel estimates, it is important to avoid serial correlation with the error term because consistent estimators are obtained only when this condition is satisfied. We performed Arellano–Bond tests to check for serial correlation. A significant first-order serial correlation is permitted in the AR (1) test, but the AR (2) test must show no statistically significant second-order serial correlation. Based on our tests, we could not reject the null hypothesis of no second-order serial correlation in Models E and F (see Table 4). In other words, the error term (ujt) did not exhibit the second-order self-correlation in either model. We then used the result of the Sargan–Hansen test (a test of exogeneity for instrumental variables) to ensure that the number of variables was not excessive and the conditions for dynamic panel estimates were satisfied. As expected, a statistically significant negative coefficient was obtained for the population density variable (representing population agglomeration). For the majority of other socioeconomic variables, such as energy price and per capita income, the sign condition was also satisfied and statistically significant values were obtained.
Table 4

Estimation results for Eq. (4) (dynamic panel data models)

Variable

Coefficient

Model E

Model F

ln(ENERGY)(−1)

λ

0.311**

(17.69)

0.327**

(14.83)

ln(P)

β1

−0.012*

(−2.01)

−0.013

(−1.86)

ln(Y)

β2

−0.082**

(−6.68)

−0.086**

(−6.51)

ln(DENS)

β3

−0.393*

(−2.33)

−0.714**

(−4.20)

dum

δ

 

0.978

(1.76)

ln(KL)

β4

−0.079

(−1.32)

−0.052

(−0.76)

ln(KL)2

β5

−0.020**

(−4.17)

−0.019**

(−4.38)

ln(IK)

β6

−0.023*

(−2.28)

−0.021

(−1.79)

ln(IK)2

β7

−0.008

(−1.43)

−0.006

(−1.03)

ln(COOL)

β8

0.011**

(4.06)

0.017**

(5.26)

ln(HEAT)

β9

0.008

(1.03)

0.005

(0.64)

Time

β10

0.003

(0.30)

−0.001

(−0.11)

Number of observations

 

893

893

J-statistic

 

41.189

41.259

Prob (J-statistic)

 

0.25

0.22

m-statistic (AR(1))

 

−2.94**

−5.65**

m-statistic (AR(2))

 

1.75

1.93

Instrumental variable

 

Energy (t − 2)

Energy (t − 2)

Estimates are two-step dynamic generalized method of moments estimates

** Significant at 1% level; * significant at 5% level

Values in parentheses are t-statistics

The J-statistic is the Sargan–Hansen test (a test of exogeneity) for the instrumental variable

The m-statistics are Arellano–Bond self-correlation tests for first- and second-order serial correlation

In this dynamic model, we estimated the model considering the influence of the regional dummy (dum) on population density. The estimation results are shown in Model F. According to the results, the regional dummy coefficients are not statistically significant. This is different from the result of the static model (Model C). This result suggests that the hypothesis (i.e., the effect of population density is different between metropolitan and rural areas) is not supported in the dynamic model. Therefore, the regional difference effects of population density on energy intensity arise solely from differences in regional data, and not from those in the estimated coefficients. The reason for these different results between the static and dynamic models, with respect to the regional dummy, might be population movement over time. Generally, people and firms tend to move among regions according to productive environmental disparities. In the short run, when population movement does not occur, there is a clear difference in the effects of population density between urban and rural regions. Meanwhile, in the long run, when the population moves freely among regions, inter-regional disparities in the effect would be diminished.

Table 5 shows the short- and long-run elasticities of each variable relative to energy intensity, which are calculated from the estimation results of Model E shown in Table 4. Population density has short- and long-run elasticities of −0.39 and −0.57, respectively, which are of bigger magnitude than the elasticities for energy price and per capita income. In other words, this finding suggests that when population density increases by 1%, energy intensity declines by about 0.4 and 0.6% over the short and long run, respectively. These estimated elasticities are smaller than the results obtained from static models, but the conclusion that population density improves energy intensity still holds. Energy price has short- and long-run elasticities of −0.01 and −0.02, respectively, which are smaller than are those for population density and per capita income. These relative elasticities suggest that changes in energy prices have less effect on energy intensity than does either income or population agglomeration.
Table 5

Factor elasticities relative to energy intensity

 

Short run

Long run

P

−0.01

−0.02

Y

−0.08

−0.12

DENS

−0.39

−0.57

KL

−0.12

−0.17

IK

−0.04

−0.06

COOL

0.01

0.02

HEAT

NS

NS

TIME

NS

NS

Each elasticity is calculated based on the results of Model E

NS not statistically significant

These findings demonstrate the existence of economies of population agglomeration on energy intensity, and this effect is relatively large. We obtained consistent signs of the estimated coefficients on population density regardless of using static or dynamic models. Therefore, we determined that urban and regional development policies that boost population agglomeration improve energy intensity.

The main question that our research sought to answer was the extent to which population density contributes to improvements in energy intensity. Under long-run equilibrium, actual energy intensity \(y_{jt}\) equals desirable energy intensity \(y_{jt}^{ * }\). In other words, \(y_{jt}^{ * } = y_{jt} = y_{jt - 1}\). This relationship can be used to express Eq. (4) as the following difference formula:
$$\Delta y_{jt} = \Delta {\mathbf{x^{\prime}}}_{jt} \frac{{{\hat{\varvec{\beta }}}}}{1 - \lambda } + \Delta \varepsilon_{jt} .$$
(6)
Using Eq. (6), we calculated each explanatory variable’s contribution (annual average, %) to changes in energy intensity. In addition, to measure the effect, we used the estimated coefficients of Model E, as shown in Table 4.
Table 6 shows these contribution levels. The largest contributing factor that reduced energy intensity in all regions is the capital–labor ratio. Population density is the second largest contributing factor for the Greater Tokyo area and Okinawa, and the third largest contributing factor in Kita-Kanto, Chubu, and Kansai. For example, these contributions from the capital–labor ratio and population density were −0.445 and −0.264%, respectively, for the Greater Tokyo area. From these results, we concluded that regions with large cities, with the exception of Kita-Kanto and Okinawa, have advanced population agglomeration over the study period, which, in combination with rising productivity, has improved energy intensity. Similarly, the majority of regions that contain large cities lack advanced population agglomeration and, thus, have not benefited from corresponding improvements in energy intensity. For example, energy intensity improved only modestly in Hokkaido, Kita-Kanto, and Chugoku regions, and even deteriorated in Tohoku and Okinawa regions.
Table 6

Contributions of each factor to changes in energy intensity (1990–2010; annual averages; %)

 

Rate of change of energy intensity (Δ Energy = a + b + c + d + e + f + g + h + i)

Δ Energy

Energy price

a

Per capita income

b

Population density

c

Capital–labor ratio

d

Capital vintage

e

Cooling degree days

f

Heating degree days

g

Time

h

Other factors

i

Hokkaido

−0.036

−0.005

−0.065

0.120

−0.401

0.136

0.054

0.009

0.023

0.094

Tohoku

0.155

−0.005

−0.109

0.169

−0.570

0.165

0.032

0.010

0.023

0.439

Kita-Kanto

−0.035

−0.005

−0.107

−0.030

−0.521

0.157

0.019

0.009

0.023

0.420

Greater Tokyo area

−0.241

−0.005

−0.028

−0.264

−0.445

0.166

0.017

0.014

0.023

0.282

Chubu

−0.934

−0.005

−0.094

−0.017

−0.523

0.160

0.011

0.010

0.023

−0.501

Hokuriku

−0.318

−0.005

−0.062

0.052

−0.430

0.146

0.016

0.013

0.023

−0.071

Kansai

−0.467

−0.005

−0.044

−0.016

−0.491

0.143

0.007

0.010

0.023

−0.093

Chugoku

−0.071

−0.005

−0.058

0.153

−0.557

0.152

0.006

0.014

0.023

0.202

Shikoku

−0.440

−0.005

−0.092

0.177

−0.572

0.141

0.007

0.008

0.023

−0.127

Kyushu

−0.471

−0.005

−0.115

0.072

−0.501

0.137

−0.002

0.014

0.023

−0.093

Okinawa

0.047

−0.005

−0.107

−0.225

−0.497

0.142

0.003

0.193

0.023

0.521

The prefectures included in each regional category are as follows: Hokkaido: Hokkaido, Tohoku: Aomori, Iwate, Miyagi, Akita, Yamagata, Fukushima, and Niigata, Kita-Kanto: Ibaraki, Tochigi, Gunma, and Yamanashi, Greater Tokyo area: Saitama, Chiba, Tokyo, and Kanagawa, Chubu: Nagano, Gifu, Shizuoka, Aichi, and Mie, Hokuriku: Toyama, Ishikawa, and Fukui, Kansai: Shiga, Kyoto, Osaka, Hyogo, Nara, and Wakayama, Chugoku: Tottori, Shimane, Okayama, Hiroshima, and Yamaguchi, Shikoku: Tokushima, Kagawa, Ehime, and Kochi, Kyushu: Fukuoka, Saga, Nagasaki, Kumamoto, Oita, Miyazaki, and Kagoshima, Okinawa: Okinawa

Each factor’s contribution was calculated based on the estimated results (Model E) from the dynamic panel analysis, as shown in Table 4

In addition to population density, other factors that contribute to improving energy intensity are per capita income and the capital–labor ratio. These factors are important in all regions, and the capital–labor ratio, in particular, explains the largest part of the observed changes in energy intensity. Conversely, energy price has little effect on energy intensity, as shown by the low price elasticity of energy demand. Thus, we concluded that energy intensity is affected more by income and capital intensity than by energy price.

4 Conclusion and policy implications

Japan’s aging and declining population, falling birthrates, and tougher environmental restrictions mean that finding a way to achieve regional economic growth is a crucial policy issue for the country. Grounded in the notion that Japan’s prefectures can combine improvements in energy efficiency and productivity for regional economic growth, this study presented an empirical analysis focusing on the effect that population agglomeration has on energy intensity.

Facing global competition, Japanese companies are developing energy-saving technologies. Consequently, the ratio of final energy consumption per unit of production value has continually declined. However, the degree of energy intensity improvement varies from region to region. For example, regions in which the commercial sector accounts for a relatively large share of the industrial structure have made insufficient progress. This suggests that the degree of energy intensity improvement in areas where the commercial sector is concentrated might be small. Previous studies conducted in Japan have shown that the higher a region’s population agglomeration rate, the lower the energy intensity in manufacturing sectors; yet, previous analyses have not adequately evaluated the nature of the effect that population agglomeration has on the entire region’s energy intensity over all sectors.

The results of this study show, for the first time, that population agglomeration brings about improvements in energy intensity across entire regions and sectors. In other words, population agglomeration brings about low energy intensity. We concluded that this improved energy intensity is created by the high elasticity enabled by population agglomeration. This finding may be applicable to other developed countries where population agglomeration is progressing. According to the United Nations (2015), the ratio of the urban population to the total population is greater in the UK, France, Germany, and the US than in Japan. In other words, urbanization is progressing in Europe and the US rather than in Japan. This implies that the improvement of energy intensity will be more advanced in European countries and the US, where higher levels of urbanization have been attained. An examination of the global trend would be an interesting research topic in the future.

Moreover, agglomeration actually reduces the costs of energy-efficiency improvements in two ways. First, proximity lowers the need for infrastructure, public transportation, and the provision of energy-saving services. Simultaneously, spillover effects, by which advances of one company’s efficiency spread rapidly to other companies, lower the overall cost threshold for improvements in energy intensity. In Japan, these effects are stronger in rural areas that contain many industrial cities compared to in large metropolitan areas. In this context, there is a difference in the industrial structure among areas.

To measure the population agglomeration effect on energy intensity, we calculated the degree to which population agglomeration contributes to changes in energy intensity during the observation period from 1990 to 2010. From the result, this study found that the population agglomeration effect is apparent in all regions. However, the trend varies dramatically from region to region. In urban areas with large cities, such as the Greater Tokyo area, Chubu, and Kansai, increased population inflow apparently decreases energy intensity, whereas in rural areas other than Okinawa and Kita-Kanto, its contribution is opposite, due to the decline in population density. Thus, we concluded that the increased population agglomeration results in decreased energy intensity in Japan for entire regions and sectors, but that the effect is not realized in rural areas, in particular, due to declining population density.

This conclusion suggests that the differences in regional population agglomeration may affect regional energy demand through regional energy intensity. This is because energy demand is calculated by multiplying energy intensity by GDP. In other words, in reviewing Japan’s Basic Energy Plan, we need to pay attention to regional energy intensity. Regional energy intensity is affected by various socioeconomic factors such as regional population and industry. Our evidence revealed that the influence of population agglomeration is strong, thus implying that it is necessary to pay attention to the influence of population density when planning national energy policies.

Otsuka and Goto (2015b) and Otsuka (2017b) revealed that many large cities that exist in the regions of the Greater Tokyo area, Chubu, and Kansai enjoy the benefits of agglomeration economies from population agglomerations. According to Fujita and Thisse (2002), these benefits stem from the diversity of the economic structure in these regions. Increasing the diversity of economic structure means improvement in both productivity and energy intensity (Parr 2002). Conversely, rural regions lack diversified cities that play a key role in their economies, and their population is widely dispersed. In order to improve the energy intensity of the whole country in the future, it is necessary to integrate the population by forming a major urban area in each rural region, so that the area can enjoy the economies brought about by population agglomeration. To enjoy the benefits without suffering from the uneconomical aspects of large-city agglomeration (e.g., overcrowding, soaring land prices, and environmental damage) (Marquez-Ramos 2016), a city should, ideally, not become too big by aggregating an excessive number of industries. Determining the appropriate size of a city from the population agglomeration perspective and guiding regions toward such a development path is an important policy issue. As a first step, the policy needs to form medium-sized cities that function as major urban areas in rural regions, in order to improve the energy intensity of Japan as a whole. Furthermore, it might be necessary to rectify the excess concentration of population in the Greater Tokyo area and encourage the migration of the population to rural areas. As such, in order to deepen further consideration, we should evaluate the role of city size in energy intensity from the urban systems perspective, considering the city size distribution. We believe that it is possible to provide clearer suggestions if we can identify the non-linear effects of population size on energy intensity.

Given these findings, this study suggests that growth strategies for urban development should be tailored by region. Exploring tailored strategies by using more detailed data is a research topic that we plan to pursue in the future. In particular, it needs to be further verified whether the hypothesis—that the effect of population density on energy intensity is different between cities and regions—is applicable to other countries by using a more extensive dataset of foreign countries. Moreover, our analysis does not consider spatial effects on energy intensity. For example, logistics, like cargo, generates spatial effects on energy intensity. To introduce spatial spillover effects in the study, we need to examine the energy consumption of cargo in detail, which is grasped in the transport sector. However, in Japan, energy consumption related to cargo is only measured at the country level. This is because it is difficult to distinguish between origin and landing in the energy consumption of cargo. Therefore, our analysis does not analyze the transport sector considering this data constraint. In order to capture the effect of logistics, it is necessary to accurately estimate the actual state of the energy consumption of cargo between regions. This is an important future extension that we should address.

Footnotes

  1. 1.

    Energy intensity is a measure of the energy efficiency of a nation or region’s economy. High energy intensity indicates a high price or cost of converting energy into GDP, while low energy intensity indicates a lower price or cost of converting energy into GDP.

  2. 2.

    We check the possible non-linearity of DENS in the estimation and confirm that a coefficient of the square term of DENS was not statistically significant at the 5% level. Therefore, we do not incorporate the square term of DENS into our models.

  3. 3.

    For a theoretical background on partial adjustment models, see Nordhaus (1979) and Cuddington and Dagher (2015).

  4. 4.

    The change in energy intensity had a correlation coefficient of −0.38 with the change in population density.

  5. 5.

    It is well known that endogeneity always exists when measuring population agglomeration using production functions (Graham 2009). As discussed in Otsuka and Goto (2015b), strong effects from endogeneity are unlikely when not using production functions to construct data on population agglomeration; however, for verification purpose, this study considers the endogeneity problem.

Notes

Acknowledgements

The authors thank reviewers whose comments have improved the quality of this study. This study was funded by Japan Society for the Promotion of Science (Grant No. 15K17067, 16K01236). In addition, Dr. Otsuka has received Grant-in-Aid as Young Scientific Research by Yokohama City University.

References

  1. Agency for Natural Resources and Energy, the Ministry of Economy, Trade and Industry (2015) Energy consumption statistics by prefectureGoogle Scholar
  2. Antweiler W, Copeland BR, Taylor MS (2001) Is free trade good for the environment? Am Econ Rev 91(4):877–908CrossRefGoogle Scholar
  3. Arellano M, Bond SR (1991) Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Rev Econ Stud 58:277–297CrossRefGoogle Scholar
  4. Bento AM, Cropper ML (2005) The effects of urban spatial structure on travel demand in the United States. Rev Econ Stat 87(3):466–478CrossRefGoogle Scholar
  5. Boyd GA, Pang JX (2000) Estimating the linkage between energy efficiency and productivity. Energy Policy 28(5):289–296CrossRefGoogle Scholar
  6. Brownstine D, Golob TF (2009) The impact of residential density on vehicle usage and energy consumption. J Urban Econ 65(1):91–98CrossRefGoogle Scholar
  7. Cabinet Office (2013) Annual report on prefectural accountsGoogle Scholar
  8. Cuddington JT, Dagher L (2015) Estimating short and long-run demand elasticities: a primer with energy-sector applications. Energy J 36(1):185–209CrossRefGoogle Scholar
  9. Fujita M, Thisse J (2002) The economics of agglomeration: cities, industrial location and regional growth. Cambridge University Press, CambridgeCrossRefGoogle Scholar
  10. Graham DJ (2009) Identifying urbanization and localization externalities in manufacturing and service industries. Pap Reg Sci 88(1):63–84CrossRefGoogle Scholar
  11. Karathodorou N, Graham DJ, Noland RB (2010) Estimating the effect of urban density on fuel demand. Energy Econ 32(1):86–92CrossRefGoogle Scholar
  12. Marquez-Ramos L (2016) The environmental impact of renewables. Int J Glob Warm, forthcomingGoogle Scholar
  13. Metcalf GE (2008) An empirical analysis of energy intensity and its determinants at the state level. Energy J 29(3):1–26CrossRefGoogle Scholar
  14. Metcalf GE, Hassett KA (1999) Measuring the energy savings from home improvement investment: evidence from monthly billing data. Rev Econ Stat 81(3):516–528CrossRefGoogle Scholar
  15. Mindali O, Raveh A, Salomon I (2004) Urban density and energy consumption: a new look at old statistics. Transp Res Part A: Policy Pract 38(2):143–162Google Scholar
  16. Morikawa M (2012) Population density and efficiency in energy consumption: an empirical analysis of service establishments. Energy Econ 34(5):1617–1622CrossRefGoogle Scholar
  17. Newman PWG, Kenworthy JR (1989) Gasoline consumption and cities: a comparison of US cities with a global survey. J Am Plan Assoc 55(1):24–37CrossRefGoogle Scholar
  18. Nordhaus WD (1979) The efficient use of energy resources. Yale University Press, New Haven and LondonGoogle Scholar
  19. Otsuka A (2017a) Regional energy demand and energy efficiency in Japan. Springer Nature, SwitzerlandCrossRefGoogle Scholar
  20. Otsuka A (2017b) Regional determinants of total factor productivity in Japan: stochastic frontier analysis. Ann Reg Sci 58(3):579–596CrossRefGoogle Scholar
  21. Otsuka A, Goto M (2015a) Estimation and determinants of energy efficiency in Japanese regional economies. Reg Sci Policy Pract 7(2):89–101CrossRefGoogle Scholar
  22. Otsuka A, Goto M (2015b) Agglomeration economies in Japanese industries: the Solow residual approach. Ann Reg Sci 54(2):401–416CrossRefGoogle Scholar
  23. Otsuka A, Goto M, Sueyoshi T (2014) Energy efficiency and agglomeration economies: the case of Japanese manufacturing industries. Reg Sci Policy Pract 6(2):195–212CrossRefGoogle Scholar
  24. Parr J (2002) Agglomeration economies: ambiguities and confusion. Environ Plan 34:717–731CrossRefGoogle Scholar
  25. Reiss PC, White MW (2008) What changes energy consumption? Prices and public pressures. RAND J Econ 39(3):636–663CrossRefGoogle Scholar
  26. Statistics Bureau, the Ministry of Internal Affairs and Communications (2015a) Basic resident register populationGoogle Scholar
  27. Statistics Bureau, the Ministry of Internal Affairs and Communications (2015b) Society/population statistical surveyGoogle Scholar
  28. Su Q (2011) The effect of population density, road network density, and congestion on household gasoline consumption in U.S. urban areas. Energy Econ 33(3):445–452CrossRefGoogle Scholar
  29. Thompson P, Taylor TG (1995) The capital-energy substitutability debate. Rev Econ Stat 77(3):565–569CrossRefGoogle Scholar
  30. United Nations (2015) World urbanization prospects. United Nations, New YorkGoogle Scholar
  31. Wu Y (2012) Energy intensity and its determinants in China’s regional economies. Energy Policy 41:703–711CrossRefGoogle Scholar

Copyright information

© The Japan Section of the Regional Science Association International 2017

Authors and Affiliations

  1. 1.Association of International Arts and ScienceYokohama City UniversityYokohamaJapan
  2. 2.School of Environment and SocietyTokyo Institute of TechnologyTokyoJapan

Personalised recommendations