12
Measurement of environmentally sensitive productivity growth in Korean industries Yeimin Chung a , Almas Heshmati b, * a Department of Food and Resource Economics, College of Life Sciences and Biotechnology, Korea University, Anam-dong, Seongbuk-gu, East Building, Seoul 136-713, Republic of Korea b Department of Economics, Sogang University, Sinsoo-dong #1, Mapo-gu, Seoul 121-742, Republic of Korea article info Article history: Received 23 September 2013 Received in revised form 12 May 2014 Accepted 10 June 2014 Available online xxx Keywords: CO 2 emission Undesirable output DEA MalmquisteLuenberger productivity (ML) index Metafrontier MalmquisteLuenberger productivity (MML) index Productivity change abstract In this study we will attempt to measure productivity growth at the industrial level using the Meta- frontier MalmquisteLuenberger (MML) productivity growth index and dissect/analyze this index to reveal further information. The results will be compared with those obtained from the conventional MalmquisteLuenberger (ML) productivity growth index. Utilizing the MML-index has two advantages when compared with the ML-index: the rst is that it is able to consider undesirable output as a by- product of production; and the second is that it can account for producer group heterogeneities such as production technology. Noting such advantages, we will model this study to achieve three objectives related to productivity, technology and policy effects. To separate the results of the productivity index, we estimate the changes in the technological gap between regional and global frontier technologies. The proposed index presents productivity growth and dissects its components into 14 Korean industrial sectors from 1981 to 2010. For the purpose of detailed analysis, we have divided the relevant period into three decades. The results show that technology innovation can be regarded as an important component of productivity growth, rather than merely efciency change. Chemical and petrochemical, iron and steel and machinery are all treated as global innovators throughout the entire period. It is also inferred that the groups with higher labor productivity obtain a higher productivity growth rate as compared with their low labor productivity counterparts. Considering the heterogeneity of production technology and the time that policy is introduced, the policy implications of the results will affect the circumstances regarding investment in environmental technology. © 2014 Elsevier Ltd. All rights reserved. 1. Introduction This study tests the effect of CO 2 emissions on production by examining the degree of and quantitative changes to productivity of Korean industries when they are faced with the regulatory pressure of reducing their CO 2 emission levels. Also, we attempt to nd evi- dence of effects of environmental policy upon productivity growth at the industrial level. Recently, Korea has employed several new policies concerning the environment, especially regarding air pollution. Most of these policies were implemented following adoption of the Kyoto Protocol in 1997. Following the Kyoto Protocol, thirty-eight developed countries that assumed duty of transition were obliged to cut down their emission of greenhouse gases. 1 Since Korea was classied as a developing country at that time, the country was given an extra adjustment period to prepare as a nation faced with the duty of transition. As a nation, Korea needed to ac- quire a cleaner production technology and economic structure to prepare for the international agreement of reducing its CO 2 transition. This study analyzes how well Korea was prepared to follow the requisite environmental regulations from 1981 to 2010. The goal of implementing environmental policy is the management of envi- ronmentally harmful by products, not the manufacturing of the primary industrial products of a nation. Therefore, it is necessary to consider outputs that can be divided into desirable and undesirable categories for establishing a suitable environmental policy. Despite the fact that rms do not intend to produce harmful and undesir- able by products, it is often the case that a government controls the amount of undesirable output to achieve the goal of environmental policy. Under such conditions, industries are often obliged to adjust to the new economic circumstances through reducing the * Corresponding author. Tel.: þ82 10 4513 1712. E-mail addresses: [email protected] (Y. Chung), [email protected] (A. Heshmati). 1 Greenhouse gases consist of six components: CO 2 , CH 4 ,N 2 O, PHC, HFC and SF 6. Contents lists available at ScienceDirect Journal of Cleaner Production journal homepage: www.elsevier.com/locate/jclepro http://dx.doi.org/10.1016/j.jclepro.2014.06.030 0959-6526/© 2014 Elsevier Ltd. All rights reserved. Journal of Cleaner Production xxx (2014) 1e12 Please cite this article in press as: Chung, Y., Heshmati, A., Measurement of environmentally sensitive productivity growth in Korean industries, Journal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2014.06.030

Measurement of environmentally sensitive productivity growth in Korean industries

  • Upload
    almas

  • View
    215

  • Download
    2

Embed Size (px)

Citation preview

Page 1: Measurement of environmentally sensitive productivity growth in Korean industries

lable at ScienceDirect

Journal of Cleaner Production xxx (2014) 1e12

Contents lists avai

Journal of Cleaner Production

journal homepage: www.elsevier .com/locate/ jc lepro

Measurement of environmentally sensitive productivity growth inKorean industries

Yeimin Chung a, Almas Heshmati b, *

a Department of Food and Resource Economics, College of Life Sciences and Biotechnology, Korea University, Anam-dong, Seongbuk-gu,East Building, Seoul 136-713, Republic of Koreab Department of Economics, Sogang University, Sinsoo-dong #1, Mapo-gu, Seoul 121-742, Republic of Korea

a r t i c l e i n f o

Article history:Received 23 September 2013Received in revised form12 May 2014Accepted 10 June 2014Available online xxx

Keywords:CO2 emissionUndesirable outputDEAMalmquisteLuenberger productivity (ML)indexMetafrontier MalmquisteLuenbergerproductivity (MML) indexProductivity change

* Corresponding author. Tel.: þ82 10 4513 1712.E-mail addresses: [email protected] (Y. Chu

(A. Heshmati).1 Greenhouse gases consist of six components: CO2

http://dx.doi.org/10.1016/j.jclepro.2014.06.0300959-6526/© 2014 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Chung, Y.,Journal of Cleaner Production (2014), http:/

a b s t r a c t

In this study we will attempt to measure productivity growth at the industrial level using the Meta-frontier MalmquisteLuenberger (MML) productivity growth index and dissect/analyze this index toreveal further information. The results will be compared with those obtained from the conventionalMalmquisteLuenberger (ML) productivity growth index. Utilizing the MML-index has two advantageswhen compared with the ML-index: the first is that it is able to consider undesirable output as a by-product of production; and the second is that it can account for producer group heterogeneities suchas production technology. Noting such advantages, we will model this study to achieve three objectivesrelated to productivity, technology and policy effects. To separate the results of the productivity index,we estimate the changes in the technological gap between regional and global frontier technologies. Theproposed index presents productivity growth and dissects its components into 14 Korean industrialsectors from 1981 to 2010. For the purpose of detailed analysis, we have divided the relevant period intothree decades. The results show that technology innovation can be regarded as an important componentof productivity growth, rather than merely efficiency change. Chemical and petrochemical, iron and steeland machinery are all treated as global innovators throughout the entire period. It is also inferred thatthe groups with higher labor productivity obtain a higher productivity growth rate as compared withtheir low labor productivity counterparts. Considering the heterogeneity of production technology andthe time that policy is introduced, the policy implications of the results will affect the circumstancesregarding investment in environmental technology.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

This study tests the effect of CO2 emissions on production byexamining the degree of and quantitative changes to productivity ofKorean industries when they are faced with the regulatory pressureof reducing their CO2 emission levels. Also, we attempt to find evi-dence of effects of environmental policy upon productivity growthat the industrial level. Recently, Korea has employed several newpolicies concerning the environment, especially regarding airpollution. Most of these policies were implemented followingadoption of the Kyoto Protocol in 1997. Following the Kyoto Protocol,thirty-eight developed countries that assumed duty of transitionwere obliged to cut down their emission of greenhouse gases.1 Since

ng), [email protected]

, CH4, N2O, PHC, HFC and SF6.

Heshmati, A., Measurement o/dx.doi.org/10.1016/j.jclepro.2

Korea was classified as a developing country at that time, thecountrywas given an extra adjustment period to prepare as a nationfaced with the duty of transition. As a nation, Korea needed to ac-quire a cleaner production technology and economic structure toprepare for the international agreement of reducing its CO2transition.

This study analyzes how well Korea was prepared to follow therequisite environmental regulations from 1981 to 2010. The goal ofimplementing environmental policy is the management of envi-ronmentally harmful by products, not the manufacturing of theprimary industrial products of a nation. Therefore, it is necessary toconsider outputs that can be divided into desirable and undesirablecategories for establishing a suitable environmental policy. Despitethe fact that firms do not intend to produce harmful and undesir-able by products, it is often the case that a government controls theamount of undesirable output to achieve the goal of environmentalpolicy. Under such conditions, industries are often obliged to adjustto the new economic circumstances through reducing the

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 2: Measurement of environmentally sensitive productivity growth in Korean industries

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e122

undesirable output or changing their production technologye evenif these actions reduce their production and profit.

Since no firm wants to decrease their desirable output on avoluntary basis, most firms and industries are forced to changetheir production processes by taking advantage of technologicalprogress. For this reason, industries are faced with productivitychanges based on internal and external technological levels. Thispoint contributes to the existing literature by employing a meth-odology superior to the commonly used environmental method byaccounting for heterogeneities in behavior by the industrial sector.This new approach is applied to data from a newly industrialeconomy such as Korea, which is intending to adopt the obligatoryenvironmental policies.

It should be noted in this context that of greater importance isthe understanding of the relationship between productivity andCO2 emission. Based on this relationship, environmental policiesshould be considered and carried into practice. However, questionsare raised about whether the relevance of introduced policies andmeasures being sufficient to counterbalance the economic lossgenerated from CO2 emissions mitigation.

By considering this problem in all its dimensions, in this studywe attempt to find evidence that CO2 emissions become aconstraint2 on the expansion of production activities and theirgrowth. According to Kim et al. (2010), most studies related to theenvironment insist upon the positive relationship between CO2

emission, economic activity and growth. In understanding such arelationship, policy makers should support firms or industries bysuggesting a set of effective policies that enhance productivitylevels of individual firms and industries. The Metafrontier Malm-quisteLuenberger (MML) productivity index is used to shed lighton these issues which embody the contribution of this paper to theexisting literature.

The rest of this study is organized as follows: we review theliterature in Section 2 to provide a detailed explanation of themodel with the background of the MML-index. The MML meth-odology and empirical models are outlined in Section 3. Section 4presents the empirical results based on Korean industrial leveldata. The final Section 5 provides a summary and policyrecommendations.

2. Literature review

Before utilizing the MML index with its fundamental assump-tion, we need to confirm the impact or causality effects betweeneconomic activity and CO2 emission. Then we can utilize the MMLindex to find the evidence of environment regulation impact uponindustry. Thus, wewill first describe the literature that studies suchrelationships and then review the methodology used in this study.

To understand the structure of relationship, several studies triedto find evidence of causality between economic activity and CO2emission using various methodologies. The conventional andwidely used methodology is time series and panel data analysis.Most previous studies which related to productivity or growth rateand environmental issue assume simple linearity with lag structurein the relationship.

However, there was evidence of divergence in the conclusion onthe matter of direction of causality under a simple linearityassumption. Therefore, Hsieh (1991) and Brock and LeBaron (1996)insisted that the assumption of non-linear relationship is appro-priate to explain two indicators due to the effect of cyclical

2 Yang et al. (2011) found a positive relationship between pollution abatementfees and R&D expenditures, but there is no evidence to support an R&D-induce-ment effect brought about by PACE.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

components of the economies. Accepting the nonlinearityassumption has led to introduction of many alternative method-ologies that were further developed to analyze the indicators'causal relationship– for example, the smooth transition autore-gressive model (STAR) that was developed by Luukkonen et al.(1988) and the Threshold Vector Error Correction Model devel-oped by Hansen and Seo (2002).

Based on a parametric analysis regarding the causality betweeneconomic activity and CO2 emission, it is justified that we couldassume (i) non-linearity and (ii) mutual causality between the twofactors when using the non-parametric method. As a nonpara-metric efficiency evaluation technique with those assumptions,DEA recently gained popularity and it adjusted productivity forenvironmental issues.

Chung et al. (1997) employed undesirable output as a by-product when a unit produced desirable output and analyzedproductivity growth using the ML index at the micro-level. Theyannounced that technical change is a primary source of produc-tivity growth in Swedish paper and pulp mills for the period of1986e1990. F€are et al. (2001) also adopted the ML-index method-ology and used a state-level panel data to analyze the relationshipbetween market output and pollution abatement cost in1974e1986.

Following the study of F€are et al. (2001), the ML-index wasused extensively as a standard best practice methodology forestimating productivity growth in wide areas of performancerelated research. In addition, Weber and Domazlicky (2001)applied a similar methodology in order to include toxic releasein the productivity analysis of the US manufacturing sector in1988e1994. Sueyoshi and Goto (2010) proposed a new use of DEAto measure the operational, environmental and unified efficiencymeasures of US coal-fired power plants. Kumar (2006) alsoemployed the ML-index to analyze the environmentally sensitiveproductivity growth of 41 countries for the period between 1973and 1992. In his study, Kumar found that the productivity growthof Annex-I countries were higher than that of non-Annex-Icountries3 and that technical change was the main contributorto productivity growth.

In a recent study, Oh and Heshmati (2010) proposed an index formeasuring environmentally sensitive productivity growth whichappropriately considers the characters of technical change in pro-duction. For incorporating this aspect in developing the index, theconventionally used MalmquisteLuenberger productivity indexwas modified to give the Sequential MalmquisteLuenberger (SML)productivity index. This index was employed to measure environ-mentally sensitive productivity growth and the analyzed compo-nents of 26 OECD countries during the period of 1970e2003.

It should be mentioned that several of the abovementionedstudies in which the ML-index was used have a weakness in termsof individual heterogeneity, where individuals are either highlyaggregated industrial sectors or countries. To account for the het-erogeneity of groups, Oh (2010a) suggested the Metafrontier ML-index (MML-index) as a preferable method to estimate productiv-ity growth. Oh presented an alternative environmentally sensitiveproductivity growth index to incorporate group heterogeneitiesinto the conventional MalmquisteLuenberger productivity growthindex. The proposed index is employed to measure productivitygrowth and its decomposed components in 46 countries observedbetween 1993 and 2003. The main finding indicates that Europehas taken the lead in the world frontier technology and Asia hasattempted to catch up by moving towards the frontier technology.

3 Annex-I countries tend to represent developed countries and non-Annex-I re-fers to developing countries.

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 3: Measurement of environmentally sensitive productivity growth in Korean industries

5 Such as production probability set (PPS), convexity, disposability, allocationefficiency, technical efficiency, and so on.

6 Douglas W. Caves, Laurits R. Christensen and Walter E. Diewert introduced theMalmquist index in the 1982 with the title “Multilateral Comparisons of Output,Input and Productivity Using Superlative Index Numbers”.

7 Linear programming was used to decide a way to get the best outcome, such asmaximum profit or lowest cost, in a given mathematical model for some list of

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e12 3

It is worth mentioning that the MML-index is a combination oftwo concepts. One is the conventional ML-index and another is theconcept of Metafrontier. The Metafrontier is the envelope of theconceived production frontiers introduced by Battese and Rao(2002) and further described in Battese et al. (2004). Battese andRao (2002) tried to solve the incomparability of performances ofvarious groups by using the concept of Metafrontier or globalfrontier.

Although a large number of studies have been performedusing the ML-index, there are few studies that have utilized theconcept of the MML-index. Oh and Heshmati (2010), Oh (2010b)and Chiu et al. (2012) are representative studies on productivitygrowth related to environmental issues employing SML andMML-indices. As we mentioned, Oh (2010b) considered incor-poration of ex-ante group heterogeneities, thus utilizing theMML-index to measure environmentally sensitive productivitygrowth at the macro-level.

The result showed that European countries are good at inno-vating and Asian countries are good at imitating and catching upwith the world frontier technology. Chiu et al. (2012) divided DMUsinto groups based on indicators of technological competitivenessand average per capita annual income in order to consider andappropriately control heterogeneity of technology. Wang Q. et al.(2013a) used the Metafrontier concept to analyze efficiency andproductivity in Chinese regions. Wang K et al. (2013b) utilized theDEA based models and Range-Adjusted Measure (RAM) to evaluatethe energy and environmental efficiency of China.

To consider environmental efficiency and the impact of regu-lation in dry land organic vine production, Ochoa et al. (2014)applied the Meta-production frontier concept with the DEAmethod. They concluded that organic agriculture is more efficientbecause it produces higher output than conventional agriculturefor the same quantity of inputs and pollutant emissions. Moreover,Jin et al. (2014) extended the DEA method with its stochasticconcept to measure environmental performance under randomconditions. They suggested that neglecting randomness wouldresult in biased environmental efficiency values. Thus, therandomness of both inputs and outputs should be seriouslyconsidered in actual applications. Research adopting the MMLmethod based on the Korean industry has not previously beenconducted to analyze environmental sensitive productivitygrowth. This research contributes to a deeper understanding of theKorean industrial structure and its performance accounting forundesirable effects.

3. Methodology and model

This section is composed of three sub-sections covering: (i)background to the MML-index, (ii) the fundamental assumptionsand (iii) the MML-index model. The first sub-section suggests thetheoretical progress to understand the index that we used. In thesecond section, wewill discuss the underlying assumptions and theissues of directional distance function, both of which are elementsof the index computation. Subsequently, we will construct the in-dex model and analyze the index into its underlying componentswith specific economic interpretation.

3.1. Background to the MML-index4

The MML-index is the expanded and developed form of theMalmquist index (M-index) which uses data envelopment analysis

4 Annex-I countries tend to represent developed countries and non-Annex-I re-fers to developing countries.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

(DEA) for its computation. By following the general character ofDEA, the M-index has many advantages in measuring productivitygrowth index.

It is widely applicable and does not require any assumption offunctional form for production function prior to estimation, as wellas assumption about distribution of any error terminology. Usingthe above characteristics, Farrell (1957) empirically obtained theinefficiency of a sample of firms with linear programming. Farrellsuggested many key concepts5 of DEA that became basic assump-tions to the measurement process. However, DEA was questionedby economics and econometric disciplines due to its disadvantageof not estimating the elasticity of input variables, as well as notconsidering characteristics of production and the statistical randomerror term. These disadvantages meant that DEA became regardedas a nonstandard econometric methodology, yet still as a frequentlyemployed benchmark alternative approach to compute productiv-ity growth.

In order to overcome the disadvantage of DEA as a non-parametric method, Aigner and Chu (1968) suggested a determin-istic parametric frontier analysis which continued to develop untilthe stochastic frontier analysis (SFA) method was introducedsimultaneously by Aigner et al. (1977) and Meeusen and Van denBroeck (1977). SFA has become a useful method to parametricallyanalyze the production and cost efficiency of producing units.

After Banker (1993) proved that DEA also has statistical prop-erties, the Malmquist index6 was utilized and continually devel-oped. Through the adoption of linear programming (LP)7 based ondirectional distance function, the MalmquisteLuenberger index(ML-index), the sequential MalmquisteLuenberger index (SML-index) and the MML-index have been developed and frequentlyapplied by many researchers to diverse areas. Regarding the matterof sensitivity of the results to the measurement method, themethods are often concurrently employed and their performancescompared.

3.2. The fundamental assumptions

This section deals with the fundamental assumptions which arerequired for defining the ML and MML-indices. The study of F€areet al. (2005) and Oh (2010b) are based on four basic assumptionsthat are followed here. They suggested that the production possi-bility set (PPS) for decision making units (DMUs) is represented bythe output set P(x), where DMUs produce M desirable outputs, Y2RMþ , and J undesirable by-products, b 2 Rjþ

8 The output set formsdesirable and undesirable output vector (y, b) that is jointly pro-duced from N inputs which is represented by the input vector, x2RNþ. The PPS is then expressed as follows:

PðxÞ ¼ fðy;bÞjx can produceðy;bÞg (1)

if x' � x then Pðx'ÞJPðxÞ (2)

Equation (1) illustrates the production technology mathemati-cally. In order to explain equation (1) clearly, it is necessary to

requirements that represented linear relationships.8 Following the study of Chung and F€are (1995), if a desirable output is produced

in a positive amount some undesirable output must also be produced. Thus, un-desirable output is bound as the positive set.

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 4: Measurement of environmentally sensitive productivity growth in Korean industries

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e124

discuss the undesirable outputs. Therefore, we employed thefollowing assumptions:

if ðx; y;bÞ2P and b ¼ 0; then y ¼ 0 (3)

ifðy;bÞ2PðxÞ and 0 � q � 1; then ðqy; qbÞ2PðxÞ (4)

if ðx; y;bÞ2P and y' � y; then�x; y';b

�2P (5)

if ðy;bÞ2PðxÞ and y' � y then�y';b

�2PðxÞ (6)

Jointness in production and disposability are the most impor-tant elements among production characteristics. Equation (3) ex-presses the characteristic called as null-jointness. It means thatdesirable output cannot be produced independently from unde-sirable output. Equation (4) shows the characteristic of weakdisposability. This condition describes that the reduction of theundesirable outputs is possible only when the desirable outputsdecreases with the same proportional contraction. Equation (5)indicates that the desirable outputs can be freely disposed. Thus,the desirable output may be decreased while simultaneouslydecreasing the undesirable outputs. The last assumption, equation(6) indicates that if an observed outputs vector is feasible, then anyoutput vector smaller than that is also feasible. This also means thatsome of the desirable outputs can always be disposed without anycost incurred. Using the above assumptions, the production possi-bility frontier combination of desirable and undesirable output isexpressed as Fig. 1. We assumed that A, B, C and D are observed asactual combined production points.

In Fig. 1, if the initial output set D, given as (y1, b1), is inefficientbecause it is placed at the interior of PPS, it means DMUs have tochoose their optimal strategy to reach the frontier of PPS. Arcelusand Arocena (2005) indicated that DMU tends to take their cir-cumstances into consideration while determining the productionlevel. In this situation, DMUs canmove to points A, B and C. Becausecurrently we discuss the method of simultaneous development ofeconomy and environment, the direction of A could be determinedas the appropriate direction that increases the desirable output anddecreases the undesirable output. Likewise, each DMU shouldmove toward the frontier in order to achieve a higher level of ef-ficiency, thus, b could be defined as an inefficiency index.

In order to draw a more detailed graph, directional distancefunction (DDF) is needed. Let g

. ¼ ð g.y; g!

bÞ be a direction vector,with g2RMþ � RJþ. Then, the directional distance function is definedas follows:

Fig. 1. Production possibility set combining production of desirable and undesirable outputs(1995), Fig. 1 can be illustrated as a convex production possibility set.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

D!�

x;y;b; g!y; g!

b

�¼max

nb :

�x;yþb g!y;b�b g!b

�2P

o(7)

There is no correct answer on how to find a right directionalvector, thus the reason/need for further research. As wementioned,the directional distance function seeks a way to increase thedesirable outputs while reducing the undesirable outputs at thesame time. Accordingly, g

.¼ð g.y; g!

bÞ determines the direction asshown in Fig. 1.

3.3. The MML-index model

With the logic described above, we are able to define the MML-index and decompose it for detailed analysis. In order to incorpo-rate metafrontier and directional distance function concepts in theindex computation, three definitions about benchmark technologyare needed. These include: (1) contemporaneous, (2) inter-temporaland (3) global benchmark technologies. The definition and notationused here are based on Oh (2010b)’s study, which in turn wereoriginally developed by Tulkens and van den Eeckaut (1995).

With theMML-indexmodel based onDEA, the concept of frontieris a benchmark against which the relative performance of industriesis measured. Industries are able to operate at an optimal efficiencylevel which is determined by the efficient industries that are locatedat the frontier. Generally, these industries use aminimumquantity ofinputs to produce the same quantity of outputs and we can beinferred that these industries have the “benchmark technology”.

The first benchmark technology, a contemporaneous benchmarktechnology of the group Gh is defined as PC

t

Gh, where the subscript h

represents individual DMU, h ¼ 1,…,H. It can be described as PCt

Gh¼

fðxt ; yt ;btÞ��xt can produce ðyt ;btÞg , where t ¼ 1, …,T. Thecontemporaneous benchmark technology indicates a referenceproduction set at time t. This set is made from observations at thattime only for the group Gh.

The second benchmark technology, the inter-temporal bench-mark technology of the group Gh is defined as PiGh

, in turn describedas PiGh

¼ PC1

Gh∪ PC

2

Gh∪ PC

3

Gh/∪ PC

T

Gh. It consists of a single reference pro-

duction set made from observations throughout the whole timeperiod for the group Gh. There can be H distinct inter-temporalbenchmark technologies. Industries in one inter-temporal bench-mark technology are assumed to be unable to easily access differentsuch technologies. Fig. 2 shows intuitively the relationship that theinter-temporal benchmark technology of a specific group of deci-sion units could envelope within its contemporaneous benchmarktechnologies.

The last benchmark technology, the global benchmark technologyof all groups is defined as PWGh

. The global benchmark technology

. According to the concept of ‘free disposal hull’ (FDH) of Tulkens and van den Eeckaut

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 5: Measurement of environmentally sensitive productivity growth in Korean industries

Fig. 2. Three concepts of benchmark technology.

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e12 5

establishes the only reference production set made from observa-tions throughout the entire time period for all groups. It can then bewritten as PWGh

¼ PiG1∪ PiG2

∪ PiG3∪/∪ PiGH

. As we can see from Fig. 2,the global benchmark technology covers all technology groups andall its enveloping inter-temporal technologies. For the purpose ofanalysis, we assume that it is possible for industries to reach theglobal technology, both theoretically and potentially, althoughthere may be obstacles in accessing other technologies.

The above three definitions allow us to formulate the MML-index, which is an advanced form of the ML-index. The contem-poraneous ML-index, which is based on the contemporaneousbenchmark technology between the time periods t and tþ1 isdefined as following suggested by Chung et al. (1997):

MLT�xt ; yt ;bt ; xtþ1; ytþ1;btþ1

�¼ 1þ D

!TC�xt ; yt ;bt

�1þ D

!TC�xtþ1; ytþ1;btþ1�

¼24 1þ D

!tc�xt ; yt ;bt

�1þ D

!tc�xtþ1; ytþ1;btþ1�$

1þ D!tþ1

c�xt ; yt ;bt

�1þ D

!tþ1c

�xtþ1; ytþ1;btþ1�

351=2

¼ ECT$TCT

(8)

where D!T

C means the contemporaneous directional distance func-tion following the logic of direction vector g

.as mentioned. For

simplicity, we replaced the directional distance functionD!ðx; y;b; g!y; g

!bÞ with D

!ðx; y;bÞ.9 Also, the superscript T ¼ t, tþ1and subscript ‘c’ indicate the ‘contemporaneous’ frontier.10

The geometric mean form of two consecutive contemporaneousML-productivity indices is typically used, expressed asMLT ¼ (MLt$MLtþ1)1/2. This ML-index can be broken down into ef-ficiency change (EC) and technical change (TC) components.

Through expansion of the logic of the ML-index and adjustmentof inter-temporal and global benchmark technologies, the MML-index can be derived as follows:

MMLT�xt ; yt ;bt ; xtþ1; ytþ1;btþ1

¼ 1þ D!T

G�xt ; yt ;bt ; yt ;bt

�1þ D

!TG�xtþ1; ytþ1;btþ1; ytþ1;btþ1� (9)

9 This replacement is applied with the MML-index.10 For comparison and analysis using both the ML and MML-indices, we tried toexplain the condition using the concept of “contemporaneous”.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

where D!T

Gðx; y;b; g!y; g!

bÞ ¼ maxfb : ðx; y þ b g!y;b� b g!bÞ2PGgand period T ¼ t, tþ1. It shows that the global directional distancefunction is defined in the global technology set, and D

!TI means the

inter-temporal directional distance function follows the logic ofdirection vector g

.. To extract more information from the MML-

index, a decomposition of productivity growth is needed. As aresult, the MML equation is written as:

MMLT�xt ; yt ;bt ; xtþ1; ytþ1;btþ1

�¼ 1þ D

!TG�xt ; yt ;bt

�1þ D

!TG

�xtþ1; ytþ1;btþ1

¼ 1þ D!t

C�xt ; yt ;bt

�1þ D

!tþ1C

�xtþ1; ytþ1;btþ1

�8<:1þ D

!TI�xt ; yt ;bt

�1þ D

!tC�xt ; yt ;bt

��1þ D

!tþ1C

�xtþ1; ytþ1;btþ1

1þ D!T

I

�xtþ1; ytþ1;btþ1

�9=;

�8<:1þ D

!TG�xt ; yt ;bt

�1þ D

!TI�xt ; yt ;bt

� �1þ D

!TI

�xtþ1; ytþ1;btþ1

1þ D!T

G

�xtþ1; ytþ1;btþ1

�9=;

¼ TEtþ1.TEt � BPRtþ1.

BPRt � TGRtþ1.TGRt

(10)

¼ EC� BPC � TGC (11)

The MML-index is broken down into a number of components,each of which is described in the above equations (10) and (11).Each of the three kinds of directional distance functions is based oneach of the three benchmark technology sets.

In Equation (10), TET means technical efficiency and BPRT in-dicates the best practice gap ratio which is derived by the gap be-tween the contemporaneous and the inter-temporal benchmarktechnologies in period T. TGRT is defined as the technology gap ratio,which is measured as the gap between the inter-temporal and theglobal benchmark technologies in period T.

In addition, equation (10) can be drawn as Fig. 2, and it showsthe three concepts of benchmark technology quite simply. It rep-resents one industry and two outputs during two time periods tomeasure productivity change.

Using Fig. 2, we can derive MML-index and its decomposedcomponents as below:

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 6: Measurement of environmentally sensitive productivity growth in Korean industries

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e126

MML�xt ; yt ;bt ; xtþ1; ytþ1;btþ1

�¼ 1þ AD=1þ EH

� �

¼1þAB1þEF

0B@1þAC=1þAB

1þEG=1þEF

1CA�

0B@1þAD=1þAC1þEH=1þEG

1CA (12)

where equation (12) shows three terms and which can be matchedwith the components of equation (10) sequentially.11

Through rewriting equation (10), we can get the form of equa-tion (11). Each term has a different meaning. At first, EC representsefficiency change and provides information on how much the gapto be closed at the contemporaneous benchmark technology at timetþ1 relative to the previous period t. If EC exceeds 1, it can beinterpreted as efficiency gain and also be meant that output levelgets closer compare to the contemporaneous benchmark technol-ogy frontier. Therefore, it can be termed the ‘catching up effect’economically.12

Next, BPC, best practice gap change, is regarded as ‘innovationefficiency’ or ‘technology change’13 because it is measured bychange in the best practice gap ratio during the two periods. Whena representative value of BPC is 1 and exceeds this value, it meansthat the contemporaneous benchmark technology frontier hasshifted towards the inter-temporal benchmark technology frontierand it can shift their contemporaneous benchmark frontier in thedirection of producing more desirable outputs and fewer undesir-able outputs.

TGC, technical gap change, denotes the ‘technology catching-upeffect’ among the DMUs, and it is related with changing of aninter-temporal benchmark technology frontier and a globalbenchmark technology frontier. When TGC is over 1, it could beinterpreted as a technological gap between any DMU at all periods,and the global frontier technology is reduced.

Before calculating the MML-index, the directional distancefunction should be estimated. There are a number of representativeways that have been previously introduced by Chung et al. (1997),Lee et al. (2002), Kumar (2006) and F€are et al. (2007). Previousstudies employed DEA with linear programming. As it wasmentioned, the DEA has both advantages and disadvantages. Theadvantages of DEA are that it does not need a specific objective(production) function and directional distance function. These ad-vantages provide convenience to interpret the results and releaserestrictions on parameters. Even though shadow price cannot beestimated by the DEA method, it is not a significant matter of thisstudy. Thus we followed the DEA with the linear programmingcalculation methodology suggested by Oh (2010a).

By using of linear programming properly and obtaining infor-mation in terms of productivity growth, setting up several differentlinear programming is an essential part of the exercise. In thisstudy, we should build six equations that are made up from acombination of three directional distance functions and two timeperiods at each DMU written as:

D!T

C;I;G

�xU;T ; yU;T ; bU;T

¼ maxnb :

�xU;T ; yU;T þ b g!y;b

U;T � b g!b

�2P

o

11 The point and D2 means the output set at t and tþ1.12 Efficiency change indicates that the change rate of efficiency is measured by theoutput per inputs.13 Technical change means that the introduction of new technology and man-agement process and the sufficient ability to utilize those factors.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

s:tPcon

lU;T$yU;TM � ð1þ bÞyU0; T

m ; m ¼ 1;/;M

Xcon

lU;T$bU;TJ ¼ ð1� bÞbU 0;Tj ; j ¼ 1;/; J

Xcon

lU;T$xU;TN � xU0;T

n ; n ¼ 1;/;N

lU � 0

where U represent each industry,Pcon

lU;T indicates the conditionfor constructing a PPS and the lU,T is the intensity variable whichindicates at what intensity a particular activity must be employedin the construction of a PPS. The value of directional distancefunction, which is estimated by DEA type linear programming, isan optimal solution for the calculation and decomposition of theMML-index.

4. Empirical analysis

4.1. Data resources and description

According to the model outlined above, we employed data forfourteen industries and its five variables of production consisting ofoutput, CO2 emission, capital stock, labor and energy input in Koreaover the period from 1981 to 2010.14 The period of study is deter-mined by availability of complete and comparable data at the in-dustry level. Gross output, capital stock and energy are expressed inmonetary terms, while CO2 and labor are measured in metric tonsand in working hours, respectively. Classification of industries usedin this study followed the division of detailed CO2 emission dataprovided by IEA CO2 emissions from Fuel Combustion Statistics. Inorder to match each sourced production component with classifi-cation of industries, we took into consideration the Korean stan-dard industry classification (KSIC) and the equivalent ofInternational standard industry classification (ISIC) Rev4. Since thedivision of detailed CO2 emission data is based on ISCI Rev4, weneed to make it consistent with the KSIC in order to merge the CO2data with other output and inputs variables.

We employed the KIP-DB data provided by Korean ProductivityCenter (KPC). It contained information about output, capital stock,energy and labor input obtained from KIP-DB. All variables exceptCO2 emission are deflated by producer price index for transformingnominal values into real values based in 2000. Table 1 showsclassification of industries and descriptive summary statistics of thevariables used.

Table 1 reports summary statistics of the data including averageand standard deviations of the variables. The summary shows thatthe Machinery industry has the largest average value of output andcapital stock among the sample industries. Transport industrygenerates the largest amount of CO2 emission; Agriculture is themost labor-intensive industry. Considering energy input, the com-bined Non-energy use industry/transformation/energy stands asthe dominant player in the market.

Under the assumption of general increasing trend, the value ofstandard deviation makes it possible to guess its dispersion andgrowth gap over the period from 1981 to 2010. The standard de-viation of output of Machinery and Transport equipment is quite

14 We have chosen variables that were considered by previous studies (Lozanoand Guti�errez, 2008; Zhou et al., 2010).

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 7: Measurement of environmentally sensitive productivity growth in Korean industries

Table 2Classification of Industry groups.

Group A Group B Group C Group D

Food andTobacco

Agriculture Chemical andPetrochemical

Non-specifiedindustry

Machinery Construction Iron and steel Paper, pulpand printing

Transportequipment

Textile andleather

Non-energyuse industry/transformation/energy

Transport

Wood andwood products

Non-metallicminerals

Table 1Descriptive statistics of the data, 1981e2010a.

ID Industry Ouput(billionwon)

CO2

(Mt ofton)

Capital(billionwon)

LABOR(millionhour)

Energy(billionwon)

1 Agriculture 36,022 6.193 43,379 6429 1013(5322) (2.667) (20,164) (2343) (218)

2 Chemical andPetrochemical

44,803 7.504 35,462 548 7545(27,687) (3.041) (23,656) (119) (3533)

3 Construction 87,618 1.492 64,739 3585 1225(36,358) (0.471) (57,075) (869) (293)

4 Food andTobacco

42,315 2.900 18,953 711 791(12,880) (0.743) (11,191) (91) (164)

5 Iron and steel 36,301 14.885 37,535 455 2629(20,088) (7.823) (24,329) (94) (1586)

6 Machinery 195,591 2.124 70,782 3755 3654(191,288) (0.688) (43,723) (687) (2383)

7 Non-energyuse industry/transformation/energy

31,253 10.650 15,143 123 19,062(16,825) (7.509) (10,994) (40) (14,408)

8 Non-metallicminerals

14,879 14.234 21,457 447 1909(7807) (4.480) (13,825) (109) (945)

9 Non-specifiedindustry

29,519 8.173 16,118 1088 1666(14,278) (4.969) (10,795) (154) (837)

10 Paper, pulp andprinting

13,167 3.271 10,599 380 561(6037) (1.335) (7498) (64) (268)

11 Textile andleather

38,331 4.483 28,421 2327 1692(5912) (1.075) (15,015) (1125) (813)

12 Transport 44,926 58.693 38,011 2295 6266(22,598) (27.903) (15,879) (436) (4886)

13 Transportequipment

61,890 1.649 51,827 1123 1090(49,456) (1.592) (38,258) (317) (665)

14 Wood andwood products

3334 0.197 2260 129 127(946) (0.074) (1304) (37) (35)

Note: Standard deviations in parenthesis.a The category of non-energy use industry/transformation/energy refers to Coke,

refined Petroleum and nuclear fuel, and it is defined as non-energy in industry,transformation and other energy industry own use. The category of non-specifiedindustry includes several industries such as rubber, plastics, other manufacturing,recycling and furniture.

16 We used the number of labor instead of cost of labor. The cost of labor accounts

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e12 7

large, meaning that these industries have been affected by thegovernment's initiated economic policy of 1980's and 1990's. Thispattern is consistent with the general development process ofKorean industry.15

4.2. Controlling for DMUs heterogeneity

The ML-index does not account for the heterogeneity origi-nating from each production technology that a DMU has. However,the MML-index considers heterogeneity of each DMU (industry inthis study) in production. To take into account such heterogeneityand to compute the MML-index, we need to categorize the DMUsinto several specific technology groups.

Because each industry has unique production technologies, in-dustries choose the optimal combination of inputs and maximizeoutputs depending on their production environment, geographicallocations and resource endowments (Huang et al., 2010). Huanget al. insisted that the heterogeneity of technology of industriesmight cause the observed units to operate under different frontiersdue to the influence of different operational environment. To acceptthis logic, some studies have employed geographical location toclassify countries into different industry groups (Battese et al.,2004; Oh and Lee, 2010).

15 From 1980's to early 90's, Korea had the fastest economic growth rate in theworld. In particular, the industry export promotion policy explains the developmentoutcome.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

On the other hands, the present study focuses on the interme-diate level of aggregation, as geographical location does not explainthe relationship between technology level and industries. TheWorld Economic Forum 2010 (WEF, 2010) asserts that the mainfactor of technology development is to be determined by learningand absorbing new technological knowledge and expenditures onR&D. Many studies related to technology, in this sense, use the levelof R&D or establishment of research centers as a proxy of uniqueproduction technology. In a similar manner, Iyer et al. (2006) andChiu et al. (2012) employed the technological competitiveness in-dicator and the average of annual income per capita to categorizethe technology levels of group-specific production frontiers.

We follow a similar approach as described above in the contextof production technology to split the sample into four kinds oftechnology groups that were defined based on two criteria. The firstcriterion is the output per unit of energy input (called energy ef-ficiency), and the second criterion is output per unit of labor (calledlabor productivity).16 The reason for choosing the first criteria is tobe able to capture technological competitiveness of energy effi-ciency in aspects of the environment friendly production progress.The reason for choosing the second criterion is to account for fea-tures of the industrial production structure.17 Table 2 shows theindustrial classification following the mentioned criterions.

As it can be seen in Table 2, Group A has the feature that higherenergy efficiency is attributed to environmental friendliness andthe high profitability nature of activities. It means that Group Acharacterizes industries that have environmentally comparativeadvantages of technology. On the contrary, Group D has theopposite character comparedwith that of Group A. Thus, Group D isdeemed to belong to industries that have comparative technolog-ical disadvantages. Group B has comparative advantages in terms ofenergy efficiency. However, it is characterized as a relatively lowlabor productivity industry. Group C holds opposite characteristicsto Group B. Consequently, Group C has comparative disadvantagesin terms of energy efficiency, but high labor productivity.

4.3. MML-index compared with ML-index

In contrast to the conventional ML-index, the MML-index canmeasure the environmentally sensitive productivity index. Thismeans that we are expected to capture the negative effect of CO2emission on the production of desirable output. Therefore, wefocused on comparing industrial productivity measured by the ML-index and the MML-index during same study period to isolate theeffects of CO2 emissions on productivity.

for the differences in wages of labor across different industries. The difference re-flects quality or human capital of labor.17 In reflecting on the feature of Korea economic development which is laborintensive industry structure, labor productivity is a proper classification standard tocontrol for technological heterogeneity.

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 8: Measurement of environmentally sensitive productivity growth in Korean industries

Table 3Summary statistics of key indicators by industry groups from 1981 to 2010.

Industrygroup

Output(billionwon)

CO2

(Mt ofton)

Capital(billionwon)

Labor(millionhour)

Energy(billionwon)

Group A 99,932 2.224 47,187 1863 1845(84,541) (1.008) (31,057) (365) (1071)

Group B 41,326 3.091 34,700 3117 1014(12,135) (1.071) (23,389) (1094) (340)

Group C 31,809 11.818 27,399 393 7786(18,101) (5.713) (18,201) (91) (5118)

Group D 29,204 23.379 21,576 1254 2831(14,304) (11.402) (11,391) (218) (1997)

Note: Standard deviations in parenthesis.

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e128

Prior to the comparative analysis, we report the average leveland growth rate of each production related variable for the periodof 1981e2010 by classification of industry groups.

In Table 3, we observe the specific different features of eachindustrial group as expected. Group A can be regarded as the largestproducer, taking the first rank in the levels of capital stock level.Group D shows the lowest level of output, whereas it emits moreCO2 than any other group. Because of the different features ofGroup B and Group C, they show the opposite tendency in the in-puts used and by-product produced. In particular, Group B is thelargest consumer of labor input and the lowest level of energy,while group C stands is at the opposite side.

When looking at the standard deviation, Table 3 shows evidenceof rapid growth during the study period. In order to verify thefeatures of rapid growth of industries in Korea, Table 4 reports theaverage annual growth rate of the three inputs, output and emis-sion indicators by each industry group. Using Table 4, Group A hasthe feature of higher energy efficiency and labor productivity dur-ing the whole period, showing the highest growth rate of output.But the average growth rate of CO2 emission of Group A is higherthan other Groups. Group B has the lowest percentage growth rateof labor input and Group C the higher value of capital and energyinput growth rates compared with other groups.

To analyze intuitively, we relate energy efficiency and laborproductivity to the criterion of grouping. However, some featuresare different than our initial considerations. Thus, we derive resultsin which we account for the possibility that the character of groupsmight be changed as time elapses. Therefore, for detailed analysisthat relates environmental policy to shape the industry

Table 4Average annual percentage growth rate of indicators by industry groups from 1981to 2010.

Group Industry Output CO2 Capital Labor Energy

Group A Aggregate 1.107 1.092 1.110 1.033 1.082Food and Tobacco 1.041 1.006 1.086 1.013 1.030Machinery 1.146 1.044 1.113 1.032 1.105Transport equipment 1.135 1.226 1.130 1.054 1.112

Group B Aggregate 1.032 1.026 1.093 0.980 1.023Agriculture 1.020 1.045 1.061 0.959 1.028Construction 1.065 1.038 1.142 1.022 1.041Textile and leather 1.013 0.994 1.076 0.959 1.001Wood and wood products 1.034 1.031 1.072 0.991 1.033

Group C Aggregate 1.085 1.071 1.104 1.035 1.100Chemical and Petrochemical 1.096 1.042 1.109 1.020 1.072Iron and steel 1.086 1.064 1.093 1.031 1.091Non-energy use industry/transformation/energy

1.073 1.106 1.111 1.054 1.138

Non-metallic minerals 1.088 1.040 1.107 1.004 1.081Group D Aggregate 1.073 1.077 1.089 1.018 1.084

Non-specified industry 1.087 1.108 1.101 1.018 1.058Paper, pulp and printing 1.071 1.045 1.114 1.016 1.071Transport 1.063 1.078 1.052 1.020 1.122

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

development, in Section 4.4 the period is divided into three sub-periods each consisting of one decade.

We computed the geometric mean of ML and MML-indices andbroke down both indices into their underlying components. Thegap of productivity change (DPC) is decomposed into the CO2emission effect and the heterogeneity effect. If any industry isidentified as having a negative DPC, we consider the industry to begood at controlling CO2 emission or having an advanced environ-ment friendly technology during the respective study period. Theresults are reported in Table 5.

In looking at details of ML index, the columns of EC and TC showthat many industries exceed 1 and the overall average level are1.0015 and 1.0160. Thus, technology advancement contributesmore to a higher level of PC than EC, resulting in the ‘catching up’effect. Following the definition of EC and TC, it could be interpretedthat the result of investment for achieving technology advance-ment gets an advantage of PC. According to the result of the ML-index, because both effects show positive values, they are to bepositive components of PC, 1.0175.

However, analysis of the MML-index tells a different storycompared to the results of the ML-index. The average of TGC in theMML-index is estimated to be below 1 and EC and BPC are close andgreater than 1. Thus, the catching up effect determined by the valueof EC does not contribute much in the case of the ML-index, andtechnology progress becomes a more important factor when weconsider CO2 emission as an undesirable output.

To be more specific regarding the MML analysis, except for Non-metallic minerals, all other industries reported an EC value greaterthan unity. As a result, those industries catch up with thecontemporaneous benchmark technology frontier. Also, Chemicaland Petrochemical, Machinery, Non-specified, Transport, TransportEquipment and Wood and Wood Product industries have BPCvalues larger than 1 which is interpreted as gain from innovationefficiency and technical change. It is worth mentioning that at last,Agriculture, Chemical and Petrochemical, Iron and Steel and Non-metallic minerals industries are shown to experience the technol-ogy ‘catching-up’ effects. Following this finding, the Chemical andPetrochemical industries can be regarded as global innovator18

with the positive and highest DPC.As a result, the value of PC is a combination of EC, BPC and TGC,

and five industries including Chemical and Petrochemical, Iron andSteel, Machinery, Non-Metallic Minerals, and Transport Equipmentachieved positive productive growth rates.

The last column of Table 5 is the gap between the PC of ML andMML. When this value is positive, it means that these groups didnot prepare efficiently the policy of reducing CO2 emission or thatthe policy may be an obstacle towards gaining higher productivity.In any case, the control of undesirable output, CO2 emission, affectnegatively the PC of MML and then DPC increase in general.

For an in-depth analysis of the MML-index, we divided the in-dustries into homogenous groups by applying the two criteria.Table 6 shows how different PC are calculated and it shows theresults in which all of the groups except Group B have smaller PCthan the case of ML e which indicates the case when CO2 emissionis not considered in the measurement of productivity.

Also, as expected the results show that the group that has atechnical advantage could gain a higher PC. The average rate ofproductivity growth of Group A (1.09%) was the highest and that ofGroup C and D (�0.27%) were the lowest. A detailed table withdisaggregated period is provided in Appendix A.

In looking at different aspects of decomposition of productivitychange, the average rate of best practice technology gap change in

18 When both terms, BPC and TGC, exceed 1, it can be called ‘global innovator’.

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 9: Measurement of environmentally sensitive productivity growth in Korean industries

Table 5Productivity growth computed based on ML-index and MML-index methods for the period from 1981 to 2010.

Group Industry ML-index MML-index DPC

EC TC PC EC BPC TGC PC

B Agriculture 0.9756 1.0169 0.9921 1.0000 0.9962 1.0029 0.9990 �0.0069C Chemical and petrochemical 1.0249 1.0300 1.0557 1.0017 1.0077 1.0098 1.0193 0.0364B Construction 1.0000 0.9931 0.9931 1.0000 1.0000 0.9994 0.9994 �0.0063A Food and tobacco 0.9948 1.0148 1.0095 1.0000 0.9965 0.9998 0.9962 0.0133C Iron and steel 1.0016 1.0418 1.0435 1.0000 1.0000 1.0192 1.0192 0.0243A Machinery 1.0241 1.0299 1.0547 1.0001 1.0223 1.0000 1.0224 0.0322C Non-energy use industry/

transformation/energy1.0000 0.9982 0.9982 1.0000 1.0000 1.0000 1.0000 �0.0018

C Non-metallic minerals 1.0271 1.0335 1.0615 0.9886 0.9990 1.0139 1.0013 0.0602D Non-specified industry 0.9970 0.9903 0.9873 1.0000 1.0049 0.9950 0.9999 �0.0126D Paper, pulp and printing 0.9838 1.0235 1.0069 1.0000 0.9994 0.9925 0.9919 0.0150B Textile and leather 0.9899 0.9863 0.9764 1.0000 0.9988 0.9943 0.9931 �0.0167D Transport 1.0061 1.0285 1.0348 1.0000 1.0153 0.9849 1.0000 0.0348A Transport equipment 1.0111 1.0222 1.0335 1.0000 1.0144 0.9996 1.0141 0.0194B Wood and wood products 0.9842 1.0145 0.9985 1.0000 1.0114 0.9863 0.9975 0.0010

Average 1.0015 1.0160 1.0175 0.9993 1.0047 0.9998 1.0038 0.0137

Notes: Efficiency change (EC), technical change (TC), productivity change (PC), best practice gap change (BPC), and the technical gap change (TGC). DPC¼PC_ML�PC_MML.

Table 7ML and MML-indices result and their underlying components by time periods.

Period ML-Index MML-Index DPC

EC TC PC EC BPC TGC PC

1981e2010 1.0016 1.0162 1.0180 0.9994 1.0052 0.9993 1.0038 0.01421981e1990 1.0132 1.0007 1.0139 0.9956 0.9976 1.0018 0.9947 0.01911991e2000 1.0109 1.0091 1.0203 0.9956 1.0177 0.9969 1.0096 0.01072001e2010 0.9827 1.0382 1.0201 1.0071 1.0002 0.9998 1.0066 0.0134

Note: Efficiency change (EC), technical change (TC), productivity change (PC), bestpractice gap change (BPC), and technology growth change(TGC),DPC ¼ PC_ML�PC_MML.

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e12 9

Group A is the highest (1.11% per year). This finding indicates thatthe main factor of enhancement of PC would be BPC, best practicegap change, which represents the inter-temporal ‘innovation effi-ciency’ or ‘technology change’ in each industry.

Another term related to technology progress, namely TGC, in-dicates that Group C is good at inventing new technology orinnovation and does play a role as a global innovator group ratherthan reducing its endowed inputs to catch up with the frontiertechnology. On the other hand, Group B and D showed similardecomposition results that the PC of both groups did not reach theunity value, and TGC could not raise its PC value.

These groups do not show a better efficiency change, whereasthe indicator of BPC became better. This implies that both groupswere shown to be good in capturing “innovation efficiency”,reaching closer to the inter-temporal frontier technology.

4.4. Further analysis of the result

In Section 4.3, we analyzed the results from MML-index by in-dustry groups. In this Section, we add another condition to analyzethe results we already have in a more detailed form.

As wementioned above, Korea has industrialized in a very shorttime, and also the structure of industry has changed a great deal.Since the 90's Korea has realized that quality of economic growth ismore important than quantitative expansion of the economy. Thus,the country began to regulate its air pollution in earnest since mid-90's. We split the study period into three sub-periods, corre-sponding to three decades, in order to capture policy effects onproductivity by accounting for heterogeneity across both industrygroups and decades.

Table 7 reports the results of the ML and MML-indices by timeperiods. The overall result in a more detailed form is presented inAppendix A and B.

As it is shown in Table 7, we can demonstrate more clearly whatwe have previously discussed. Due to considering the negative

Table 6MML-index result and its decomposition by groups of industries from 1981 to 2010.

Group EC BPC TGC PC DPC

A 1.0000 1.0111 0.9998 1.0109 0.0217B 1.0000 1.0016 0.9957 0.9973 �0.0072C 0.9976 1.0017 1.0107 1.0099 0.0298D 1.0000 1.0066 0.9908 0.9973 0.0124

Note: Efficiency change (EC), productivity change (PC), best practice gap change(BPC), and the technical gap change (TGC), DPC¼PC_ML�PC_MML.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

outputs and heterogeneity of each industry production technology,PC of MML is shown to be of lower value in all decades. There aredifferences among the decades. However, we get implicitly theresult that the strengthening trend of environmental policy in the90's gave a bigger negative impact on the productivity change.

In the ML-index section, EC is the highest (1.32%) in the1981e1990 and lowest (�1.73%) in the 2001e2010. On the otherhand, TC is the highest (3.82%) in the 2001e2010 and the lowestlevel (0.07%) is attributed to the 1981e90. While in the case of theMML-index, the highest EC value is 0.71% in the third sub-periodand the lowest value is �0.44%. The best practice gap change, BPC,shows the highest value (1.77%) in the 1991e2000 and the lowest(�0.24%) in the 1981e1990. TGC in the first sub-period showing agood result in catching up with the frontier technology.

According to Table 7 and Appendix A, there is evidence thattechnological progress plays an important role in achieving thegoals of production. The indicators related to technical progress aremore important factors to maintain or increase productivitychange, rather than factor inputs in the form of greater endowmentresources that industries can use in their production activities.19

In summary, we can derive some implicit conclusions throughAppendix A. The highest MML-index among all industries and timeperiods is 4.53% in Machinery in the second sub-period and such aresult was caused by its high value of BPC (5.31%). The dramaticchange in DPC is shown in Iron and Steel (7.19%) from2001e2010,and Transport Equipment (�2.83%) in the 1981e1990. Iron andSteel are the representative industries to explain government

19 Most of industries in each sub-period in MML show the unity estimates of EC,however, PC exceeds 1. Thus technology indicators, BPC and TGC, contribute toobtain a positive value of PC.

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 10: Measurement of environmentally sensitive productivity growth in Korean industries

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e1210

initiated economic policy in 80's. When environmental policy wasnot sufficiently introduced, those industries gained an advantagecompared with the recent period.

However, as eco-friendly policies are adopted, this industryshould follow the policy to survive and finally maintain acomparatively higher PC compared with other industries in the2000's. There was no demands on Transport Equipment to considerCO2 in relation with its production in the early stage of the Koreaneconomic development. This industry became one of the mainpillars of Korean economy growth. During this period, the demandfor transport equipment increased rapidly, and the industryexpanded its supply without considering CO2 emission. In this case,this particular industry lost their productivity when interconnect-ing with environmental policy. Thus, Iron and Steel is able to beinterpreted as the most well-prepared and adaptable industry inthe era of reducing CO2 emission. On the other hand, TransportEquipment faced difficulties in attaining its high rate of produc-tivity as before.

After introducing the environmental policy, some of the in-dustries resisted environmental regulation and others adjustedwell in overcoming those policies. The global innovators in eachsub-period can be regarded as well-adjusted industries usingtechnological progress to move through production constraints.From 1981 to 1990, non-specified industries which include rubber,plastics, other manufacturing, recycling and furniture are globalinnovators. However, it is hard to define a specific type of industry.In 1991e2000, Food and Tobacco, Agriculture, Chemical andPetrochemical and Non-Metallic Minerals became global in-novators. In 2001e2010, Machinery, Chemical and Petrochemicalalso became global innovators.

To show the general results, we reported the disaggregated formfor each industry group in Appendix B. According to the criteria ofgrouping and PC in Appendix B, we could derive the implicitconclusion that environment policy effects obviously occurred inthe 2001e2010. Many environmental policies started in the 90'sand this explains the difference before the 80's and 2000's. Thegroups with comparatively higher labor productivity (Group A andC) are better able to follow and invest in technological progressthan others. Indeed, the lower labor productivity groups, Group Band Group D change their negative PC value to positive. Thesetendencies mean that these industries suffered from changing theirproduction technology in the short run to adapt to environmentalpolicies.

Also, comparing theMML-indextotheML-index,Korean industriesneed more concentration in finding the proper direction productionneeds to take towards more production and less CO2 emission.

5. Conclusion

The objective of this study was to find evidence about theimpact of environmental policies upon industries productivity. Anattempt was made to identify a leading industry of technologyprogress and derive the answer as to whether industry preparationin facing environmental regulations is productive or not.

In order to achieve the goal of this study, we employed theMML-index which relates environmental constraint, reducing CO2

emission, and productivity change. Also we compared the MML-index to the ML-index to amplify two concepts. The first weaddressed was whether bad output affected productivity and thesecond was heterogeneity of DMUs and its implications. For con-trolling the heterogeneity of DMUs caused from difference in pro-duction technology, we divided the industries into four groups withtwo criteria which were energy efficiency and labor productivity.

To consider the two concepts, we are able to identify whichindustry and industrial groups did experience technological

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

progress during the study period. Also, our results providedanalyzed productivity growth regarding efficiency change andtechnological change. For an in-depth study, we divided the part oftechnology change into innovation efficiency and the technological‘catch up’ effect. Through the MML results, we found that most ofthe industries have increased the average productivity growth rateas time elapsed. However, lower estimates were generally reportedthan those of the ML-index. This means that industry suffered fromconstraints during production. Therefore, we reached one of ourobjectives of this study e to know the negative impacts of envi-ronmental policy of reducing CO2 emissions upon economy.

Though CO2 emission was considered to be a constraint onproduction, the MML-index of a few industries show a higherproductivity value than the ML-index approach. Notably, Chemicaland Petrochemical industries lead the whole industrial nation asglobal innovators with higher technology progress during theentire sample period. These industries are representative industriesthat enhance the internal and external technological levels.Therefore, we also achieved the second objective of this study.

And lastly, we attempted to answerwhether preparation againstchanging production circumstances is productive or not. Thoughsome exceptions exist, most average technology indicated positiveprogress under environmental constraints and low efficiencychange. These results indicated that most industries carried out aninvestment aimed at reducing CO2. This shows that Korean indus-trial sectors are regarded as well prepared in adapting to new andrestrictive environmental policies.

To conclude, our findings indicate that industries that can beregarded as the primary contributor to the Korean economicgrowth e such as Food and Tobacco, Agriculture, Machinery,Chemical and Petrochemical and Non-Metallic Minerals e havehad problems in facing the tightened environmental policy ineach period. Primary contributors like these industries have thewide spillover effect of investment in technology. By leading andsolving the matter of the environment in its economy, the Koreangovernment needs to set up favorable environment policies forthe primary contributors to assist them in investing in thedevelopment of new technologies. Effectiveness of such policies isobserved in Krautzberger and Wetzel (2012) in context of EU andNorway.

Considering the analysis by sub-periods, we could also find evi-dence of heterogeneity among industrial sectors. The groups with ahigh level of labor productivity tended to get a higher annual rate ofproductivity change. We can understand that the groups producingconsiderably pooroutputs tend to have a strongmotivation to reducetheir CO2 emission, to invest inpromoting technological progress andto utilize technology to use energy more efficiently. Furthermore,such tendencies are clearly evident as time passes. Our results couldbe interpreted as restructuring at the industrial level changes in in-dustrial production and environmental conditions.

After 2015, Korea may be obliged to reduce its GHG emissions.Given this situation, each industry would face new market environ-mental conditions in earnest. In order to maintain global competi-tiveness, comprehensive investment programs to change productionprocesses or to innovate new products and processes should beachieved in the near future. In conclusion, the government has tofocus on supporting in the short-run the primary contributors toeconomic growth in order to maintain competitiveness in the globalmarket. As far as the long-run, the government also should be con-cerned about the acquisition of technology for less profitable in-dustries which do not possess enough funds to invest in new energysaving and environmental friendly technologies.

This study attempted to make a contribution to the literature byconsidering undesirable output in the calculation of productivitygrowth at the industry level. However, it could not present clearly

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 11: Measurement of environmentally sensitive productivity growth in Korean industries

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e12 11

the interaction process of desirable and undesirable outputs, due tothe imposition of necessary and excessive assumptions.

In addition, we were not able to consider the issue ofmanufacturing production relocated abroad serving as exportinggreenhouse gases (Grimes, 2003). Thus, CO2 could be under-estimated and productivity changemight have been overestimated.The data set used here has a limitation in terms of partial dataavailability at the industry level. Also, due to the data availabilityproblem, we could not employ other inputs like material, servicesand information technology. Consequently, the results could bebiased due to the lack of full information about the production

ML and MML-indices results and their components by industries, industry groups and p

Group Industry Period ML-index

EC TC

A Food and tobacco 1981e2010 0.9948 1.01481981e1990 1.0000 1.00111991e2000 1.0000 1.02552001e2010 0.9850 1.0166

Machinery 1981e2010 1.0241 1.02991981e1990 1.0097 1.00961991e2000 1.0622 1.02372001e2010 1.0000 1.0547

Transport equipment 1981e2010 1.0111 1.02221981e1990 1.0015 1.01031991e2000 1.0351 1.03302001e2010 0.9961 1.0222

B Agriculture 1981e2010 0.9756 1.01691981e1990 0.9541 1.03121991e2000 0.9849 1.02702001e2010 0.9860 0.9943

Construction 1981e2010 1.0000 0.99311981e1990 1.0000 0.99191991e2000 1.0000 0.99372001e2010 1.0000 0.9936

Textile and leather 1981e2010 0.9899 0.98631981e1990 0.9834 0.97771991e2000 1.0080 0.96972001e2010 0.9779 1.0112

Wood and wood products 1981e2010 0.9842 1.01451981e1990 1.0034 1.00101991e2000 0.9958 0.97662001e2010 0.9560 1.0667

C Chemical and petro-chemical 1981e2010 1.0249 1.03001981e1990 1.0253 1.03241991e2000 1.0634 1.02142001e2010 0.9875 1.0364

Iron and steel 1981e2010 1.0016 1.04181981e1990 1.0407 1.03071991e2000 1.0005 1.03842001e2010 0.9687 1.0553

Non-energy use industry/transformation/energy

1981e2010 1.0000 0.99821981e1990 1.0000 1.00551991e2000 1.0000 0.97842001e2010 1.0000 1.0117

Non-metallic minerals 1981e2010 1.0271 1.03351981e1990 1.0529 1.00941991e2000 1.0189 1.03212001e2010 1.0126 1.0571

D Non-specified industry 1981e2010 0.9970 0.99031981e1990 1.0654 0.96961991e2000 0.9800 0.96282001e2010 0.9554 1.0380

Paper, pulp and printing 1981e2010 0.9838 1.02351981e1990 1.0055 0.96511991e2000 0.9931 1.01952001e2010 0.9557 1.0833

Transport 1981e2010 1.0061 1.02851981e1990 1.0317 0.99311991e2000 1.0062 1.01652001e2010 0.9837 1.0739

Notes: Efficiency change (EC), technical change (TC), productivity change (PC), best practi

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

structure of industries. We will take up consideration of theselimitations in our follow-up study.

Acknowledgments

The authorswish to thank D.H. Oh, two anonymous referees and aneditorof the journal forcommentsandsuggestionsonanearlier versionof this paper. Almas Heshmati gratefully acknowledges financial sup-port from Sogang University Climate Change Research Fund (SRF).

Appendix A

eriods of time.

MML-index DPC

PC EC BPC TGC PC

1.0095 1.0000 0.9965 0.9998 0.9962 0.01331.0011 1.0000 0.9530 0.9969 0.9500 0.05111.0255 1.0000 1.0380 1.0004 1.0384 �0.01301.0014 1.0000 0.9957 1.0018 0.9975 0.00391.0547 1.0001 1.0223 1.0000 1.0224 0.03221.0194 1.0005 1.0038 1.0000 1.0043 0.01511.0874 1.0000 1.0531 0.9926 1.0453 0.04211.0547 1.0000 1.0087 1.0075 1.0163 0.03841.0335 1.0000 1.0144 0.9996 1.0141 0.01941.0118 1.0000 1.0401 1.0000 1.0401 �0.02831.0692 1.0000 1.0078 0.9982 1.0060 0.06321.0182 1.0000 0.9984 1.0007 0.9992 0.01910.9921 1.0000 0.9962 1.0029 0.9990 �0.00690.9839 1.0000 0.9881 1.0090 0.9970 �0.01311.0115 0.9420 1.0587 1.0029 1.0002 0.01130.9804 1.0616 0.9442 0.9974 0.9998 �0.01930.9931 1.0000 1.0000 0.9994 0.9994 �0.00630.9919 1.0000 1.0000 0.9778 0.9778 0.01400.9937 1.0000 0.9964 1.0193 1.0156 �0.02190.9936 1.0000 1.0037 0.9992 1.0029 �0.00920.9764 1.0000 0.9988 0.9943 0.9931 �0.01670.9616 0.9440 0.9971 1.0379 0.9768 �0.01530.9775 1.0533 0.9989 0.9502 0.9998 �0.02230.9889 1.0000 1.0002 1.0011 1.0013 �0.01240.9985 1.0000 1.0114 0.9863 0.9975 0.00101.0044 1.0000 1.0113 0.9782 0.9893 0.01520.9726 1.0000 1.0267 0.9739 0.9999 �0.02731.0197 1.0000 0.9963 1.0063 1.0025 0.01721.0557 1.0017 1.0077 1.0098 1.0193 0.03641.0585 0.9858 1.0182 0.9966 1.0003 0.05831.0862 1.0179 1.0057 1.0007 1.0245 0.06171.0235 1.0000 1.0003 1.0311 1.0314 �0.00791.0435 1.0000 1.0000 1.0192 1.0192 0.02431.0727 1.0000 1.0000 1.0008 1.0008 0.07191.0390 1.0000 1.0000 1.0155 1.0155 0.02341.0222 1.0000 1.0000 1.0397 1.0397 �0.01750.9982 1.0000 1.0000 1.0000 1.0000 �0.00181.0055 1.0000 1.0000 0.9933 0.9933 0.01230.9784 1.0000 0.9945 1.0059 1.0004 �0.02201.0117 1.0000 1.0055 1.0002 1.0057 0.00601.0615 0.9886 0.9990 1.0139 1.0013 0.06021.0628 1.0000 0.9512 1.0515 1.0001 0.06271.0516 0.9708 1.0238 1.0062 1.0001 0.05161.0704 0.9964 1.0189 0.9887 1.0037 0.06670.9873 1.0000 1.0049 0.9950 0.9999 �0.01261.0330 1.0000 1.0159 1.0020 1.0180 0.01510.9435 1.0000 1.0000 0.9829 0.9829 �0.03940.9917 1.0000 1.0000 1.0008 1.0008 �0.00911.0069 1.0000 0.9994 0.9925 0.9919 0.01500.9704 1.0000 0.9924 0.9807 0.9732 �0.00281.0125 1.0000 0.9930 1.0073 1.0002 0.01231.0353 1.0000 1.0124 0.9884 1.0007 0.03461.0348 1.0000 1.0153 0.9849 1.0000 0.03481.0246 1.0000 0.9920 1.0079 0.9998 0.02481.0227 0.9597 1.0421 0.9999 1.0000 0.02271.0564 1.0420 1.0102 0.9501 1.0001 0.0564

ce gap change (BPC), and technology growth change (TGC).DPC ¼ PC_ML�PC_MML.

f environmentally sensitive productivity growth in Korean industries,014.06.030

Page 12: Measurement of environmentally sensitive productivity growth in Korean industries

Y. Chung, A. Heshmati / Journal of Cleaner Production xxx (2014) 1e1212

Appendix B

ML and MML-indices results and their decomposition by industry groups and time periods.

Group Period ML-index MML-index DPC

EC TC PC EC BPC TGC PC

A 1981e2010 1.0100 1.0223 1.0326 1.0000 1.0111 0.9998 1.0109 0.02171981e1990 1.0037 1.0070 1.0108 1.0002 0.9990 0.9990 0.9981 0.01261991e2000 1.0324 1.0274 1.0607 1.0000 1.0330 0.9971 1.0299 0.03082001e2010 0.9937 1.0312 1.0248 1.0000 1.0010 1.0033 1.0043 0.0205

B 1981e2010 0.9874 1.0027 0.9900 1.0000 1.0016 0.9957 0.9973 �0.00721981e1990 0.9852 1.0005 0.9854 0.9860 0.9991 1.0007 0.9852 0.00021991e2000 0.9972 0.9918 0.9888 0.9988 1.0202 0.9866 1.0039 �0.01512001e2010 0.9800 1.0165 0.9957 1.0154 0.9861 1.0010 1.0016 �0.0059

C 1981e2010 1.0134 1.0259 1.0397 0.9976 1.0017 1.0107 1.0099 0.02981981e1990 1.0297 1.0195 1.0499 0.9964 0.9923 1.0105 0.9986 0.05131991e2000 1.0207 1.0176 1.0388 0.9972 1.0060 1.0071 1.0101 0.02872001e2010 0.9922 1.0401 1.0320 0.9991 1.0062 1.0149 1.0201 0.0118

D 1981e2010 0.9956 1.0141 1.0097 1.0000 1.0066 0.9908 0.9973 0.01241981e1990 1.0342 0.9759 1.0093 1.0000 1.0001 0.9969 0.9970 0.01231991e2000 0.9931 0.9996 0.9929 0.9866 1.0117 0.9967 0.9944 �0.00152001e2010 0.9649 1.0651 1.0278 1.0140 1.0075 0.9798 1.0005 0.0273

Averages 1981e2010 1.0016 1.0162 1.0180 0.9994 1.0052 0.9993 1.0038 0.01421981e1990 1.0132 1.0007 1.0139 0.9956 0.9976 1.0018 0.9947 0.01911991e2000 1.0109 1.0091 1.0203 0.9956 1.0177 0.9969 1.0096 0.01072001e2010 0.9827 1.0382 1.0201 1.0071 1.0002 0.9998 1.0066 0.0134

Notes: Efficiency change (EC), technical change (TC), productivity change (PC), best practice gap change (BPC), and technology growth change(TGC).DPC ¼ PC_ML�PC_MMLD1.

References

Aigner, D.J., Chu, S.F., 1968. On estimating the industry production function. Am.Econ. Rev. 58 (4), 826e839.

Aigner, D., Lovell, C.A.A., Schmidt, P., 1977. Formulation and estimation of stochasticfrontier production function models. J. Econ. 6 (1), 21e37.

Arcelus, F.J., Arocena, P., 2005. Productivity differences across OECD countries in thepresence of environmental constraints. J. Operat. Res. Soc. 56, 1352e1362.

Banker, R.D., 1993. Maximum likelihood, consistency and data envelopment anal-ysis: a statistical foundation. Manag. Sci. 39 (10), 1265e1273.

Battese, G.E., Rao, D.S.P., 2002. Technology gap, efficiency, and a stochastic meta-frontier function. Int. J. Bus. Econ. 1, 87e93.

Battese, G.E., Rao, D.S.P., O'Donnell, C.J., 2004. A metafrontier production functionfor estimation of technical efficiencies and technology gaps for firms operatingunder different technologies. J. Prod. Anal. 21, 91e103.

Brock, W.A., LeBaron, B., 1996. A dynamic structural model for stock return volatilityand trading volume. Rev. Econ. Stat. 78, 94e110.

Caves, D.W., Christensen, L.R., Diewert, W.E., 1982. Multilateral comparisons ofoutput, input, and productivity using superlative index numbers. Econ. J. 92(365), 73e86.

Chiu, C.E., Liou, J.L., Wu, P.I., Fang, C.L., 2012. Decomposition of the environmentalinefficiency of the Metafrontier with undesirable output. Energy Econ. 34,1391e1399.

Chung, Y.H., F€are, R., 1995. Productivity and Inseparable Outputs: A DirectionalDistance Function Approach. In: Discussion Paper Series No. 95e24 November.

Chung, Y.H., F€are, R., Grosskopf, S., 1997. Productivity and undesirable outputs: adirectional distance function approach. J. Environ. Manag. 51, 229e240.

F€are, R., Grosskopf, S., Noh, D.-W., Weber, W., 2005. Characteristics of a pollutingtechnology: theory and practice. J. Econ. 126, 469e492.

F€are, R., Grosskopf, S., Pasurka Jr., C.A., 2001. Accounting for air pollution emissions inmeasures of state manufacturing productivity growth. J. Regional Sci. 41, 381e409.

F€are, R., Grosskopf, S., Pasurka Jr., C.A., 2007. Environmental production functionsand environmental directional distance functions. Energy 32, 1055e1066.

Farrell, M.J., 1957. The measurement of productive efficiency. J. Royal Stat. Soc. Ser.General. 120 (3), 253e290.

Grimes, P., 2003. Exporting the greenhouse: foreign capital penetration and CO2emissions 1980e1996. J. World-Syst. Res. 9 (2), 261e275.

Hansen, B., Seo, B.S., 2002. Testing for two-regime threshold cointegration in vectorerror-correction models. J. Econ. 110 (2), 293e318.

Hsieh, D., 1991. Chaos and nonlinear dynamics: applications to financial markets.J. Finance 47, 1145e1189.

Huang, Y.J., Chen, K.H., Yang, C.H., 2010. Cost efficiency and optimal scale of elec-tricity distribution firms in Taiwan: an application of metafrontier analysis.Energy Econ. 32, 15e23.

Iyer, K., Rambaldi, A., Tang, K.K., 2006. Globalization and the Technology Gap:Regional and Time Evidence, Leading Economic and Managerial issuesInvolving Globalization. Nova Science New York, United States, pp. 213e227.

Jin, J., Zhou, D., Zhou, P., 2014. Measuring environmental performance with sto-chastic environmental DEA: the case of APEC economies. Econ. Model. 38,80e86.

Please cite this article in press as: Chung, Y., Heshmati, A., Measurement oJournal of Cleaner Production (2014), http://dx.doi.org/10.1016/j.jclepro.2

Kim, S., Lee, K., Nam, K., 2010. The relationship between CO2 emissions and eco-nomic growth: the case of Korea with nonlinear evidence. Energy Policy 38,5938e5946.

Krautzberger, L., Wetzel, H., 2012. Transport and CO2: productivity growth andcarbon dioxide emission in the European commercial transport industry. En-viron. Resour. Econ. 53, 435e454.

Kumar, S., 2006. Environmentally sensitive productivity growth: a global analysisusing MalmquisteLuenberger index. Ecol. Econ. 56, 280e293.

Lee, J.D., Park, J.B., Kim, T.Y., 2002. Estimation of the shadow prices of pollutantswith production/environment inefficiency taken into account: a nonparametricdirectional distance function approach. J. Environ. Manag. 64, 365e375.

Lozano, S., Guti�errez, E., 2008. Non-parametric frontier approach to modelling therelationships among population, GDP, energy consumption and CO2 emissions.Ecol. Econ. 66, 687e699.

Luukkonen, R., Saikkonien, P., Ter€asvirta, T., 1988. Testing linearity against smoothtransition autoregressive models. Biometrika 75 (3), 491e499.

Meeusen, W., Van den Broeck, J., 1977. Technical efficiency and dimension of thefirm: some results on the use of frontier production functions. Empir. Econ. 2(2), 109e122.

Ochoa, A., Oliva, V., Miura, A., 2014. Environmental efficiency and the impact ofregulation in dryland organic vine production. Land Use Policy 36, 275e284.

Oh, D.H., 2010a. A global MalmquisteLuenberger productivity index. J. Prod. Anal.34, 183e197.

Oh, D.H., 2010b. A metafrontier approach for measuring an environmentally sen-sitive productivity growth index. Energy Econ. 32, 146e157.

Oh, D.H., Heshmati, A., 2010. A sequential MalmquisteLuenberger productivityindex: environmentally sensitive productivity growth considering the pro-gressive nature of technology. Energy Econ. 32, 1345e1355.

Oh, D.H., Lee, J.D., 2010. A metafrontier approach for measuring Malmquist pro-ductivity index. Empir. Econ. 38, 47e64.

Sueyoshi, T., Goto, M., 2010. Should the US clean air act include CO2 emissioncontrol?: examination by data envelopment analysis. Energy Policy 38,5902e5911.

Tulkens, H., van den Eeckaut, P., 1995. Non-parametric efficiency, progress andregress measures for panel data: methodological aspects. Eur. J. Operat. Res. 80,474e499.

Wang, Q., Zhao, Z., Zhou, P., Zhou, D., 2013a. Energy efficiency and productiontechnology heterogeneity in China: a meta-frontier DEA approach. Econ. Model.35, 283e289.

Wang, K., Wei, Y.M., Zhang, X., 2013b. Energy and emissions efficiency patterns ofChinese regions: a multi-directional efficiency analysis. Appl. Energy 104,105e116.

Weber, W., Domazlicky, B., 2001. Productivity growth and pollution in statemanufacturing. Rev. Econ. Stat. 83, 195e199.

Yang, C.H., Tseng, Y.H., Chen, C.P., 2011. Environmental Regulations, induced R&D,and Productivity: Evidence from Taiwan’s Manufacturing Industries. In:Working Paper Series, vol. 2011e18. November 2011.

Zhou, P., Ang, B.W., Han, J.Y., 2010. Total factor carbon emission performance: aMalmquist index analysis. Energy Econ. 32, 194e201.

f environmentally sensitive productivity growth in Korean industries,014.06.030