Evaluation of environmental energy efficiency and its influencing factors: a prefecture-level analysis of Japanese manufacturing industries

Abstract

This study evaluates the progress of efficient energy use and the control of carbon dioxide (CO₂) emissions in Japan between 1990 and 2012. A new indicator of energy performance is presented called environmental energy efficiency (EEE). The EEE of manufacturing industries was measured by each prefecture in Japan. We estimated the influencing factors of EEE for each industry by applying the pooled mean group (PMG) method. Our findings are as follows: First, the Japanese manufacturing industry has not been in line with the EEE improvement goals since the adoption of the Kyoto Protocol. However, the progress of each industry was relatively consistent by region. Second, EEE tends to improve and then deteriorate or monotonically increase as economic development progresses. Third, EEE is raised by expanding industry share. Finally, EEE, which focuses on energy reduction, is likely to increase with the progress of energy-saving technology.

1 Introduction

Since the adoption of the Kyoto Protocol in 1997, Japan has promoted efforts to reduce greenhouse gas emissions to mitigate global warming. The Protocol imposed an obligation on Japan to reduce greenhouse gas emissions by 6% by 2012 compared to its 1990 level. This reduction target, however, was not achieved by that deadline. Energy-related carbon dioxide (CO₂) emissions increased from 1068 million tons in 1990 to 1227 million tons in 2012, and the manufacturing sector, which is the main source of emissions, accounted for 25–33% of these emissions during this period.¹ Cooperation with manufacturing companies is indispensable to cut CO₂ emissions, and environmental operations are inevitable because they are expected to address the issue of global warming as part of their corporate social responsibility (Managi and Kuriyama 2017, pp. 164–170).

Introducing cleaner production approaches can help decrease CO₂ emissions. Cleaner production aims to reduce the burden on the environment in all production processes, from the input of raw materials to the output and disposal of products (Managi and Kuriyama 2017, pp. 206–207). Since the enforcement of the Protocol, the approach has become more widespread in Japanese firms. As a result, manufacturing companies may have promoted a review of energy use, such as using clean energy, adopting energy-saving technology, and recycling energy.

Companies have an inherent incentive to improve energy performance, which is defined as energy use per unit of production, even in a situation that has not introduced cleaner production approaches. This is because it has the potential to reduce production costs through energy savings. However, in a situation where cleaner production is required to reduce CO₂ emissions, energy performance may not improve as expected immediately. This is due to the possibility that all production methods are reviewed considering long-term investments. Ideally, it is desirable for the companies that energy use remains the most efficient in either situation, considering it with and without pollution abatement.

The first objective of our study is to apply a new indicator of energy performance, considering the aforementioned situations in subnational (prefectural) scope. We refer to this new indicator as environmental energy efficiency (EEE), which has been employed at a multinational but regional level to date (Zhou and Ang 2008; Rakshit and Mandal 2020). Besides, our EEE takes a ratio to create an EEE index; this is different from previous studies. This indicator is an application of environmental efficiency proposed by Färe et al. (1996) and Kumar and Khanna (2009), which is measured by data envelopment analysis (DEA) using a distance function approach. For the period between 1990 and 2012, which was the framework for the Kyoto Protocol, we evaluate whether efficient use of energy and CO₂ emissions control has progressed by measuring the EEE of each prefecture in Japan. We focus on measuring the EEE of manufacturing industries because they are the main sources of CO₂ emissions. This study extracts seven industries from the manufacturing sector, in addition to the entire manufacturing industry, and provides policymakers with detailed information about industry-wise trends of energy performance in Japan.

Further, it is necessary to analyze the influencing factors of EEE to derive useful policy guidelines. The second objective of our study is to estimate EEE factors using a panel data analysis for the period of 1990–2012. We demonstrate that there is a significant relationship between EEE and three factors: economic development, industrial structure, and technological progress. This study investigates the influencing factors using a pooled mean group (PMG) estimator based on the autoregressive distributed lag (ARDL) approach of Pesaran et al. (1999) while simultaneously considering the problems of cross-sectional dependence, unit roots, and cointegration in the panels. We ultimately propose appropriate environmental energy strategies for policymakers.

The remainder of this paper is organized as follows. “Literature review” Section presents a brief literature review and the contributions of our study. “Measuring environmental energy efficiency” Section provides the measurement method of EEE, including the data used and the measurement results. “Panel data analysis” Section explains our panel data models, applied panel data, and reports the estimation results. “Conclusion” Section summarizes our empirical results and suggests directions for environmental energy policies in Japan.

2 Literature review

Energy input plays a pivotal role in the production of products. However, it generates undesirable byproducts, such as greenhouse gas emissions, which raises environmental concerns. The concept of energy efficiency is associated with controlling the adverse environmental effect and has become increasingly crucial. It refers to the minimized use of energy in the production process without compromising the same level of output.

The DEA method has been increasingly used in the past decades to investigate the nexus of environment and energy efficiency (Mardani et al. 2017). The DEA models primarily apply radial or non-radial measures, while our EEE is constructed using the radial model. Previous studies on environment and energy efficiency included a wide range of analysis targets from a multi-country perspective as examples of radial measures (Zhou and Ang 2008; Rakshit and Mandal 2020). Most empirical studies were focused on China (Wu et al. 2012) and India (Mandal 2010) because they are the world’s top energy consumers and CO₂ emitters among developing countries. However, other prior studies focus on Japan (Sueyoshi and Goto 2011; Fukuyama et al. 2020). We examine the empirical results of previous studies based on the radial model that serve as benchmarks for this study.

In previous research targeting Japan, the topic of energy efficiency has been researched sequentially since Hu and Wang (2006) proposed the index of total-factor energy efficiency (TFEE) using a DEA. Honma and Hu (2008) divide energy input into 11 sources between 1993 and 2003 in earlier applications of the region-wise TFEE assessment by Japanese prefectures. They found that, unlike most efficient inland and coastal (along the Sea of Japan) regions, Niigata and other prefectures (especially Chiba, Wakayama, and Yamaguchi) located in the Pacific Belt Zone, where energy-intensive industries have developed, were less energy efficient in the use of heavy oil and coal.

Moreover, Honma and Hu (2014b, 2018) supported these results. They estimated the effects of manufacturing share on the TFEE using panel data of the Japanese prefecture from 1996–2008.² The results showed that increased shares of the manufacturing industry (especially energy-intensive industry) are associated with a decline in energy efficiency.

Additionally, the sector-wise TFEE in the period of 1998–2005 was examined by the work of Honma and Hu (2013). In the Japanese manufacturing sector, they confirmed that as opposed to machinery industries (i.e., general, electric, and transportation machinery), energy-intensive industries (i.e., pulp and paper, chemical, cement and ceramics, and primary metal) have worse inefficiency on energy use in all studied years. However, according to Honma and Hu (2014a), whose study measured the country-wise TFEE for seven industries comparing Japan and another 13 developed countries in the period of 1995–2005, the results indicated that the industrial energy efficiency of Japan is at a higher level among developed countries through international comparison.

Honma and Hu have mainly focused on energy efficiency, not simultaneously accounting for environmental efficiency, which measures how much pollution emissions can be reduced in a production process. Although several prior studies investigated environmental efficiency in Japanese prefectures, such as Honma and Hu (2009) and Eguchi (2017), energy input was not incorporated into their DEA applications. Consequently, this study has three differences or contributions compared to previous studies. First, the current study proposes the EEE indicator, which simultaneously accounts for energy and environmental efficiency. Second, the study focuses on the manufacturing sector, including its seven subsectors, in Japanese prefectures. Third, it investigates the influencing factors of EEE in each industry using a PMG method. Thus, our study applies prefecture- and industry-level data in the Japanese manufacturing sector, where the level of industrial development differs among prefectures, and scrutinizes the movements of environmental energy performance in Japan.

3 Measuring environmental energy efficiency

3.1 Methodology and data

To present EEE, this study uses environmental or regulated technology, which is assumed for the joint production of desirable and undesirable outputs, and traditional or unregulated technology without simultaneously accounting for undesirable outputs (Färe et al. 2007). In the regulated technology, three inputs, including labor (l), capital stock (k), and energy (e), are used for producing the added value (y), which is a desirable output, while CO₂ (c) is an undesirable output. The unregulated technology does not consider CO₂ emissions in our approach. The regulated technology can then be represented as follows:

$$T = \left\{ \left( l,k,e,y,c \right): \left(l,k,e \right){\text{ can produce }} \left( y,c \right) \right\}$$

(1)

The technology that handles finite inputs and outputs is assumed to satisfy the standard properties: T is compact, while inputs and desirable outputs are strongly or freely disposable, as described by Färe and Grosskopf (2003). To show that it is a regulated technology, T is assumed to incorporate two additional conditions: null-jointness and weak disposability of outputs. Null-jointness is expressed as follows:

$${\text{if }}\left( {l,k,e,y,c} \right)\, \in T\,{\text{and}}\, c = 0{\text{ then}}\, y = 0$$

(2)

This means that CO₂ is emitted to generate an added value. The weak disposability of the outputs is as follows:

$${\text{if }}\left( {l,k,e,y,c} \right) \,\in T\,{\text{and}}\,0\leq\varphi\leq1\,{\text{then }}(l, k, e, {\varphi}y,{\varphi}c)\, \in T$$

(3)

This implies that if CO₂ emissions are reduced at the rate of $\varphi$, the added value simultaneously decreases at the same rate when the inputs are kept constant.³ This assumes that it is necessary to continue to divert some of the inputs in order to continuously reduce undesirable outputs, resulting in a proportionate decrease in desirable outputs under a given production technology. This assumption is introduced by Färe et al. (1989) and shows that it is usually necessary to incur regulatory costs to control emissions. However, the unregulated technology does not incorporate CO₂ emissions in T and is represented as follows:

$$\widehat{T} = \, \left\{ {\left( {l,k,e,y} \right): \, \left( {l,k,e} \right){\text{ can produce}}\,y} \right\}$$

(4)

$\widehat{\text{T}}$ assumes free disposability of undesirable outputs; that is, there is no regulatory cost to control CO₂ emissions.

Let i = (1, …, I) index be represented by the observations of inputs and outputs, (l_i, k_i, e_i, y_i, c_i) for i = 1, …, I; the regulated DEA technology of T consists of the following:

$$T =\{ \begin{array}{llllllll} ( {l,k,e,y,c} ): & \sum\limits_{i = 1}^{I} {z_{i} y_{i} \ge y,} & \sum\limits_{i = 1}^{I} {z_{i} c_{i} = c,} & \sum\limits_{i = 1}^{I} {z_{i} l_{i} \le l,} & \sum\limits_{i = 1}^{I} {z_{i} k_{i} \le k,} & \sum\limits_{i = 1}^{I} {z_{i} e_{i} \le e,} & \sum\limits_{i = 1}^{I} {z_{i} = 1,} & z_{i} \ge 0,\,i = 1,...,I \end{array}\}$$

(5)

In Eq. (5), the second constraint sets strict equality and imposes weak disposability on the undesirable output. Null-jointness is assumed to satisfy the condition ${\sum }_{{\text{i}}= \text{1} }^{\text{I}}{{\text{c}}}_{\text{i}}> \text{0}$. z_i as an intensity variable, while the sixth constraint is assumed to impose variable returns to scale (VRS) on the technology. The unregulated DEA technology of $\widehat{\text{T}}$ is established by excluding CO₂ emissions and dropping the second constraint in Eq. (5).

In this study, we propose two types of EEE by applying a distance function approach. To derive the first EEE, this study adopts the input distance function, accounting for both the reduction of energy use and CO₂ emissions, which applies Tyteca’s (1997) input-undesirable output model.⁴ The input distance function is defined as follows:

$$D_{c} \left( {l,k,e,y,c} \right) \, = {\text{ sup}}\{ \beta :\left( {l, \, k, \, e/\beta , y, c/\beta } \right)\, \in T\}$$

(6)

This distance function is a modification of the Shephard carbon distance function defined by Zhou et al. (2010).⁵ The value of β indicates how much energy use and CO₂ emissions can be simultaneously reduced. We assume that the input distance function is separable in desirable and undesirable outputs, following Kumar and Khanna (2009).⁶ Equation (6) can be rewritten as:

$$D_{c} \left( {l,k,e,y,c} \right) \, = C\left( c \right)\widehat{D}_{c} \left( {l,k,e,y} \right)$$

(7)

where

$$\widehat{D}_{c}\left({l,k,e,y} \right)\,={\text{sup}}\{\lambda:\left({l,\,k,\,e/\lambda, y}\right)\in\widehat{T}\}$$

(8)

Equation (8) is the energy-oriented input distance function without considering CO₂ emissions. The value of λ indicates the extent to which energy use can be reduced, not considering CO₂ emissions.

However, this study also presents the second EEE using the input distance function, which only accounts for the reduction in energy use. The input distance function is defined as follows:

$$D_{e} \left( {l,k,e,y,c} \right) \, = {\text{ sup}}\{ \theta : \, \left(l,k,{e/\theta,y,c} \right)\in{T}\}$$

(9)

This distance function is the Shephard energy distance function defined by Wu et al. (2012). The value of θ indicates the extent to which energy use can be reduced, considering CO₂ emissions. Based on Färe et al. (1996), we can assume that the input distance function is separable in desirable and undesirable outputs.⁷ Equation (9) can be rewritten as:

$$D_{e} \left( {l,k,e,y,c} \right) \, = C\left( c \right)\widehat{D}_{e} \left( {l,k,e,y} \right)$$

(10)

where $\widehat{\text{D}}$_e(l, k, e, y) =$\widehat{\text{D}}$_c(l, k, e, y).

According to Färe et al. (1996) and Kumar and Khanna (2009), Eqs. (7) and (10) are decomposed into the terms that reveal the impact of CO₂ emissions, C(c), and the term that captures the effect of energy efficiency, $\widehat{\text{D}}$_c(l, k, e, y) and $\widehat{\text{D}}$_e(l, k, e, y). Consequently, EEE1 and EEE2 were obtained as follows:

$$EEE1 = C\left(c\right) \, = \frac{{D_{c} \left( {l,\,k,\,e,y,c} \right)}}{{\widehat{D}\left( {l,\,k,\,e,y} \right)}}$$

(11)

and

$$EEE2 = C\left( c \right) \, = \frac{{D_{e} \left( {l,k,e,y,c} \right)}}{{\widehat{D}\left( {l,k,e,y} \right)}}$$

(12)

where $\widehat{\text{D}}$(l, k, e, y) = $\widehat{\text{D}}$_c(l, k, e, y) = $\widehat{\text{D}}$_e(l, k, e, y). If the EEE1 and EEE2 equal 1, it can be concluded that the observation is considered environmentally energy efficient. However, if the EEE1 and EEE2 is less than 1, the observation is regarded as environmentally energy inefficient in this study.

As three input distance functions need to be computed, this study solves the following linear programming problems:

$$\begin{gathered} \left[ {D_{c} \left( {l^{\prime}_{i} ,k^{\prime}_{i} ,e^{\prime}_{i} ,y^{\prime}_{i} ,c^{\prime}_{i} } \right)} \right]^{{ - 1}} = \,\min \,\beta \hfill \\ {\text{s.t.}}\sum\limits_{{i = 1}}^{I} {z_{i} } y_{i} \ge y^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } c_{i} = \,\,\beta c^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } l_{i} \le l^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } k_{i} \le k^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } e_{i} \le \beta e^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } = 1 \hfill \\ \,\,\,\,\,\,z_{i} \ge 0,\,i = 1,...,I\end{gathered}$$

(13)

$$\begin{gathered} \left[ {D_{e} \left( {l^{\prime}_{i} ,k^{\prime}_{i} ,e^{\prime}_{i} ,y^{\prime}_{i} ,c^{\prime}_{i} } \right)} \right]^{{ - 1}} = \,\min \,\theta \hfill \\ {\text{s.t.}}\sum\limits_{{i = 1}}^{I} {z_{i} } y_{i} \ge y^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } c_{i} = \,\,c^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } l_{i} \le l^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } k_{i} \le k^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } e_{i} \le \theta e^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } = 1 \hfill \\ \,\,\,\,\,\,z_{i} \ge 0,\,i = 1,...,I \hfill \\ \end{gathered}$$

(14)

$$\begin{gathered} \left[ {\widehat{D} \left( {l^{\prime}_{i} ,k^{\prime}_{i} ,e^{\prime}_{i} ,y^{\prime}_{i} } \right)} \right]^{{ - 1}} = \,\min \,\lambda \hfill \\ {\text{s.t.}}\sum\limits_{{i = 1}}^{I} {z_{i} } y_{i} \ge y^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } l_{i} \le l^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } k_{i} \le k^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } e_{i} \le \lambda e^{\prime}_{i} \hfill \\ \,\,\,\,\,\,\sum\limits_{{i = 1}}^{I} {z_{i} } = 1 \hfill \\ \,\,\,\,\,\,z_{i} \ge 0,\,i = 1,...,I \hfill \\ \end{gathered}$$

(15)

To measure the EEE, this study uses one desirable and undesirable output each, and three inputs in the Japanese manufacturing industry covering 47 prefectures from 1990 to 2012. This study extracts seven industries from the manufacturing sector: (1) food, beverages, and tobacco; (2) textile; (3) pulp and paper; (4) chemical, oil, and coal; (5) ceramic, stone, and clay; (6) iron and steel; and (7) machinery, while simultaneously analyzing the manufacturing industry as a whole. Thus, because our study applies prefecture- and industry-level data in the Japanese manufacturing sector with a focus on energy use, we selected variables for which data can be obtained consistently. The variables used in this study are as follows:

• Desirable output: real added value (y).

• Undesirable output: CO₂ emissions (c).

• Inputs: employed people (l), real capital stock (k), and energy use (e).⁸

Real added value, employed people, and real capital stock are sourced from the Regional-Level Japan Industrial Productivity (R-JIP) Database 2017.⁹ The data on CO₂ emissions and energy use were sourced from the Energy Consumption Statistics by Prefecture (ECSP).¹⁰ This study uses CO₂ emitted from the combustion of fossil fuels (coal, coal products, oil, oil products, natural gas, and town gas), excluding non-energy utilization, while energy use is based on total energy consumption.¹¹Finally, for the above seven manufacturing industries, we selected them by extracting only those industries for which data can be obtained consistently from the R-JIP 2017 and the ECSP.¹² The summary statistics for variables are shown in Table 1.

Table 1 Summary statistics of variables

3.2 Measuring results

Figure 1 shows the changes in EEE1 and EEE2 for the manufacturing sector in 1990–2012, where the scores of EEE1 are significantly higher than those of EEE2 (EEE1 > EEE2). Thus, incorporating CO₂ emission reduction with energy use reduction such as of EEE1 yields a higher index than its EEE2 counterpart which only considers the reduction of energy use. Moreover, the results of EEE2 clearly show that the EEE had shifted from an upward to a downward trend since the end of the 1990s, especially following the Asian Crisis (1997–1998) and the Financial Crisis (2007–2008). In particular, the EEE plunged after 2008, owing to declining added value and subsequent energy efficiency. This phenomenon was observed not only in Japan but also in other high-income countries (Rakshit and Mandal 2020).

Fig. 1

Changes in EEE1 and EEE2 for the manufacturing sector (1990–2012). The values in the figure are the arithmetic mean based on the measurement results by prefecture

Figure 2 shows the changes in EEE1 and EEE2 by light industry. Considering the EEE1 measurement for the light industry between 1990 and 2012, higher scores are observed in “food, beverages, and tobacco” followed by “textile” and “pulp and paper” consecutively, whose annual trends are almost strictly untangled. In the industry of “pulp and paper,” a noteworthy improvement was discovered between 1994 and 1995, while a notable decline occurred between 2010 and 2012. Meanwhile, in the EEE2 measurement, annual trends between 1990 and 2012 for “textile” and “pulp and paper” showed a tendency of improvement albeit inconsistent, where its conspicuous advancement is observed between 1993 and 1997, and 1994 and 1997, respectively. However, a decline occurred in “pulp and paper” between 2010 and 2012. In the industry of “food, beverages, and tobacco,” despite relative stability in the studied period, its score was gradually reduced between 1990 and 2012. Therefore, these results indicate that the light industry had failed to improve the EEE continuously since the end of the 1990s.

Fig. 2

Changes in EEE1 and EEE2 by light industry (1990–2012). The values in the figure are the arithmetic mean of each industry based on the measurement results by prefecture

Figure 3 shows the changes in EEE1 and EEE2 by heavy industry, which have a fluctuating trend than light industries. Although the scores of “iron and steel” were higher among others between 1990 and 2005, they had reduced significantly since the late 2000s. This downtrend had also occurred with “ceramic, stone, and clay” at the end of the 2000s, which was the rock-bottom for the whole heavy industry. Meanwhile, “machinery” had led as the best frontier. Likewise, an increasing trend was observed with “chemical, oil, and coal” since the late 2000s. Therefore, these results indicate that “ceramic, stone, and clay” and “iron and steel” had tended to deteriorate the EEE since the late 2000s albeit rather inconsistent for “ceramic, stone, and clay,” as opposed to “chemical, oil, and coal” and “machinery.” These results are consistent with the findings of a previous study that higher energy efficiency in machinery industries was also found by Honma and Hu (2013).

Fig. 3

Changes in EEE1 and EEE2 by heavy industry (1990–2012). The values in the figure are the arithmetic mean of each industry based on the measurement results by prefecture

However, Table 2 shows the annual mean of EEE1 and EEE2 for the manufacturing sector by prefecture. The scores vary by prefecture, with ranges being 0.150–1.000 and 0.142–1.000, respectively. Regardless of the type of EEE, Tokyo, Aichi, and Okinawa lead as the frontier prefectures, whereas Hokkaido, Hiroshima, and Fukuoka are among the least efficient prefectures. In comparison to previous studies focusing on the results of EEE2, which is the indicator considered for energy reduction, Hokkaido is the prefecture with the lowest score, as well as Niigata, Ehime, and other prefectures located in the Pacific Belt Zone (i.e., Ibaraki, Okayama, Hiroshima, and Fukuoka) tend to be more inefficient. These results are consistent with the findings of Honma and Hu (2008, 2009) on the lowest environmental efficiency in Hokkaido and lower energy efficiency in some regions within Pacific Belt Zone.

Table 2 Annual mean of EEE1 and EEE2 for the manufacturing sector by prefecture (1990–2012)

Figures 4 and 5 show the annual mean of EEE1 and EEE2 in seven manufacturing industries by region. Compared to Figs. 4 and 5, the scores of EEE2 vary widely across regions than those of EEE1. To grasp trends between regions, focusing on the results of EEE1, the results present a significant variation in scores among industries in the Hokkaido and Tohoku, Chugoku and Shikoku, and Kyushu and Okinawa regions. Meanwhile, more efficient industries are comparatively concentrated in the Kanto, Chubu, and Kinki regions, which include a major metropolitan area. Therefore, intensive improvement in local regions might be essential.

Fig. 4

Annual mean of the EEE1 in seven manufacturing industries by region (1990–2012). The regional classification is partially modified and integrated based on the Ministry of Economy, Trade and Industry (METI) of Japan

Fig. 5

Annual mean of the EEE2 in seven manufacturing industries by region (1990–2012). The regional classification is partially modified and integrated based on the Ministry of Economy, Trade and Industry (METI) of Japan

4 Panel data analysis

4.1 Models and data

To investigate the influencing factors of EEE, this study estimates the following panel data models for EEE1 and EEE2:

[Model 1]

$${{EEE1}}_{i,t} = \beta_{1} \,GDPC_{i,t} + \beta_{2} \left( {GDPC}_{i,t} \right)^{2} + \beta_{3} IS_{i,t} + \beta_{4} EI_{i,t} + \beta_{5} \mu_{t} + \eta_{i} + \varepsilon_{i,t}$$

(16)

and

[Model 2]

$$EEE2_{i,t} = \beta_{1} \,GDPC_{i,t} + \beta_{2} IS_{i,t} + \beta_{3} EI_{i,t} + \beta_{4} \mu_{t} + \eta_{i} + \varepsilon_{i,t}$$

(17)

where the subscripts i and t represent the prefecture and year, respectively. The dependent variables EEE1 and EEE2 are measured in the previous section. For the explanatory variables, GDPC stands for gross domestic product (GDP) per capita, IS is the industry share, and EI is the energy intensity. η_i is the unobservable fixed effect by prefecture, μ_t is the linear trend showing a common time effect, and ε is an error term.

Following Kumar and Khanna (2009), who conducted a panel data analysis, our model includes the EEE factors associated with development level, industry structure, and technology progress. In this research, our panel data are constructed for each industry and analyzed empirically for the period 1990–2012.

GDPC is the proxy for the level of economic development and is shown by the real GDP divided by population. Equation (16) shows a U-shaped relationship between EEE1 and GDP per capita, which is intended to investigate the environmental Kuznets curve (EKC) hypothesis. This is because EEE1 considers not only saving energy but also reducing CO₂. Therefore, the estimated parameters of the GDPC are expected to satisfy β₁ < 0 and β₂ > 0. The turning point of GDP per capita was calculated as (− β₁/2β₂). However, Eq. (17) examines if there is a linear relationship between EEE2 and GDP per capita, and the estimated parameter is expected to be positive. This is because there is always an incentive to improve energy efficiency, and EEE2 only focuses on saving energy without considering CO₂ reduction. Therefore, the model is not incorporated into testing the EKC hypothesis. The real GDP at 2000 constant prices (million yen) and the total population are sourced from the R-JIP 2017.

IS is the proxy for the change in industrial structure and is represented by the ratio of each industry to manufacturing total, which is calculated using the nominal added value.¹³ As industry share reflects the degree of industrial concentration, the EEE is expected to improve as its power increases. Thus, the estimated parameter of IS is expected to be positive. The nominal added value was sourced from the R-JIP 2017.

EI is a proxy for the progress of energy-saving technology and is used as energy intensity, which is measured by the total energy consumption divided by the real added value, for each industry. Given that the EEE is expected to improve as technology advances, the estimated parameter of EI is expected to be negative. The real added value at 2000 constant prices (million yen) and total energy consumption (terajoule) were sourced from the R-JIP 2017 and the ECSP, respectively. The descriptive statistics of variables are summarized in Table 3.

Table 3 Descriptive statistics of variables

To examine the influencing factors, this study adopts the error correction model of the ARDL approach presented by Pesaran et al. (1999). There are three advantages to employing the model, considering the use of long-term panel data in our study. First, if there are data with unit roots, the long-term panel data are likely to be exposed to risk of non-stationary effects. However, the model can simultaneously incorporate stationary and non-stationary variables. Second, if the slope parameters are not homogeneous across cross-sectional units, slope heterogeneity could bias the estimation results (Pesaran and Smith 1995). However, this model allows parameter heterogeneity. Third, the model can be divided into a long-run effect and a short-run effect on the dependent variable. It is important to clarify the long-run relationship to provide policymakers with practical information.

In this study, our panel data models are estimated using the PMG estimator developed by Pesaran et al. (1999), which has been applied to empirical studies related to CO₂ emissions, such as Martínez-Zarzoso and Bengochea-Morancho (2004) and Iwata et al. (2011). The estimator assumes that the long-run coefficients are homogeneous across cross-sectional units, while the short-run coefficients are heterogeneous, as are the intercepts and the error correction coefficients. The PMG method is established by the ARDL approach, and Eqs. (16) and (17) are built into an ARDL (p, q) model as follows:

$$EEE_{i,t} = \sum\nolimits_{j = 1}^{p} {\lambda_{i,j} } EEE_{i,t - j} + \sum\nolimits_{j = 0}^{q} \delta^{\prime }_{i,j} X_{i,t-j} + \gamma_{i} \mu_{t} + \eta_{i} + \varepsilon_{i,t}$$

(18)

where X_i,t is the K × 1 vector of explanatory variables, δ_i,j is the K × 1 coefficient vector, λ_i,j is the coefficient of the lagged dependent variables, and γ_i is the coefficient of the linear trend. Equation (18) can be rewritten as the following error correction model:

$$\Delta EEE_{i,t} = \phi_{i} \left( {EEE_{i,t - 1} - \theta^{\prime}_{i} X_{i,t} } \right) + \sum\nolimits_{j = 1}^{p - 1} {\lambda^*_{ij} } \Delta EEE_{i,t-j} + \sum\nolimits_{j = 0}^{q - 1} {\delta_{i,j}^{{*^{\prime}}} \Delta X_{{_{i,t-j} }} } + \gamma_{i} \mu_{t} + \eta_{i} + \varepsilon_{i,t}$$

(19)

where

$$\phi_{i} = - \left( {1 - \sum\limits_{j = 1}^{p} {\lambda_{i,j} } } \right),$$

$$\theta_{i} = - \frac{{\sum\nolimits_{j = 0}^{q} {\delta_{i,j} } }}{{\phi_{i} }},$$

$$\lambda_{i,j}^{ * } = - \sum\limits_{m = j + 1}^{p} {\lambda_{i,m} } ,$$

$$\delta_{i,j}^{ * } = - \sum\limits_{m = j + 1}^{q} {\delta_{i,m} }$$

In Eq. (19), the parenthesis is the error correction term, and $\phi_{i}$ is its coefficient, which is expected to be significantly negative to exist a long-run relationship. In the PMG estimation, the short-run coefficient vector is $\delta_{i,\,j}^{ * }$. The long-run coefficient vector is θ_i, which is assumed to be homogeneous (that is, θ_i = θ for all i) and is estimated using the maximum likelihood method (Pesaran et al. 1999).

This study uses the following four procedures to estimate the EEE factors.¹⁴ First, to examine the independence among cross-sectional units in the panels, the cross-section dependence (CD) tests are implemented by a variety of tests: the Lagrange multiplier (LM) test of Breusch and Pagan (1980), the scaled LM and CD tests of Pesaran (2004), and the bias-corrected scaled LM test of Baltagi et al. (2012). Second, if there is cross-sectional dependence, to investigate the stationarity of the panels, the panel unit root test is selected for the cross-sectionally augmented Im, Pesaran, and Shin (CIPS) test as done by Pesaran (2007). The CIPS panel unit root test can control the problem of cross-sectional dependence and then identify whether each variable is integrated with order zero or one, that is, I(0) or I(1). Third, if there are unit roots in the panels, the panel cointegration tests of Pedroni (1999, 2004) are conducted to confirm a cointegration relationship among the variables. Finally, the PMG estimation is carried out based on the ARDL (1, 1) model by setting p = q = 1 in Eq. (19), which has been broadly used in empirical studies such as Li et al. (2016) and Salahuddin et al. (2016).¹⁵ However, the cross-sectional dependence, which is often caused by unobservable common factors, is modeled by including a linear trend.

4.2 Estimation results

Table 4 presents the results of the CD tests. The null hypothesis of no cross-section correlation was mostly rejected at 1% for all the variables. This result indicates that there is a cross-sectional dependence in each variable, and consequently, it is appropriate to choose the CIPS panel unit root test. Table 5 presents the results of the CIPS test, which is the null hypothesis of the unit root. The results show a mixed order of integration in our variables, and consequently, the PMG estimation is adequate in our panel data. Moreover, we examine whether there is a cointegrating relationship among the variables using the Pedroni panel cointegration tests. Table 6 presents the results of the Pedroni test. In all the models, the null hypothesis of no cointegration is rejected at 5% for four of the seven statistics, excluding Model 2 for machinery. Hence, it is difficult to analyze whether the variables can be assumed to have a cointegrating relationship.

Table 4 Cross-section dependence (CD) tests

Table 5 CIPS panel unit root tests

Table 6 Pedroni panel cointegration tests

However, Tables 7 and 8 report the estimation results of the PMG for EEE1 and EEE2, respectively. In both tables, the error correction coefficients are significantly estimated at the 1% level and are confirmed to be negative in all the models. This means that the models have a long-run equilibrium relationship among the variables.

Table 7 Pooled mean group (PMG) estimation results in the EEE1

Table 8 Pooled mean group (PMG) estimation results in the EEE2

In the long-run coefficients of Table 7, the parameters of GDPC and its square are significant at the 5% level, excluding “chemical, oil, and coal” and “iron and steel.” The estimated parameters reflect an inverted U-shaped relationship between EEE1 and GDP per capita, while the turning point of GDP per capita falls within the collected data. This result indicates that there is no EKC pattern, contrary to our expectations. However, in the long-run coefficients of Table 8, the parameters of GDPC are significantly estimated to be positive at 5% level, excluding “food, beverages, and tobacco” and “textile.” As expected, the estimated parameters show a linear relationship between EEE2 and GDP per capita. Therefore, the EEE in the Japanese manufacturing industry tends to improve and then deteriorate or monotonically increase as economic development progresses.

Likewise, the estimation results of the other variables are also meaningful. The parameters of IS are significantly estimated to be positive at 5% level in Tables 7 and 8, or both, excluding “pulp and paper,” as expected. This result means that the EEE in the Japanese manufacturing industry is raised by expanding the industry share. This finding suggests that the EEE is likely to improve as industry concentration increases. For the industry of “pulp and paper,” which is significantly estimated to be negative, the results should be carefully interpreted and seem to be influenced by the consolidation of companies. However, it should be reconsidered in subsequent analysis.

Meanwhile, although the parameters of EI are significantly estimated in Tables 7 and 8, or both, the sign of the coefficients is mixed. However, in the results of EEE2, which is the indicator focused on energy reduction, the parameters are estimated to be significantly negative at the 5% level (Table 8), excluding “chemical, oil, and coal” and “iron and steel,” which is consistent with our expectations. This result suggests that the progress of energy-saving technology could lead to an increase in EEE through energy reduction in the Japanese manufacturing industry. In this regard, it is too early to draw any conclusions from the result. This is because most industries in Table 7 are significantly estimated to be positive, contrary to our expectations. The findings suggest that manufacturing companies have pushed for a review of energy use, such as alternatives to clean energy, despite little progress in reducing energy consumption. Therefore, it can be analyzed in future studies whether EI is appropriate as a proxy for the progress of energy-saving technology, as it has been typically used in empirical studies (Kumar and Khanna 2009).

5 Conclusion

This section summarizes the empirical results and derives the policy implications. For changes in the energy performance of Japan, some practical information was confirmed from the measurement results of the EEE in the manufacturing industry. The EEE of the entire manufacturing industry was at almost the same level or on a downward trend from late 1990 to 2012. However, the changes in the EEE have different movements by individual industries. Notably, the EEE had improved in “chemical, oil, and coal” and “machinery,” although it had deteriorated in “iron and steel” since the late 2000s. This finding indicates that the Japanese manufacturing industry had not been keeping in step with the improvement of EEE since the adoption of the Kyoto Protocol.

Meanwhile, the EEE of the entire manufacturing industry has different trends by each prefecture during the studied period. Tokyo, Aichi, and Okinawa were the most efficient prefectures, whereas Hokkaido, Hiroshima, and Fukuoka were inefficient. However, more efficient industries tended to have comparatively concentrated in the Kanto, Chubu, and Kinki regions. Consequently, this finding suggests that the progress of each industry was relatively consistent by region.

Furthermore, some interesting results were obtained from the panel data analysis. First, the EEE was likely to improve and then deteriorate or monotonically increase with economic development between 1990 and 2012. This implies that the improvement phase can occur because more advanced systems, skilled labor, and sophisticated facilities are prepared as development increases. Therefore, policymakers should consider creating an institutional environment that encourages investment in human and physical capital in the environmental energy fields. Moreover, the EEE was raised by expanding the industry share during the study period. This indicates that strengthening industrial concentration on specific industries within the manufacturing industry is an improvement factor. It is necessary to promote the selection and concentration of industries as a part of the environmental energy policy. Finally, the EEE, focusing on energy reduction, was likely to increase with the progress of energy-saving technology during that period. This suggests that efforts to conserve energy could be another improvement factor. It might be important to support the innovation of energy-saving technologies as part of the environmental energy strategy.

However, it should be noted that some analytical challenges remain in our research.　Although the study period is limited to 1990–2012 due to data constraints, it would be valuable to extend the period and include the impact of the Paris Agreement in future research.¹⁶ In this study, it would be more beneficial to subdivide the manufacturing industry more than it is classified in the current survey. In addition, it would be desirable to verify the industry-wise trend and its influencing factors of the EEE using micro-level firm data. If the study can resolve these challenges, it will be possible to provide policymakers with more detailed information and solid evidence to help implement an effective environmentally friendly energy policy in Japan.