**2. Materials and methods**

*Energy Policy*

analysis (DEA).

econometric models, including SFA.

There is no clear and accepted definition of energy efficiency, but according to Bhattacharyya [5], most definitions are based on a simple ratio of "useful output of a process/energy input into a process." Additionally, Patterson [6] shows several ways to quantify the output and input of this ratio. One of the ratios most frequently used in energy analysis at the macro level is the energy-GDP ratio, called energy intensity, which is in fact the reciprocal of the economic-thermodynamic index of energy efficiency identified by Patterson [6]. Energy intensity has been traditionally used as an indicator of energy efficiency. However, this approach has been disputed by the claims that energy intensity may not reflect the specific factors that enable energy intensity to accurately approximate energy efficiency [7–9]. An Energy Information Administration (EIA) [7] report first highlighted that energy intensity and efficiency are often used interchangeably and discussed the use of energy intensity as a measure of energy efficiency. Energy intensity is thus susceptible to socioeconomic factors other than energy efficiency, such as energy price, income, and production environment. Given this energy intensity problem, we need to control other important factors to obtain a pure measure of energy efficiency. Therefore, numerous studies attempted to measure the energy efficiency indices by conducting stochastic frontier analysis (SFA) and data envelopment

For instance, Huntington [10] discusses the relationship between energy and production efficiency using the framework of production theory. Feijoo et al. [11] conduct SFA to measure the energy efficiency of Spanish industries and Buck and Young [12] to estimate the energy efficiency of commercial buildings in Canada. Similarly, Boyd [13] analyzes the energy efficiency of wet corn milling plants and highlights the advantage of not having to define the problem of energy intensity in an SFA. Further, Zhou and Ang [14] measure the energy efficiency of 21 OECD countries using DEA. On the other hand, Filippini and Hunt measure the energy efficiency of 29 OECD countries [15] and calculate the energy efficiency of the US household sector using SFA [16]. The authors show that the energy efficiency level measured by conducting an SFA is not correlated with energy intensity, thus concluding that energy intensity is not a suitable proxy for energy efficiency. Carvalho [17] follows a time frame similar to that of Filippini and Hunt [15] and covers a series of non-OECD countries. Aranda-Uson et al. [18] perform an SFA to measure the energy efficiency for Spain's grocery and tobacco manufacturing, textile, chemical, and nonferrous metal product manufacturing industries. China-based studies have also applied SFA to measure the energy efficiency of the thermal power [19], iron and steel, and chemical industries [20, 21]. Lin and Du [22] and Filippini and Lin [23] compare energy efficiency levels across Chinese provinces using various

In sum, numerous studies support the use of an SFA instead of energy intensity as an indicator of energy efficiency. Moreover, SFA is a parametric approach that can tackle statistical noise and thus, is more desirable than DEA, a nonparametric approach. To this effect, Zhou et al. [24] evaluate the energy efficiency index using both approaches and show SFA is more desirable than DEA. A large body of research focuses on measuring energy efficiency values using SFA, whereas few studies explore the individual factors determining energy efficiency levels, such as the empirical works by Otsuka [25, 26]. These studies analyze the energy consumption trends of households and reveal that resident characteristics determine energy and electricity efficiency. However, to the best of the author's knowledge, there is a scarcity of research on economic production sectors. Particularly, how mechaniza-

tion and electrification affect the energy efficiency have not been clarified.

the determinants of the improvements in energy efficiency for Japan's industrial

This study thus measures the level of energy efficiency by using SFA and clarifies

**86**

#### **2.1 Econometric model for energy efficiency**

This study assumes the following aggregated energy demand function, *f*, exists at the Japanese prefectural level. That is,

$$E\_{\rm jet} = f\{P\_t, Y\_{\rm ft}, KL\_{\rm jt}, IK\_{\rm jt}, CDD\_{\rm jt}, HDD\_{\rm jt}, EF\_{\rm jt}\}\tag{1}$$

where *j* denotes a region (j = 1, …, J), *t* is time (t = 1, …, T) and *E* is the final energy consumption for the industrial sector. *P* is the energy price index for the sector and *Y* income. *KL* is the capital-labor ratio and represents the degree of mechanization in a factory or office. Thompson and Taylor [30] show that capital and energy both have short- and long-term relationships. *IK* is the proportion of investment in capital stock and represents the degree of vintage. *CDD* and *HDD* are the cooling and heating degree days and represent temperature. In regions with severe temperatures, energy consumption is more likely to be associated with air conditioning. Previous studies have shown that *CDD* and *HDD*, as indicators of cooling and heating, are related to energy consumption [31, 32]. *EF* is the level of energy efficiency in a region.

It is necessary to estimate energy efficiency, particularly because it is not directly observable in an economic system. Therefore, this study estimates energy efficiency using a stochastic frontier energy demand function. Stochastic frontier functions generally measure the economic performance of production and operation processes and have therefore been applied to production or cost theory using an econometric approach. This approach is based on the notion that frontier functions produce the maximum output or minimum cost levels achievable by a producer. In a production function, the frontier represents the maximum production level for a given input. In a cost function, the frontier is the minimum cost for a given output. An energy demand function can thus be considered similar to a cost function. In other words, the difference between observed energy demand and minimized demand is the technical inefficiency observed when the output for a production activity is given. In an aggregate energy demand function, the frontier denotes the

minimum energy level needed for the production activities in a region to achieve a given production level. In other words, by estimating an energy demand frontier function, it is possible to determine the baseline energy demand that reflects the energy demand in a region that is efficiently managing energy use through its production and operational processes. Additionally, it allows us to ascertain whether a region is on the frontier. If a region is not on the frontier, the distance from the frontier indicates the rate of energy consumption exceeding baseline demand (i.e., energy inefficiency) [33].

The panel SFA in this study follows the premise of Aigner et al. [34]. Further, this study adopts the one-step approach of Battese and Coelli [35]. It thus estimates the energy frontier function and the determinants of the energy inefficiency term simultaneously. Traditionally, a two-step estimation method is adopted, in which inefficiency is obtained by estimating the stochastic frontier function, and the value is regressed by determinants. In this case, a contradiction arises between the assumption of the distribution on the inefficiency term of the frontier function and the regression analyzing the inefficiency determinant. As such, the consistency of the estimation result is not guaranteed [36]. By adopting the one-step approach, we can avoid this problem. An SFA model using this approach approximates an economy's energy efficiency level based on a one-sided non-negative error term. That is, this study assumes the log-log function type in Eq. (1) can be specified as follows:

$$\begin{array}{rcl} \ln E\_{\text{jt}} & = & a + a\_{\text{P}} \ln P\_{\text{t}} + a\_{\text{Y}} \ln Y\_{\text{jt}} + a\_{\text{KL}} \ln \text{KL}\_{\text{jt}} + a\_{\text{IK}} \ln \text{IK}\_{\text{jt}} \\ & + a\_{\text{CDD}} \ln \text{CDD}\_{\text{jt}} + a\_{\text{HDD}} \ln \text{HDD}\_{\text{jt}} + v\_{\text{jt}} + u\_{\text{jt}}, \end{array} \tag{2}$$

where α is an estimated parameter. The error term (*vjt* + *ujt*) consists of two parts, a random error term *vjt* and an error term for inefficiency, *ujt*. It is assumed that *vjt* has a distribution *N*(0,σ<sup>2</sup> ) and is independent of *ujt* and all explanatory variables. *ujt* is a non-negative random variable and follows the distribution *N*(*μ*,σ*<sup>u</sup>* 2 ). *ujt* indicates that the efficiency energy level *EF* in Eq. (1) is an energy inefficiency index. Given Eq. (2), the energy efficiency level *EFjt* is estimated using the conditional expectation *E*(*ujt*|*vjt* + *ujt*) for the efficiency term [37]. Specifically, the energy efficiency level *EFjt* is measured by the ratio of the estimated energy demand frontier *Ejt F* to the observed energy demand *Ejt*. In other words, *EFjt* ≡ *Ejt F* /*Ejt* = *e* −*ujt* ,0 < *EFjt* ≤ 1.

Improvements in energy efficiency can be achieved through social innovation in the production and consumption processes of energy services, as well as the technical and organizational factors of energy demand. Average energy efficiency in this study is formulated as:

$$
\mu\_{\rm jet} = \beta + \beta\_{\rm KL} \ln \text{KL}\_{\rm jet} \star \beta\_{\rm ER} \ln \text{ER}\_{\rm jet}, \tag{3}
$$

**89**

model.

**2.3 Data**

*Determinants of Energy Demand Efficiency: Evidence from Japan's Industrial Sector*

industries, this study considers capital-labor ratio. The coefficient values for *KL* are

Regions that use coal and kerosene tend to report higher carbon dioxide emissions than those using electricity. Further, areas with a low electrification rate are considered wasteful in terms of energy use. Electrification of factories and offices enables an efficient use of energy. For example, a factory energy management system (FEMS) can be introduced to electrify a factory. A FEMS functions in coordination with power generation, power storage, and energy saving devices, allowing for energy saving that industries have been unable to hitherto realize. Furthermore, the implementation of a building energy management system (BEMS) for commercial buildings could reduce energy consumption and control energy-related facilities. Consequently, energy efficiency could increase with a rise in electricity usage through promoting electrification. Therefore, the coefficient values for *ER* are

Electrification can significantly influence the improvement of energy efficiency in a region. Therefore, this study conducts a quantitative analysis as an additional regression that account for the characteristics of factories and offices that may be

The variables in the following equation are assumed to be determinants of a

+ *δHDD* ln*HDDjt* + *δ<sup>j</sup>* + ε*jt*, (4)

where *j* is a region (j = 1, …, J), *t* is the time (t = 1, …, T), *ER* is the electrification rate in the industrial sector, and *LN is* the number of employees per establishment, comprising offices and factories, and denotes the scale of an establishment. *OR* is the ratio of the number of offices to that of establishments; *TFP* is the total factor productivity and represents an establishment's productivity level; *CDD* and *HDD* are cooling and heating degree days, respectively; and *δ* is an estimated parameter. Since this study uses panel data, δj denotes the fixed effect. In estimation of (4), it is necessary to consider endogeneity between the productivity and the electrification rate. It would be possible that a higher electrification rate also influences productivity. Although these endogeneity effects can be treated with a fixed effect model, it is not sufficient. To obtain robust results, this study calculates the estimates by panel GMM using instrumental variables in addition to the fixed effect

The data used for the analysis are 1990–2010 panel data for 47 prefectures. Data on the final energy consumption (*E*) of the sectors of each prefecture are taken from the Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade, and Industry). The energy price index (*P*) is estimated using the real energy price index for the respective sector by the International Energy Agency (IEA). Income (*Y*) is a real gross regional expenditure, data for which are available in the Annual Report on Prefectural Accounts (Cabinet Office). The capital-labor ratio (*KL*) is the ratio of capital stock to the number of employees, and data for the number of employed persons are available in the Annual Report on Prefectural Accounts

ln*ERjt* = *δLN* ln*LNjt* + *δOR* ln*ORjt* + *δTFP* ln*TFPjt* + *δCDD* ln*CDDjt*

*DOI: http://dx.doi.org/10.5772/intechopen.81482*

**2.2 Determinant model for the electrification rate**

expected to be positive.

expected to be negative.

electrification determinants.

region's electrification rate:

where *β* is an estimated parameter, *KL* is the capital-labor ratio, and *ER* is the electrification rate for the industrial sector. If the factor of the inefficiency term improves the efficiency, the sign of *β* is negative.

Factories and offices with large-scale facilities have high energy consumption and low energy efficiency in production. For example, a petrochemical complex, the paper pulp manufacturing industry, and the steel industry have large-scale production facilities. Therefore, the energy efficiency levels of these industries are low. Meanwhile, labor-intensive factories and offices have compact-scale production facilities, thus low energy consumption and high energy efficiency. For example, labor-intensive process-assembled industries are more energy efficient than material-based industries [38]. To control the differences in local production

*Determinants of Energy Demand Efficiency: Evidence from Japan's Industrial Sector DOI: http://dx.doi.org/10.5772/intechopen.81482*

industries, this study considers capital-labor ratio. The coefficient values for *KL* are expected to be positive.

Regions that use coal and kerosene tend to report higher carbon dioxide emissions than those using electricity. Further, areas with a low electrification rate are considered wasteful in terms of energy use. Electrification of factories and offices enables an efficient use of energy. For example, a factory energy management system (FEMS) can be introduced to electrify a factory. A FEMS functions in coordination with power generation, power storage, and energy saving devices, allowing for energy saving that industries have been unable to hitherto realize. Furthermore, the implementation of a building energy management system (BEMS) for commercial buildings could reduce energy consumption and control energy-related facilities. Consequently, energy efficiency could increase with a rise in electricity usage through promoting electrification. Therefore, the coefficient values for *ER* are expected to be negative.

#### **2.2 Determinant model for the electrification rate**

Electrification can significantly influence the improvement of energy efficiency in a region. Therefore, this study conducts a quantitative analysis as an additional regression that account for the characteristics of factories and offices that may be electrification determinants.

The variables in the following equation are assumed to be determinants of a region's electrification rate:

$$\begin{array}{rcl} \ln ER\_{\text{jt}} &=& \delta\_{LN} \ln LN\_{\text{jt}} + \delta\_{OR} \ln OR\_{\text{jt}} + \delta\_{TFP} \ln TFP\_{\text{jt}} + \delta\_{CDD} \ln CDD\_{\text{jt}}\\ &+ \ \delta\_{HDD} \ln HDD\_{\text{jt}} + \delta\_{\text{j}} + \mathbf{e}\_{\text{jt}},\end{array} \tag{4}$$

where *j* is a region (j = 1, …, J), *t* is the time (t = 1, …, T), *ER* is the electrification rate in the industrial sector, and *LN is* the number of employees per establishment, comprising offices and factories, and denotes the scale of an establishment. *OR* is the ratio of the number of offices to that of establishments; *TFP* is the total factor productivity and represents an establishment's productivity level; *CDD* and *HDD* are cooling and heating degree days, respectively; and *δ* is an estimated parameter. Since this study uses panel data, δj denotes the fixed effect. In estimation of (4), it is necessary to consider endogeneity between the productivity and the electrification rate. It would be possible that a higher electrification rate also influences productivity. Although these endogeneity effects can be treated with a fixed effect model, it is not sufficient. To obtain robust results, this study calculates the estimates by panel GMM using instrumental variables in addition to the fixed effect model.

#### **2.3 Data**

*Energy Policy*

follows:

a distribution *N*(0,σ<sup>2</sup>

study is formulated as:

energy inefficiency) [33].

minimum energy level needed for the production activities in a region to achieve a given production level. In other words, by estimating an energy demand frontier function, it is possible to determine the baseline energy demand that reflects the energy demand in a region that is efficiently managing energy use through its production and operational processes. Additionally, it allows us to ascertain whether a region is on the frontier. If a region is not on the frontier, the distance from the frontier indicates the rate of energy consumption exceeding baseline demand (i.e.,

The panel SFA in this study follows the premise of Aigner et al. [34]. Further, this study adopts the one-step approach of Battese and Coelli [35]. It thus estimates the energy frontier function and the determinants of the energy inefficiency term simultaneously. Traditionally, a two-step estimation method is adopted, in which inefficiency is obtained by estimating the stochastic frontier function, and the value is regressed by determinants. In this case, a contradiction arises between the assumption of the distribution on the inefficiency term of the frontier function and the regression analyzing the inefficiency determinant. As such, the consistency of the estimation result is not guaranteed [36]. By adopting the one-step approach, we can avoid this problem. An SFA model using this approach approximates an economy's energy efficiency level based on a one-sided non-negative error term. That is, this study assumes the log-log function type in Eq. (1) can be specified as

ln*Ejt* = *α* + *α<sup>P</sup>* ln*Pt* + *α<sup>Y</sup>* ln*Yjt* + *αKL* ln*KLjt* + *αIK* ln*IKjt*

where α is an estimated parameter. The error term (*vjt* + *ujt*) consists of two parts, a random error term *vjt* and an error term for inefficiency, *ujt*. It is assumed that *vjt* has

that the efficiency energy level *EF* in Eq. (1) is an energy inefficiency index. Given Eq. (2), the energy efficiency level *EFjt* is estimated using the conditional expectation *E*(*ujt*|*vjt* + *ujt*) for the efficiency term [37]. Specifically, the energy efficiency level *EFjt* is measured by the ratio of the estimated energy demand frontier *Ejt*

Improvements in energy efficiency can be achieved through social innovation in the production and consumption processes of energy services, as well as the technical and organizational factors of energy demand. Average energy efficiency in this

*μjt* = *β* + *βKL* ln*KLjt* + *βER* ln*ERjt*, (3)

where *β* is an estimated parameter, *KL* is the capital-labor ratio, and *ER* is the electrification rate for the industrial sector. If the factor of the inefficiency term

Factories and offices with large-scale facilities have high energy consumption and low energy efficiency in production. For example, a petrochemical complex, the paper pulp manufacturing industry, and the steel industry have large-scale production facilities. Therefore, the energy efficiency levels of these industries are low. Meanwhile, labor-intensive factories and offices have compact-scale production facilities, thus low energy consumption and high energy efficiency. For example, labor-intensive process-assembled industries are more energy efficient than material-based industries [38]. To control the differences in local production

a non-negative random variable and follows the distribution *N*(*μ*,σ*<sup>u</sup>*

observed energy demand *Ejt*. In other words, *EFjt* ≡ *Ejt*

improves the efficiency, the sign of *β* is negative.

+ *αCDD* ln*CDDjt* + *αHDD* ln*HDDjt* + *vjt* + *ujt*, (2)

) and is independent of *ujt* and all explanatory variables. *ujt* is

*F* /*Ejt* = *e* 2

,0 < *EFjt* ≤ 1.

−*ujt*

). *ujt* indicates

*F* to the

**88**

The data used for the analysis are 1990–2010 panel data for 47 prefectures. Data on the final energy consumption (*E*) of the sectors of each prefecture are taken from the Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade, and Industry). The energy price index (*P*) is estimated using the real energy price index for the respective sector by the International Energy Agency (IEA). Income (*Y*) is a real gross regional expenditure, data for which are available in the Annual Report on Prefectural Accounts (Cabinet Office). The capital-labor ratio (*KL*) is the ratio of capital stock to the number of employees, and data for the number of employed persons are available in the Annual Report on Prefectural Accounts


*Sources: For final energy consumption, see Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade and Industry: http://www.enecho.meti.go.jp/statistics/energy\_consumption/ec002/); for energy price index, see International Energy Agency databases; for income, see Annual Report on Prefectural Accounts (Cabinet Office: http://www.esri.cao.go.jp/jp/sna/sonota/kenmin/kenmin\_top.html); for capital-labor ratio, see Central Research Institute of Electric Power Industry databases; for vintage, see Central Research Institute of Electric Power Industry databases; for electrification rate, see Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade and Industry: http://www.enecho.meti.go.jp/statistics/energy\_consumption/ec002/); for establishment size and office ratio, see Economic Census (Statistics Bureau, Ministry of Internal Affairs and Communications: http:// www.stat.go.jp/data/e-census/index.html); and for TFP index, see Otsuka and Goto [29]: https://link.springer.com/ article/10.1007/s00168-016-0745-x.*

#### **Table 1.**

*Descriptive statistics.*


**91**

high here.

descriptive statistics.

*Determinants of Energy Demand Efficiency: Evidence from Japan's Industrial Sector*

**Electrification rate (%) Establishment size** 

**(person)**

46.78 9.53 95.23 0.27

30.67 11.08 96.82 0.40

**Region ER LN OR TFP** Hokkaido 34.69 9.84 97.67 0.25 Tohoku 44.58 9.27 96.33 0.15

Chubu 44.25 9.70 94.77 0.26 Hokuriku 51.66 8.84 95.00 0.22 Kansai 40.63 9.34 95.82 0.32 Chugoku 28.35 9.53 96.47 0.19 Shikoku 38.06 8.55 96.59 0.21 Kyushu 38.78 9.53 97.13 0.15 Okinawa 49.24 8.24 98.20 0.14

**Office ratio (%) TFP** 

**index**

(Cabinet Office). Capital vintage (*IK*) is the ratio of capital investment to capital stock. Data on capital investment and stock are based on the data published by the Central Research Institute of Electric Power Industry. Data on *CDD* and *HDD* are from the prefectural government's location and weather station—cooling degree day is the sum of the difference between average temperature on the days exceeding 24 and 22°C, while heating degree day is the sum of the difference between average temperatures below 14°C and above 14°C. The *ER* is estimated from the data in the Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade, and Industry). The estimation for the percentage of offices for all establishments (*OR*) is based on the number of business establishments listed by the Economic Census (Ministry of Economy, Trade, and Industry). Data for productivity (*TFP*) are the total factor productivity calculated by Otsuka and Goto [39]. **Table 1** presents the

*Notes: the regional classification is as follows: Hokkaido (Hokkaido), Tohoku (Aomori, Iwate, Miyagi, Akita, Yamagata, Fukushima, and Niigata), Tokyo (Saitama, Chiba, Tokyo, Kanagawa, Ibaraki, Tochigi, Gunma, and Yamanashi), Hokuriku (Toyama, Ishikawa, and Fukui), Chubu (Nagano, Gifu, Shizuoka, Aichi, and Mie), 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,* 

**Table 2** presents the regional characteristics for Japan as of 2010. Particularly, large metropolitan areas, such as the Greater Tokyo Area, Kansai, and Chubu, report high energy consumption. Moreover, the income scale is large and vintage is high in these areas. The capital-labor ratio is relatively high because the manufacturing industry is concentrated in the Chubu and Hokuriku regions. The degree of air conditioning usage is significant in the warm western Japan, and the number of heating days is high in eastern Japan. Further, the Greater Tokyo Area, Kansai, and Chubu have several large-scale business establishments and productivity tends to be

*DOI: http://dx.doi.org/10.5772/intechopen.81482*

*Oita, Miyazaki, and Kagoshima), and Okinawa (Okinawa).*

*Regional characteristics for Japan (as of 2010).*

**Panel B**

Kita-Kanto

Greater Tokyo area

**Table 2.**


*Determinants of Energy Demand Efficiency: Evidence from Japan's Industrial Sector DOI: http://dx.doi.org/10.5772/intechopen.81482*

*Notes: the regional classification is as follows: Hokkaido (Hokkaido), Tohoku (Aomori, Iwate, Miyagi, Akita, Yamagata, Fukushima, and Niigata), Tokyo (Saitama, Chiba, Tokyo, Kanagawa, Ibaraki, Tochigi, Gunma, and Yamanashi), Hokuriku (Toyama, Ishikawa, and Fukui), Chubu (Nagano, Gifu, Shizuoka, Aichi, and Mie), 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), and Okinawa (Okinawa).*

#### **Table 2.**

*Energy Policy*

Final energy consumption (TJ)

Energy price index (2010 = 100)

Establishment size (person)

**90**

**Panel A**

*Descriptive statistics.*

**Table 1.**

Kita-Kanto

Greater Tokyo area

**Final energy consumption (TJ)**

*article/10.1007/s00168-016-0745-x.*

**Energy price index (2010 = 100)** **Income (JPY, millions)**

**Description Variable Mean Std. dev. Maximum Minimum**

Income (JPY, millions) Y 10,422,755 14,063,661 100,931,767 1,865,830 Capital-labor ratio KL 15.48 3.51 26.19 7.27 Vintage IK 0.061 0.016 0.118 0.035 Cooling degree day CDD 367.0 175.6 1186.1 0.0 Heating degree day HDD 1106.3 470.9 2769.2 0.2 Electrification rate (%) ER 36.18 12.08 59.29 8.55

Office ratio (%) OR 94.92 1.62 98.20 89.67 TFP index TFP 0.185 0.112 0.662 −0.055 *Sources: For final energy consumption, see Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade and Industry: http://www.enecho.meti.go.jp/statistics/energy\_consumption/ec002/); for energy price index, see International Energy Agency databases; for income, see Annual Report on Prefectural Accounts (Cabinet Office: http://www.esri.cao.go.jp/jp/sna/sonota/kenmin/kenmin\_top.html); for capital-labor ratio, see Central Research Institute of Electric Power Industry databases; for vintage, see Central Research Institute of Electric Power Industry databases; for electrification rate, see Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade and Industry: http://www.enecho.meti.go.jp/statistics/energy\_consumption/ec002/); for establishment size and office ratio, see Economic Census (Statistics Bureau, Ministry of Internal Affairs and Communications: http:// www.stat.go.jp/data/e-census/index.html); and for TFP index, see Otsuka and Goto [29]: https://link.springer.com/*

E 199,829 214,836 1,181,999 24,530

P 86.5 8.9 111.5 77.7

LN 9.34 0.87 12.12 7.08

**Region E P Y KL IK CDD HDD** Hokkaido 322,771 100.00 19,199,451 15.09 0.035 124.0 2591.2 Tohoku 84,446 100.00 5,948,899 18.69 0.042 315.4 1907.7

Chubu 234,453 100.00 14,956,162 21.40 0.046 511.0 1270.3 Hokuriku 55,461 100.00 4,209,377 20.81 0.043 476.2 1522.8 Kansai 211,767 100.00 13,566,039 20.86 0.047 556.5 1116.0 Chugoku 269,740 100.00 5,900,042 20.52 0.045 539.2 1194.3 Shikoku 75,918 100.00 3,549,752 19.05 0.046 572.4 910.6 Kyushu 138,517 100.00 6,618,666 18.32 0.047 545.1 911.8 Okinawa 38,462 100.00 3,850,416 12.97 0.052 909.0 122.2

**Capitallabor ratio**

170,132 100.00 7,908,596 19.67 0.046 450.4 1407.0

613,806 100.00 41,983,820 19.21 0.046 492.5 1060.9

**Vintage Cooling** 

**degree day**

**Heating degree day**

*Regional characteristics for Japan (as of 2010).*

(Cabinet Office). Capital vintage (*IK*) is the ratio of capital investment to capital stock. Data on capital investment and stock are based on the data published by the Central Research Institute of Electric Power Industry. Data on *CDD* and *HDD* are from the prefectural government's location and weather station—cooling degree day is the sum of the difference between average temperature on the days exceeding 24 and 22°C, while heating degree day is the sum of the difference between average temperatures below 14°C and above 14°C. The *ER* is estimated from the data in the Energy Consumption Statistics by Prefecture (Ministry of Economy, Trade, and Industry). The estimation for the percentage of offices for all establishments (*OR*) is based on the number of business establishments listed by the Economic Census (Ministry of Economy, Trade, and Industry). Data for productivity (*TFP*) are the total factor productivity calculated by Otsuka and Goto [39]. **Table 1** presents the descriptive statistics.

**Table 2** presents the regional characteristics for Japan as of 2010. Particularly, large metropolitan areas, such as the Greater Tokyo Area, Kansai, and Chubu, report high energy consumption. Moreover, the income scale is large and vintage is high in these areas. The capital-labor ratio is relatively high because the manufacturing industry is concentrated in the Chubu and Hokuriku regions. The degree of air conditioning usage is significant in the warm western Japan, and the number of heating days is high in eastern Japan. Further, the Greater Tokyo Area, Kansai, and Chubu have several large-scale business establishments and productivity tends to be high here.
