Energy Economics 44 (2014) 89–98

Contents lists available at ScienceDirect

Energy Economics

j ourna l homepage: www.e lsev ie r .com/ locate /eneco

‘Green’ productivity growth in China's industrial economy

Shiyi Chen a,⁎, Jane Golley b,1

a School of Economics, China Center for Economic Studies, Fudan University, Shanghai, Chinab Australian Centre on China in the World, College of Asia and the Pacific, Coombs Building, The Australian National University, Canberra, Australia

⁎ Corresponding author. Tel.: +86 21 6564 2050; fax: +E-mail addresses: shiyichen@fudan.edu.cn (S. Chen), jan

1 Tel.: +61 2 6125 3366; fax: +61 2 6125 0745.2 Wen Jiabao first made this declaration in March 2007

then, most recently in his final report as Premier in March

http://dx.doi.org/10.1016/j.eneco.2014.04.0020140-9883/© 2014 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 25 August 2013Received in revised form 24 March 2014Accepted 2 April 2014Available online 15 April 2014

Jel classification:Q51Q56Q40O14C53C14

Keywords:ChinaIndustryCarbon dioxide emissionsUndesirable output‘Green’ TFP growthReform

This paper uses a Directional Distance Function (DDF) and the Malmquist–Luenberger Productivity Index toestimate the changing patterns of ‘green’ total factor productivity (GTFP) growth of 38 Chinese industrial sectorsduring the period 1980–2010. Unlike the measures of traditional total factor productivity (TFP) growth, the DDFincorporates carbon dioxide emissions as an undesirable output directly into the production technology, whichcredit sectors for simultaneously reducing their emissions and increasing their output. Our estimates of aggregateand sector-level GTFP growth reveal that Chinese industry is not yet on the path towards sustainable, low-carbongrowth. A dynamic panel data analysis of the determinants of GTFP across sectors is used to identify factors thatmight rectify this situation, including state owned enterprise (SOE) reform, the growth of small private enterprises,continued openness to foreign investment and higher spending on R&D, particularly in emission-intensive sectors.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

China's economic model has delivered phenomenal rates of growthover the last three decades, resulting in the country's rise to the frontand centre of the global economic stage. However, that model has alsofavoured exports and investment over domestic consumption, capitalover labour, state-owned enterprises (SOEs) over the private sector,and the economy over the environment, culminating in an economydescribed by former Premier Wen Jiabao as ‘unstable, unbalanced,uncoordinated and ultimately unsustainable’.2

The imperative for a new model of growth is reflected in the WorldBank's (2012) China 2030 report, which proposes a ‘new developmentstrategy for China through to 2030’. This strategy highlights the needfor structural reforms that will strengthen the foundations of a market-based economy (including restructuring SOEs, encouraging the privatesector and reforming capital, land, labour and energy markets) and

86 21 6564 3056.e.golley@anu.edu.au (J. Golley).

and has repeated it often since2013.

accelerate innovation and technological progress. The strategy alsostresses the benefits of ‘green development’: ‘a pattern of developmentthat decouples growth from heavy dependence on resource use, emis-sions and environmental damage, and promotes growth through the cre-ation of new green products, technologies, investments, and changes inconsumption and conservation behavior’ (p. 233). The report makes itabundantly clear that solutions to China's environmental problems areinextricably linked to reforms that will rebalance the economy and setit on course for sustainable growth in the decades ahead, with many ofits recommendations being adopted at the Third Plenum of the CentralCommittee of the Chinese Communist Party in November 2013.

An appropriate measure of ‘sustainable growth’ is essential forassessing whether the Chinese government stands any chance ofsucceeding in their latest reform endeavour. Traditionally, a risingshare of total factor productivity (TFP) in output growth has beentaken as a signal of the transformation towards a ‘sustainable’ develop-ment model based on quality rather than quantity; that is, on intensiverather than extensive growth (Solow, 1957; Krugman, 1994; Young,1995). Using a range of different methods – including Solow residualsor regressions based on Cobb–Douglas (CD) or translog productionfunctions, parametric stochastic frontier production functions and dataenvelopment analysis (DEA) – ongoing debates revolve around whether

5 The approach can readily be applied to other emissions as well. For ease of reference,we use ‘emissions’ to refer to carbon dioxide emissions for the remainder of the paper.

6 The DEA method assumes that DMUs are homogeneous in terms of the nature of op-erations they perform, and the conditions underwhich they operate. At the industry level,this implies, for example, that each sector is operating under a similar market structure,

90 S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

and when China made this transformation.3 However, in most of thisresearch, TFP is calculated by using just capital and labour as inputsinto the production function, neglecting both the energy inputs requiredfor economic growth and their environmental impacts. This neglectdiminishes the relevance of this literature for assessing the true sustain-ability of China's evolving growth model, particularly given the clearpreference of the Chinese government for a ‘green, low-carbon model’in the future.

The directional distance function (DDF) introduced by Chung et al.(1997) addresses this issue by incorporating an emissions variable (orvariables) as an undesirable output directly into the production technolo-gy, with the underlying presumption being that ‘consumers have prefer-ences for reducing bad outputs regardless of the actual damage resultingfrom those outputs’ (Färe et al., 2001). This method simultaneouslycredits reductions in bads (emissions) and increases in goods in the esti-mation of a production frontier under the framework of data envelopeanalysis (DEA). TheMalmquist–Luenberger productivity index calculatedby using the estimatedDDF scores is ameasure of TFP knownvariously asactual, environmentally-sensitive, or green TFP (henceforth GTFP).

There have been wide applications of this approach outside China, in-cluding firm level, industrial level and cross-country analyses,4 but only afew on China to date. Hu et al. (2008) focus on carbon dioxide emissionsand use DDF to calculate the GTFP of Chinese provinces, which leads todifferent provincial rankings compared with the more traditional TFPmeasures of, say, Zheng and Hu (2005). Wang et al. (2010) estimate Chi-nese regional-level TFP and find that changes in GTFP and TFP deviatefrom each other, with GTFP suffering mainly from the emissions of sul-phur dioxide and chemical oxygen demand. Likewise, Zhang et al.(2011) also conduct a provincial-level analysis and, by using amore com-plete set of pollutants, demonstrate that GTFP growth is lower whenthese pollutants are incorporated as undesirable outputs in a DDF frame-work. The regional focus of these papers, however, is quite different fromthe sector-level analysis conducted here.

This paper uses directional distance function (DDF) andMalmquist–Luenberger productivity index to estimate the GTFP growth of 38Chinese industrial sectors between 1980 and 2010. Our results revealthat China's industrial GTFP growth estimated in thisway is significantlylower than traditional TFP growth estimates that credit a producer forexpanding the production of good outputs but do not consider the outputof bads, such as emissions. The most worrying sign is that GTFP growthin the last decade was not only low, but also lower than it was in thepreceding decade. An examination of the determinants of GTFP growth,and how these determinants differ from those of TFP growth, confirmsthat many of the reforms currently under consideration by the Chinesegovernment could place China on a truly sustainable path towards low-carbon growth in the future.

2. Measuring ‘sustainable low-carbon’ growth

DEA is a nonparametric linear programming method for estimatinga production frontier with multiple inputs and outputs, originating inthe pioneeringwork of Farrell (1957) and Charnes et al. (1978). A com-parisonwith the best-practice frontier enables the identification of eachinefficient decision-making unit (DMU) and its relative efficiency value,revealed by its distance from the frontier. There are numerous differentspecifications of emissions within the DEA framework, which rely ondifferent distance functions to calculate productivity indexes. Thispaper considers two alternative specifications: one in which carbon

3 See Chen et al. (2011) for a comprehensive survey.4 At the firm level see for example, Boyd et al. (2002) and Picazo-Tadeo et al. (2005); at

the industry level see Färe et al. (2001), Shestalova's (2003), Camioto et al. (2014) andOlanrewaju et al. (2012); and at the country level see Jeon and Sickles (2004), Kumar(2006) and a comprehensive survey by Zhou et al. (2008). In a more distantly-related pa-per, Fujii et al. (2010) use firm-level data to assess the energy efficiency of China's iron andsteel sector in the 1990s, during which time they find a continuous improvement inenvironmentally-sensitive productivities (i.e. GTFPs).

dioxide emissions (henceforth ‘emissions’) are ignored altogether(Model 1); and one that treats those emissions as an undesirable output(or bad) using the directional distance function (DDF) proposed byChung et al. (1997) (Model 2).5

Assume that there are n DMUs at time t, k types of input, l types ofdesirable output (or goods), and m types of undesirable output (orbads) for each DMU. For the ith DMU (i = 1, 2, …,n), the columnvectors xi, yi and bi represent the inputs, goods and bads, respectively.

Xkxn, Ylxn, Ct ¼ ∑3

i¼1Ci;t ¼ ∑

3

i¼1Ei;t � NCVi � CEFi � COFi � 44=12ð Þ and

Bm × n are the input and output matrices containing all of the DMUs.In this study, each DMU is one of the 38 Chinese industrial sectors withk=3, corresponding to capital, labour and energy and l=1, correspond-ing to output.6 In the case ofmodel 1,m=0,while in the case ofmodel 2,m= 1, correspond to the emissions associated with the energy input.

Fig. 1 illustrates the principle of the directional distance function(DDF) for our preferred model 2. Technology is represented by the out-put set P(x) towhich the output vector of point (y,b) belongs, where y isthe desirable output (goods) and b is the undesirable output (bads).Linear programming is used to calculate the value of the distance func-tion for each DMU at a fixed point in time (as detailed in Appendix 1).The DDF increases desirable output and simultaneously reduces undesir-able output for a given level of inputs, by scaling frompoint A in the direc-tion along AB, represented by the direction vector g= (y,−b).7 The keydifference between the two models therefore relates to their treatmentof emissions: model 1 excludes emissions and only credits a producerwith an increase in good outputs, while model 2 credits producers witha reduction in emissions and a simultaneous increase in good outputs.The weak disposability assumption of emissions in the DDF of model 2implies that the disposal of emissions is costly (i.e., requiring the diversionof inputs to achieve this end, or non-zero mitigation costs).

In model 1, TFP growth is estimated by computing the change in theMalmquist productivity index (MPI) between time t and t + 1, whereasin model 2, GTFP growth is estimated by computing the change in theMalmquist–Luenberger productivity index (MPLI) between time t andt + 1. Both MPI and MPLI can be decomposed into an efficiency changeindex and a technical progress index (see Appendix 1 for further de-tails). If there have been no changes in either inputs or outputs betweentwo points in time, then both productivity indexes for a given DMUwillequal one, while an improvement (deterioration) in productivity is sig-nalled by an index greater (less) than one. If a DMU hasmoved closer to(further away) from the production frontier between two points intime, then the efficiency change index will be greater (less) than one.The technical progress indexmeasures the shift in the production frontieritself. If technical change enables higher (lower) output and lower(higher) emissions then the index is greater (less) than one.

All of these indices can be converted to average annual growth ratesto provide a more familiar indication of the performance of Chinese in-dustry over time in terms of productivity growth, efficiency change andtechnical progress. Positive (negative) growth rates in all cases corre-spond to indexes greater (less) than one. This method also enables usto calculate the contribution of productivity growth to output growth,with the remainder stemming from growth of inputs. Consistent with

with similar access to technology and factor input supplies, and so on. We acknowledgethat this assumption is likely to be violated in our analysis here, as indeed it would be inregional level or cross-country analyses, and that this is a weakness in the empiricalanalysis.

7 Note that this is not the only specification of a DDF that can be used to treat emissionswithin the DEA framework: another possible specification is one that increases the goodwhile holding the bad constant (i.e., with the DDF heading vertically from point (y,b) tothe production frontier), as in Färe et al. (2007), Boyd et al. (2002) and Jeon and Sickles(2004).

Fig. 1. Principle of directional distance function.

Table 1Descriptive statistics of key variables.

Descriptive statistics Mean S.D. Minimum Maximum

IVA (RMB 100 million yuan) 634 1407 1 20,142Emissions (10,000 tons) 8027 26,009 16 305,967Capital stock (100 million yuan) 1018 1842 13 24,592Labour (10,000 workers) 243 227 10 1279Energy consumption (10,000 tce) 3093 9901 6 111,728SOEs output share (%) 38 30 0.3 100Small enterprises output share (%) 45 21 0.7 87HMTFEs output share (%) 25 19 0.01 84R&D intensity (%) 4 5 0.1 24

Source: Industrial dataset described in text and authors' calculations.

91S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

the traditional literature, a rising share of productivity growth isassumed to indicate a shift towards sustainable growth. Critically,when productivity is measured by GTFP, a rising share would signal ashift towards sustainable, low-carbon growth.

We use a panel dataset of annual data on 38 two-digit industrialsectors between 1980 and 2010, classified according to the 2002 ver-sion of the National Standard of Industrial Classification (GB/T4754)and developed by Chen (2011). The annual output of each sector isgiven by industrial value added (IVA) output, measured at 1990prices. The capital stock cannot be obtained directly and is estimatedby using the perpetual inventory approach, depreciated at constant1990 prices of investment in fixed assets. The other inputs are labour,given by the average number of employed workers, and energy, mea-sured in tons of coal equivalent (TCE).8 Total (direct) energy consump-tion is sourced fromcoal, petroleumandnatural gas, each ofwhichhas adifferent emission factor. Energy-induced emissions (measured in10,000 tons) are calculated by multiplying the quantity of each energysource by its emission factor and summing across all sources (seeAppendix 2 for further details).9 These are the variables required forour TFP and GTFP growth estimates.

The dataset also provides other variables of interest for our investi-gation into the determinants of GTFP growth below, namely the outputshares of SOEs, small enterprises (SEs) and enterprises funded by HongKong, Macau, Taiwan and foreign investors (HMTFEs) and the R&Dintensity (R&D expenditure as a share of output) of each sector. The de-scriptive statistics for all of these key variables are provided in Table 1,revealing the substantial heterogeneity across sectors in terms of size,energy use and emissions, industrial structure (ownership, enterprisesize and openness) and spending on innovation and technologicalcapacity.

8 In traditional measures of total factor productivity, value added is used as the outputand labour and capital as the inputs, without the consideration of intermediate inputs likeenergy. However, in analyses that take emissions into account, some researchers includeenergy as an input, and we choose to follow their lead. For examples, see Jeon and Sickles(2004), Watanabe and Tanaka (2007) and Wang et al. (2010).

9 This approach to measuring ‘energy-induced emissions’ only takes into account theemissions generated directly from the combustion of three primary fossil energy (coal, pe-troleumandnatural gas); that is, total (direct) energy consumption. It does not account forthe emissions generated indirectly via other inputs that themselves use energy in theirproduction processes (including electricity), nor those emissions generated via industrialprocesses such as the manufacturing of cement, lime, iron and steel. The aggregated sec-toral CO2 emissions estimatedbyus are very similar to China's fossil-fuel relatedCO2 emis-sions provided by IEA (http://www.iea.org/statistics/topics/CO2emissions/). See Golleyand XinMeng (2012) for further discussion of the various methods for measuringenergy-induced emissions in the Chinese context.

3. The GTFP growth of Chinese industry, 1980–2010

3.1. Aggregate results

Fig. 2 illustrates the aggregate trends in levels of energy consumptionand emissions, along with their intensities (with aggregate IVA in the de-nominator) between 1980 and 2010, as revealed by our dataset. Based onthe varying patterns of energy consumption and emissions revealed inthis figure, and following Chen et al. (2011), we divide the entire periodinto three sub-periods: the slow-and-steady period (1980–1995), the sta-ble period (1996–2002), and the rapid-growth period (2003–2010). Thefirst five columns of Table 2 reports the annual growth rates of outputand inputs averaged over the 38 sectors. Capital and energy are clearlythe dominant contributors to output growth in terms of inputs. Duringthe period 1996–2002, even though energy consumption and emissionsgrew by their lowest annual rate and labour growth was negative, IVAgrew at an annual rate of 12.6%. The combination of extremely rapidIVA growth and rapid energy consumption growth between 2003 and2010 explains the dramatic surge in aggregate emissions, despite the de-cline in emission intensity during this period (as shown in Fig. 2).

The final two columns of Table 2 report our estimated average annualgrowth rates of TFP and GTFP10 for Chinese industry between 1980 and2010, weighted by the IVA share for each sector at the sectoral dimensionfirstly and then averaged geometrically over the entire sample period andfor the three sub-periods at the time horizontal. Between 1980 and 2010TFP grew at an average annual rate of 5.5%, compared with GTFP growthof 1.8%. Indeed, by accounting for the emissions, model 2 results in GTFPgrowth estimates that are statistically significantly and substantiallybelow their traditional counterparts in all sub-periods.11

Table 2 also reveals significant differences between the contributionsof TFP and GTFP growth to output growth, over the entire period (at44% and 10% respectively) and in all sub-periods. TFP growth accountedfor 80% of output growth between 1996 and 2002, and one half of it be-tween 2003 and 2010, which would support the claim that growth hasbeen predominantly intensive since the mid-1990s, that is, ‘sustainable’in the traditional sense. However, GTFP growth accounted for only 31%

10 When estimating the inter-period DEA for the calculation of productivity index likeTFP/GTFP and one of their components, technical progress, the infeasible LP problemsmay occasionally happen because contemporaneous DEA allows for the possibility thatthe technology of previous periods may become infeasible in the following periods. Whileit is true that using 2-yearwindows or sequential DEA eliminate infeasible inter-period LPproblemswhen the technology of period t + 1 is used to evaluate an observation fromperiod t, sequential DEA does not guarantee the elimination of infeasible inter-periodLPmodelswhen the technologyof period t evaluates anobservation fromperiod t+1.How-ever, a modified specification of weak disposability has now been proposed by Färe et al.(2014) that eliminates both (1) downward sloping good output–bad output productionfrontiers and (2) infeasible LP problems for inter-period problems evenwhenusing contem-poraneous frontiers. We thank the reviewer for he/she clarifies the issue even if such infea-sible LP problems are not encountered in this paper.11 We ran twonon-parametric tests (theKolmogorov–Smirnov Z test andWilcoxon ranksum test) plus a paired-sample t test to test whether the estimated TFP and GTFP serieswere statistically different. In all three tests, we can reject the null hypothesis that the TFPand GTFP series are the same at an extremely high level of significance (with p-values lessthan 0.004).

Fig. 2. The aggregate trends in levels of energy consumption, CO2 emissions and value-added and the intensities of energy and emissions between 1980 and 2010.

Table 2Average annual growth rates of outputs, inputs and productivity (%).

IVA Emissions Capital Labour Energy TFP GTFP

Period (Model 1) (Model 2)

1980–2010 12.6 6.4 9.3 2.6 6.3 5.5 (44%) 1.8 (10%)1980–1995 8.5 5.9 9.8 3.7 5.7 1.4 (17%) 0.9 (10%)1996–2002 12.6 2.8 5.8 −2.9 3.0 10.1 (80%) 3.9 (31%)

92 S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

of output growth between1996 and2002, and just 10% in themost recentperiod: a far cry from the kind of sustainability we are interested in here.This decline is consistent with themajority view in the existing literatureusing traditional TFP measures (for example, Perkins and Rawski, 2008;Wu, 2008; Zheng et al., 2009; Chen et al., 2011).

Fig. 3 illustrates the change in aggregate industrial GTFP growth andits components over time. Through the 1980s, GTFP growth remainedlow, and was even negative at times, before experiencing a long andsteady increase through the 1990s to peak at 4.2% in 2002, followedby a continuous decline to reach just 0.5% in 2010. As the figure illus-trates, the rise and fall of China's GTFP growth have been driven bythe rise and fall in technical change, while efficiency has fluctuated atlow levels throughout the period of analysis.12

3.2. Sector-level results

Table 3 presents the average annual growth rates of TFP, and GTFP,efficiency change and technical change components for each sector be-tween 1980 and 2010. The combination of growth rates of IVA andemissions for each sector shown in the final two columns of the tableare useful for understanding the differences between the estimates ofTFP and GTFP growth. Each sector's productivity growth depends onboth the change in that sector's input–output combination over time,and the shape of the production frontier near that combination at theinitial point in time. As long as frontier production of the outputincreases when emissions increase (as we would typically expect), therelative values of TFP and GTFP growth depend not only on the growthof emissions between the two points in time, but also on this growthrelative to that of output (Jeon and Sickles, 2004). The standard andmost intuitive case is where GTFP growth is lower than TFP growthbecause emission growth is positive, and is therefore penalised in the‘green’ estimate. This occurred in coal mining, ferrous metal mining,and another 28 sectors. The other intuitive case, in which GTFP growthwas higher than TFP growth because emission growth was negative,occurred in just two sectors, Logging and ‘Others’.

12 It is coincidental for the trend of technical change to be opposite the trend of efficiencyduring most of the period because they are decomposed from the GTFP index, seeEqs. (5)–(8) in Appendix 1.

Another possibility is that even though emission growth was nega-tive, GTFP growth is still lower than TFP growth because the decline inemission growth is small relative to the increase in output growth. Thisoccurred in five sectors (non-ferrous metal mining, furniture, printing,general machines and measuring instruments). This leaves just oneperverse result, the petroleum extraction sector, in which emissiongrowth exceeded output growth and yet GTFP was greater than TFP(0.1% compared with −1.3%). As Jeon and Sickles (2004) explain, thismight be because the production frontier in the neighbourhood of thissector does not expand in the direction of more output even with an in-crease in emissions, or because of limited data in this neighbourhood –

where the latter is likely here given its high ranking in terms of emissionintensity (see Table 5 below) combined with its low rate of IVA growth.

Table 3 also presents the growth rates of efficiency change and tech-nical change for each sector between 1980 and 2010. As for the aggre-gate results illustrated in Fig. 3, technical progress was the dominantsource of GTFP growth in virtually all sectors over this period, peakingat an average annual rate of 6.3% in the electronics sector, and recordingpositive values in all but one sector (petroleum, coking and nuclear fuelprocessing, listed as ‘fuel processing’ in the tables). In contrast, efficiencychange was negative in 28 sectors and was only greater than 1% in twosectors (electronics and electric and heat power).

Table 4 shows the contribution of GTFP growth to output growthin each sector for the three sub-periods. Between 1980–95 and1996–2002, this contribution rose in 28 of the 38 sectors, accountingin the latter period for more than half of IVA growth in six sectors,and for more than 20% in another twelve sectors. However, between

2003–2010 21.1 10.2 11.0 4.6 10.1 10.6 (50%) 2.1 (10%)

Note: Percentage shares in brackets are contribution of TFP or GTFP to growth, which are cal-culated followingWu (2008) by dividing the TFP orGTFP growth rate by the IVAgrowth rate.Source: Industrial dataset described in the text and authors' calculations.

Fig. 3. The change in aggregate industrial GTFP growth and its components over time.

Table 3Growth of TFP, GTFP, output and emissions by sector (1980–2010) (annual percentchange).

Sector TFP GTFP E. change T. progress IVA Emissions

Coal mining 2.7 0.0 −0.1 0.1 8.7 5.5Petrol. extraction −1.3 0.1 −0.7 0.8 2.0 4.0Ferrous metal mining 4.1 0.7 −0.2 0.9 14.7 3.4Non–F. metal mining 3.7 0.8 −0.4 1.1 10.1 −0.2Non–metal mining 1.5 −0.1 −1.0 0.9 7.3 5.5Logging −0.4 0.1 −0.6 0.7 1.1 −2.5Food processing 1.7 0.3 −0.2 0.5 13.2 3.8Food manu. 3.2 0.3 −0.1 0.4 13.9 2.7Beverages 1.4 0.5 −0.2 0.8 13.3 3.1Tobacco 5.6 2.6 0.0 2.6 12.1 1.6Textiles 0.3 0.3 −0.4 0.7 9.1 1.7Apparel 5.9 1.7 −1.5 3.3 14.3 6.5Leather 6.6 2.1 −0.2 2.3 13.3 0.2Wood processing 5.5 1.0 −0.1 1.1 16.6 3.2Furniture 7.1 2.6 −0.1 2.7 13.6 −0.1Paper 2.2 0.1 −0.1 0.2 12.2 5.9Printing 9.1 2.7 −0.3 3.0 11.2 −1.6Cultural articles 8.4 3.7 −0.6 4.3 13.5 1.0Fuel processing −0.6 −2.4 −1.5 −0.9 2.6 6.7Chemicals 3.3 0.1 −0.1 0.1 11.3 4.1Medicine 4.2 2.5 0.0 2.5 16.5 3.6Fibres 8.9 0.4 0.0 0.4 13.9 1.8Rubber 1.0 0.4 −0.3 0.7 10.9 2.0Plastic 7.1 1.4 −0.7 2.1 15.4 6.2Nonmetal manu. 2.6 0.0 −0.1 0.1 11.1 6.2Ferrous smelt/press 3.1 0.0 −0.1 0.1 10.6 5.3Non–F. smelt/press 3.4 0.1 −0.3 0.4 12.8 6.8Metal products 6.6 1.4 −0.5 1.9 12.2 1.4General machines 9.1 1.7 0.0 1.7 12.6 −1.7Special machines 8.1 1.0 −0.2 1.2 12.2 0.6Transport equip. 9.7 1.5 0.1 1.4 17.9 1.2Electrical equip. 9.4 3.3 −0.2 3.5 15.9 3.4Electronic equip. 19.3 8.4 1.9 6.3 24.3 1.5Measuring inst. 11.2 5.4 0.2 5.2 14.0 −1.0Electric & heat power 7.5 1.5 1.3 0.2 9.9 8.6Gas 5.5 0.0 0.0 0.0 11.4 3.3Water 1.1 −0.1 −2.2 2.1 6.7 4.1Others 2.4 2.5 0.1 2.4 12.5 −1.0

Source: Industrial dataset described in text and authors' calculations.

Table 4GTFP growth and its share in output growth by sector (annual percent change).

1980–95 1996–2002 2003–10

Sector Growth Share Growth Share Growth Share

Coal mining 0.0 0.2 0.0 0.6 0.0 0.0Petrol. extraction 0.1 15.5 0.0 −1.4 0.0 0.2Ferrous metal mining 0.8 7.4 0.6 7.5 0.5 1.7Non–F. metal mining 0.5 5.8 1.6 24.8 0.7 4.2Non–metal mining −0.3 −4.2 0.0 −4.4 0.1 0.6Logging −0.1 −4.6 0.5 19.8 0.1 −9.0Food processing 0.2 2.1 0.4 3.6 0.3 1.5Food manu. 0.2 2.2 0.5 4.2 0.2 0.9Beverages 0.5 3.7 0.9 12.7 0.3 1.9Tobacco −1.5 −12.4 3.4 44.3 10.7 68.7Textiles 0.2 5.0 0.7 7.1 0.2 1.1Apparel 1.9 13.4 2.2 24.4 1.0 5.3Leather 2.1 16.8 2.9 31.0 1.5 8.2Wood processing 1.0 7.1 1.8 15.9 0.5 1.9Furniture 2.0 19.0 2.6 30.7 3.6 15.2Paper 0.1 1.0 0.2 1.3 0.0 0.2Printing 2.3 32.0 3.8 27.0 2.7 15.3Cultural articles 3.1 23.6 5.9 53.3 3.4 20.9Fuel processing −3.8 315.5 3.1 56.1 −3.4 −42.3Chemicals 0.1 0.9 0.1 0.6 0.0 0.2Medicine 0.6 4.8 7.0 34.8 3.0 14.6Fibres 0.2 1.6 0.3 4.4 0.6 3.9Rubber 0.3 4.7 0.7 6.9 0.3 1.4Plastic 1.1 7.4 3.8 29.8 0.2 1.2Nonmetal manu. 0.0 0.1 0.1 1.1 0.0 0.2Ferrous smelt/press 0.0 0.7 0.0 0.4 0.0 −0.1Non–F. smelt/press 0.0 0.1 0.2 1.5 0.1 0.4Metal products 1.3 13.2 2.3 24.9 1.1 5.2General machines 1.3 16.7 2.4 25.3 1.9 7.6Special machines 0.9 14.2 1.3 13.4 0.9 3.2Transport equip. 1.5 10.6 1.4 8.6 1.6 6.3Electrical equip. 2.5 20.8 6.1 36.2 2.8 12.2Electronic equip. 6.8 30.7 17.5 54.8 5.1 21.9Measuring inst. 2.6 25.3 8.5 64.4 8.8 39.0Electric & heat power 1.8 29.5 4.9 43.6 −1.7 −10.0Gas 0.0 0.5 0.0 0.0 0.0 0.1Water −0.1 −1.9 −0.7 50.4 0.4 3.3Others 0.9 6.9 6.8 109.3 2.6 15.2

Source: Industrial dataset described in text and authors' calculations.

93S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

94 S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

1996–2002 and 2003–2010, GTFP's contribution to output growthfell in 34 of the 38 sectors, only exceeding a half in one sector (tobacco)and 20% in just another three in the most recent period. This confirmsthat the setback in the low-carbon transformation of Chinese industry iswidespread, and not merely concentrated in a few problematic sectors.

3.3. Discussion

China's average annual growth of aggregate industrial output was12.6% between 1996 and 2002, a period inwhich the vastmajority of sec-tors recorded lower output growth but also higher GTFP growth than inthe period that followed. Indeed, aggregate output grew at a robust21.2% in 2003–2010, with virtually all sectors recording their highest out-put growth in this period but also reduced rates of GTFP growth. Do theseresults imply that rapid industrial growth can only be achieved by incur-ring high, and increasingly intolerable, environmental costs?

Not necessarily. The rise in aggregate GTFP growth between 1993and 2000 (shown in Fig. 3) overlaps significantlywith the ‘stable’ periodof energy and emission growth between 1996 and 2002. During thisperiod, growing inefficiencies within the state-owned sector of theeconomy (reflected in the declining efficiency index from 1992 on-wards) resulted in the deepening of ownership reforms. These includedthe policy of ‘grasping the large and letting go of the small’ (zhua da fangxiao), which saw the closure of 84,000 small energy- and emission-intensive enterprises, and the laying off of millions of excess workersfrom SOEs (xia gang) (contributing to the negative rate of labour forcegrowth in Table 3). Growing awareness of China's emerging environmen-tal problems also saw the commencement of a range of policies to devel-op renewable energy, with the share of coal – the most emission-intensive source of energy – in energy consumption falling from 74.7%in 1993 to 66% in 2000. Admittedly, growth of industrial output duringthis period was slower than in the period that followed. However, at anannual average rate of 12.6%, it would still be classified as rapid by anyother than Chinese standards. Crucially, it was also far less damaging interms of emission growth, hence the rising GTFP growth over this period.

In contrast, the results above confirm that China's industrial growthin the last decade was underpinned by rapid growth of capital andenergy consumption, with GTFP growth declining steadily as a result.Much of this investment and energy consumption was undertaken bylarge, inefficient state-owned enterprises (SOEs), which received prefer-ential access to subsidised capital, energy and other factor inputs.13 Fur-thermore, as noted by Wong (2013), SOEs have continued to be themain culprits when it comes to violating new environmental regulationsand restrictions, generating not only economic efficiencies but also sub-stantial environmental costs in the process. Thus while China's model ofrapid industrial growth in the last decade has indeed delivered high,and increasingly intolerable, environmental costs, it does not followthat allmodels of rapid industrial growthwill do the same. The regressionanalysis below builds on this point.

4. Determinants of China's industrial TFP and GTFP growth

The determinants of GTFP growth, and how these differ from those oftraditional TFP growth, are of fundamental importance for understandingwhy China's industrial growth model has resulted in such high emissiongrowth to date, andwhether anything can be done to facilitate the transi-tion towards sustainable, low-carbon growth in the future.

The traditional literature on the sources of Chinese productivity offersa number of candidates for consideration. For example, the productivityof SOEs has long been compared with that of other ownership types,with a recent paper by Song et al. (2011) finding that SOEs have lowerproductivity than private firms that use more productive technologiesbut suffer because of limited access to credit markets.14 TFP growth has

13 See Sheng and Zhao (2013) for a highly detailed analysis of this claim.14 See also Jefferson et al. (2008) and Jefferson and Su (2006).

also been found to be higher in light industries (Zheng et al., 2003),with smaller enterprises (LMEs) (Tu and Xiao, 2005), greater opennessto foreign direct investment (FDI) (Hong and Sun, 2011; Tuan et al.,2009) and higher spending on R&D (Zhou and Xia, 2010).

Beyond these sources of TFP, the other most obvious contender as adeterminant of GTFP is emission (or energy) intensity, which is in turnlikely to be linked to the traditional sources above.15 To motivate ourfinal choice of independent variables below, Table 5 ranks sectors inorder of their emission intensity (emission per 10,000 yuan of IVA) in2010, alongside a rangeof variables reflecting the traditional sources iden-tified above.While emission intensity is not the only factor constraining asector's GTFP growth potential, it certainly appears to be amajor one: thetop twelve sectors in terms of emission intensity are the bottom twelvesectors in terms of GTFP growth over the period 1980–2010.

As shown in the bottom row of Table 5, emission intensity is mosthighly correlated with the capital–labour ratio (10,000 yuan of capitalstock per worker), followed by its correlation with the share of SOEsin total output (driven by the top five most emission-intensive sectorsall having SOE shares above 40%). In contrast, correlations betweenemission intensity and small enterprises and HMTFEs are negative.Most strikingly, the correlation between emission intensity and smallenterprises is equal in magnitude and opposite in sign to that betweenemission intensity and SOEs.

One final point of interest in Table 5 is the high positive correlationbetween emission intensity and R&D intensity (of 0.36 in 2010), withthe top four most R&D intensive sectors all among the six mostemission-intensive sectors aswell. The even higher correlation betweenR&D intensity and the SOE share in each sector (of 0.42 in 2010) reflectsthe fact that much of China's R&D is undertaken by large SOEs, and inthe most emission-intensive sectors. These correlations raise some in-teresting possibilities about the various forces that might raise GTFPgrowth in the future.

To analyse these forcesmore comprehensively, we run the followingregression:

Gð ÞTFPi;t ¼ α þ β Gð ÞTFPi;t−1 þ γXi;t þ t þ t2 þ ui þ εi;t

where the dependent variable is either TFP or GTFP growth of sector ibetween time t and t + 1. To avoid the resulting error term reflectinga systematic pattern due to the influence of lagged productivity growthon current productivity growth, we include the lagged dependent vari-able as an independent variable. This enables us to extract the history ofthe other independent variables in the regression so that their inclusionrepresents the impact of new information. A time trend and its squareterm are used to capture the nonlinear change of productivity growthover time, as depicted in Fig. 3. Individual (sector) effects are capturedby ui and εi,t is the disturbance term. The matrix Xi,t contains those vari-ables most likely to impact on GTFP growth. In the initial regression,which covers the 1980–2010 period, this includes the capital–labourratio and emissions intensity. To assess the impacts of ownership, enter-prise size and openness on productivity growth, we extend the model toinclude the output shares of SOEs, small enterprises and HMTFEs respec-tively. We also include R&D intensity as a proxy for each sector's techno-logical capacity. Due to data limitations, this extended regression onlycovers the period 1993–2010.

Because productivity growth is constructed by using the capitalstock, labour and IVA, it is possible that all of these variables may beendogenously determined. To address this problem, we estimate adynamic panel data model using system GMM, as proposed byArellano and Bover (1995). System GMM assumes no autocorrelationin the disturbance terms and requires valid instrumental variables. For

15 Fisher-Vanden et al. (2006), for example, find that firm-level R&D expenditures, own-ership reform and changes in industrial structure are the primary factors underlyingChina's declining energy intensity over the period 1997–1999.

Table 5Key variables by sector (2010).

Sector Emission intensity (tons/10,000 yuan) K-L ratio (10,000 yuan/worker) SOEs (%) SmallEs (%) HKTMEs (%) R&D (%)

Total industry 6.3 9.3 27 38 27 6Fuel processing 381.8 37.1 71 11 13 16Electric & heat power 101.5 76.4 92 13 7 5Coal mining 28.8 9.5 56 30 4 16Petrol. extraction 27.7 53.0 95 6 7 24Gas 15.2 27.5 44 48 36 1Ferrous smelt/press 13.2 25.4 39 18 13 22Nonmetal manu. 9.0 6.4 10 63 14 3Chemicals 6.6 12.9 19 48 26 7Paper 4.8 10.0 8 50 30 4Non–F. smelt/press 2.3 16.1 28 40 14 9Non-metal mining 2.3 4.1 11 81 4 2Logging 1.8 4.4 99.6 66 0.2 0.4Fibres 1.7 16.0 9 25 32 7Food manu. 1.3 5.7 7 46 32 4Textiles 1.0 4.8 2 53 21 3Food processing 1.0 5.6 6 60 22 2Rubber 0.8 6.8 13 38 32 6Beverages 0.7 7.2 16 40 31 4Water 0.6 21.9 69 47 17 8Others 0.5 2.9 6 61 29 1Ferrous metal mining 0.5 9.2 14 58 2 3Wood processing 0.5 3.6 2 79 12 1Medicine 0.4 8.1 13 41 27 5Non–F. metal mining 0.4 7.3 27 54 4 4Plastic 0.3 4.3 3 69 31 3Special machines 0.3 5.3 22 47 25 9Metal products 0.2 4.3 5 63 25 3Apparel 0.2 1.7 1 57 38 1Transport equip. 0.2 7.9 47 20 44 10Leather 0.2 1.7 0.3 46 45 1General machines 0.2 5.4 13 56 23 6Electrical equip. 0.1 4.8 9 37 31 8Printing 0.1 5.2 12 68 23 1Furniture 0.1 2.9 3 59 33 2Tobacco 0.1 24.9 99 2 0.1 1Cultural articles 0.1 2.0 1 57 52 5Measuring inst. 0.03 4.0 10 39 48 5Electronic equip. 0.02 6.6 8 12 77 4Correlation with EI: 0.49 0.37 −0.37 −0.19 0.36

Source: Industrial dataset described in text and authors' calculations.

95S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

all specifications presented in Table 6, we run the Arellano–Bond testfor first and second order autocorrelation and confirm that there isno autocorrelation in the disturbance term. The Sargan test for over-identifying restrictions confirms that all the instrumental variables for

Table 6The determinants of TFP and GTFP growth.

Regressions Regression 1 (1980–2010)

Dependent variables TFP GTFP

Coef. se Coef.

Constant −0.31* 0.1193 −1.46***(G)TFP_lag1 0.79*** 0.1827 0.04***Capital per capita −0.04 0.0228 −0.01**CO2 emission intensity 0.005* 0.0024 −0.003**Time trend 0.55*** 0.0973 0.38*Time trend square −0.01*** 0.0025 −0.01***Share of SOEs outputShare of SmallEs outputShare of HMTFEs outputShare of R&D to GDP

Diagnostic tests Statistic p value Statistic

AR(1) test −1.39 0.1394 −1.54AR(2) test 0.56 0.4026 0.52Sargan test 44.82 1.0000 36.48Wald test 33,817 0.0000 6093

Number of observations 1102 1102

Note: ***,**,* indicate that the levels of significance are 1%, 5% and 10%, respectively. Source: In

the system GMM method are exogenous. The Wald chi-square testconfirms the overall significance of each regression specification.

The results for the baseline regression are presented in Columns 1and 2. The positive and significant coefficient on the lagged value of

Regression 2 (1993–2010)

TFP GTFP

se Coef. se Coef. se

0.1264 −1.24*** 0.4537 −2.27*** 0.25350.0124 0.67*** 0.1501 0.12** 0.04460.0041 −0.07 0.0749 −0.02*** 0.00430.0014 −0.003 0.0126 −0.001** 0.00060.2288 0.11* 0.0612 0.26* 0.15940.0043 −0.057*** 0.0021 −0.06*** 0.0066

0.03 0.0443 −0.03* 0.0190−0.02 0.0342 0.04*** 0.0033−0.01 0.0136 0.12** 0.0567

0.14*** 0.0543 0.18*** 0.0754

p value Statistic p value Statistic p value

0.1233 −1.08 0.2784 −1.45 0.14590.5996 −1.56 0.1193 −1.22 0.22341.0000 33.08 1.0000 33.43 1.00000.0000 172,000 0.0000 191,063 0.0000

646 646

dustrial dataset described in text and authors' calculations.

96 S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

TFP and GTFP growth indicates that historical information has strongexplanatory power for current levels of productivity growth. The coeffi-cients on the time trend and its square confirm the inverted U-shapedpattern of GTFP (and TFP) growth shown in Fig. 3. For the remaining co-efficients there are significant differences between the determinants ofTFP and GTFP growth. Capital intensity does not impact significantly onTFP, while it has a negative and significant impact on GTFP, such that arise of 10,000 yuan of capital per worker decreases GTFP growth by0.01 percentage points.16 Emission intensity, on the other hand, ispositively associated with TFP growth but negatively associated withGTFP growth, such that a reduction of 1 ton of emissions for every10,000 yuan of output will increase GTFP growth by 0.003 percentagepoints. To put this in perspective, the coal-mining sector reduced itsemission intensity by 55 tons of emissions for every 10,000 yuan be-tween 1993 and 2010. For a sector that averaged GTFP growth of just0.01% during the last three decades (as shown in Table 3), this impliesthat emission-intensity reductions could deliver significant increasesin GTFP growth, particularly in high emission-intensity sectors.

The results for the complete specification of the regression arereported in columns 3 and 4 of Table 6. For TFP growth, the inclusionof additional variablesmeans that emission intensity is no longer signif-icant, possibly being replaced by the high significance on R&D intensity(noting the correlation above) – indeed, R&D intensity is the only signif-icant determinant of TFP growth other than lagged TFP growth, the timetrend and its square. The ratio of R&D expenditure to output is the onlyother variable that has a similar (and large) impact on both TFP andGTFP growth, confirming that technological capacity impacts positivelyof productivity growth, whether the measurement of that growthembodies emission reductions or not.

The shares of SOEs, small enterprises and HMTFEs in each sector'soutput are all statistically significant determinants of GTFP growth,and in the predicted directions. In particular, a 1-percentage point risein the output share of SOEs translates into a fall in GTFP growth of0.03 percentage points, significant at the 10% level. This indicates thatquite substantial increases in GTFP growth might be achieved vialarge-scale privatisation in sectors dominated by SOEs, especially in thefourmost highly emission-intensive sectors, all of which have SOE sharesof output above 50% (see Table 5). Higher shares of small enterprises andHMTFEs are associatedwith higherGTFP growth,with a rise of 1 percent-age point increasing GTFP growth by 0.04 and 0.12 percentage pointsrespectively.

In sum, these results indicate that sectors dominated by capital-intensive, energy-hungry, large SOEs have suffered from lower rates ofGTFP growth, compared to those sectors with high shares of small(often private) and foreign enterprises. This not only helps to explainwhy China's industrial GTFP growth has been on a downward trend inthe last decade, but also points to some possible mechanisms for in-creasing it in the future.

5. Conclusions

This paper used DEA analysis and the DDF to estimate the ‘green’TFP growth of 38 Chinese industrial sectors between 1980 and 2010.In aggregate, and in virtually all sectors, these growth estimates weresubstantially below their traditional TFP growth counterparts, andtheir contributions to output growth were well below the 50% marktaken to signal a successful transition to a sustainable low-carbon

16 While the capital–labour ratio is often found to be a significant determinant of TFP inthe traditional growth literature, this is not always the case for China. This relates toChina's distinctive reformand development process, which has involved a shift away frominefficient heavy industries towards light industries (i.e., thosewith low capital–labour ra-tios), while markets for both capital and labour have remained imperfectly competitive(with, for example, limited capital mobility even across provinces, and restrictions onthe movement of labour from rural-to-urban areas). These features suggest that the im-pact of changing capital–labour ratios on Chinese productivity growth is likely from thoseelsewhere.

model of economic growth. The fact that in the majority of sectorsboth GTFP growth rates and their share of total output growth havefallen in the last decade is particularly worrying.

Our regression analysis identified a number of variables with strongexplanatory power for the low rates of GTFP growth observed in manyof China's industrial sectors. In particular, GTPF growth was lower inmore emission-intensive sectors with higher capital–labour ratios, andin sectors with larger shares of SOEs, and smaller shares of small enter-prises and enterprises funded outside of Mainland China. R&D intensitywas the one factor that impacted in a highly significant and positivewayon both GTFP and TFP growth during the period under study.

Looking to the future, these results can be interpreted with cautiousoptimism. In particular, the World Bank's (2012) China 2030 reportoutlines the mix of market incentives, regulations, public invest-ments, industrial policy and institutional developments that Chinaplans to use in order to seize the opportunity to ‘go green’, includingthe promotion of emerging green industries, including clean energymarkets (solar-, wind-, and hydro-power), biotechnology, high-endmanufactures and clean-energy vehicles, and the use of the bestenergy-saving technologies currently available. These specifically‘green’ measures, coupled with the report's broader recommendationsto reduce the role of the state sector and encourage private entrepre-neurship and innovation, could result in the significant expansion notonly of China's own production frontier, but of the global frontier aswell, enabling technical progress to become a strong driver of produc-tivity growth again in the future.

This does not imply, however, that the path ahead is straightfor-ward. In particular, the central government in Beijing presides over aneconomic system that they officially describe as ‘socialismwith Chinesecharacteristics’, in which a dominant role for state ownership is a keyprinciple. It is highly unlikely that full-scale privatisation will be under-taken in the foreseeable future, particularly in ‘strategic’ sectors of theeconomy,which include themost capital- and emission-intensive energyand mining sectors. Furthermore, it is not a foregone conclusion thatsmall, private Chinese firms will be willing and able to undertake thelevel of R&D required to reduce emissions and boost ‘green’ productivityin the future: in a Schumpeterian world of creative destruction, large in-novative (and not necessarily private) firms would do amuch better job.A deeper understanding of the complex nexus between state ownership,firm size, innovation and emission reductions will be critical for under-standing whether China's reform agenda can possibly set the countryon the path towards sustainable low-carbon development in the decadeahead.

Acknowledgements

The authors thank the Editor, Beng W. Ang and Richard S.J. Tol, andtwo reviewers for their constructive comments. This work is supportedby the National Natural Science Foundation (71173048), National SocialScience Foundation (12AZD047), Ministry of Education (11JJD790007),Shanghai Leading Talent Project and Fudan Zhuo-Shi Talent Plan.

Appendix A

Appendix 1. Linear programming and the Malmquist and Malmquist–Luenberger productivity indexes

In model 1, the distance function of the basic DEA for the ith DMU attime t is constructed using the following linear programming (LP):

Dt0 xt

i ; yti

� �� �−1 ¼ Maxλ;δi δis:t: Yλ≥δiyi; Xλ≤xi; λ≥0

ð1Þ

where D represents the distance function and the scalar δi, whichsatisfies δi ≥ 1, is the efficiency score of the ith DMU based on

97S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

observations and technology at time t. The scalar vector λ of order(n × 1) are the weights with which the inefficient DMUs are mappedonto the production frontier. The inequality in the constraints for boththe desirable output (y) and inputs (x) are based on the assumptionthat they are freely disposable.

In model 2, in order to credit sectors for reductions in bads as a wellas expansion of goods, the directional distance function, D

!, for the i th

DMUusing technology and input–output observations in t time is calcu-lated by using the following LP:

D!t

o xti ; y

ti ;b

ti ; y

ti ;−bt

i

� �¼ Maxλ;β β

s:t: Yλ≥ 1þ βð Þyi; Bλ ¼ 1−βð Þbi; Xλ≤xi; λ≥0ð2Þ

in which β is the maximum feasible expansion of the good output andcontraction of the bad output (b) in identical proportions for a givenlevel of inputs, which amounts to the value of the DDF to be measured.The inequality in the constraints for goods and inputsmakes them freelydisposable as inModel 1, while the equality constraint on the badmakesit weakly disposable, meaning that disposing of bad outputs is eithercostly or restricted (see Färe and Grosskopf, 2004).

The distance function in model 1 and model 2 is calculated con-temporaneously; that is, each sector is compared to all other sectorsat one point in time and the frontier for each time point envelops theobservations from this period only. To ensure that productivitygrowth, technical progress and efficiency change for each sector areinter-temporally comparable over the sample period, adjacent- ratherthan base-period productivity indexes are used in the calculations (i.e.t to t + 1 in the equations below).

Specifically, the traditional TFP used inModel 1 is obtained by calcu-lating the following Malmquist productivity index (MPI):

MPIt;tþ1 ¼Dto xtþ1

i ; ytþ1i

� �

Dto xt

i ; yti

� � �Dtþ1o xtþ1

i ; ytþ1i

� �

Dtþ1o xt

i ; yti

� �0@

1A

1=2

¼Dtþ1o xtþ1

i ; ytþ1i

� �

Dto xt

i ; yti

� �Dto xtþ1

i ; ytþ1i

� �

Dtþ1o xtþ1

i ; ytþ1i

� � �Dto xt

i ; yti

� �

Dtþ1o xt

i ; yti

� �0@

1A

1=2 ð3Þ

where the first termmeasures the change in relative efficiency betweentime t and t + 1 (MECH), and the second term captures the shift intechnology between the two periods (MTCH). That is to say, Malmquistproductivity index is the geometric mean of efficiency and technicalchange as:

MPIt;tþ1 ¼ MECHt;tþ1 �MTCHt;tþ1 ð4Þ

GTFP is estimated by calculating theMalmquist–Luenberger Produc-tivity Index (MLPI) as below:

MLPIt;tþ1¼1þ D

!to xt

i ; yti ;b

ti ; y

ti ;−bt

i

� �

1þ D!to xtþ1

i ; ytþ1i ;btþ1

i ; ytþ1i ;−btþ1

i

� ��1þ D

!tþ1o xt

i ; yti ;b

ti ; y

ti ;−bt

i

� �

1þ D!tþ1o xtþ1

i ; ytþ1i ;btþ1

i ;ytþ1i ;−btþ1

i

� �24

351=2

ð5Þ

The MLPI can be decomposed as the product of two terms: thechange of technical progress (MLTCH) and the change of productionefficiency (MLECH); that is

MLPIt;tþ1 ¼ MLECHt;tþ1 �MLTCHt;tþ1 ð6Þ

where,

MLECHt;tþ1 ¼1þ D

!to xt

i ; yti ;b

ti ; y

ti ;−bt

i

� �

1þ D!tþ1

o xtþ1i ; ytþ1

i ;btþ1i ; ytþ1

i ;−btþ1i

� � ð7Þ

MLTCHt;tþ1 ¼1þ D

!tþ1o xtþ1

i ;ytþ1i ;btþ1

i ;ytþ1i ;−btþ1

i

� �

1þ D!to xtþ1

i ; ytþ1i ;btþ1

i ; ytþ1i ;−btþ1

i

� � �1þ D

!tþ1o xt

i ; yti ;b

ti ;y

ti ;−bt

i

� �

1þ D!to xt

i ;yti ;b

ti ; y

ti ;−bt

i

� �0@

1A

1=2

ð8Þ

Appendix 2. Calculating energy-induced CO2 emissions

According to the World Bank definition, CO2 emissions are thosestemming from the burning of fossil fuels and the process of industrialproduction such as manufacturing of cement, lime, iron and steel,metal and so on; the former of which accounts for at least 70% of totalCO2 emission worldwide, and more than 85% in China because of thedominance of coal in its fossil fuel consumption. Due to data limitations,the CO2 emissions calculated in this paper only relate to fossil fuel com-bustion; that is to say, those emissions generated in the production ofoutput stem from the consumption of three types of primary energy:coal, petroleum and natural gas. This is calculated using the followingexpression for each industrial sector:

CO2 ¼X3

i¼1

CO2;i ¼X3

i¼1

Ei � NCVi � CEFi � COFi � 44=12ð Þ

where CO2 represents the flow of carbon dioxide in units of ten thou-sand tons, i = 1,2,3, correspond to three types of primary energy(coal, petroleum and natural gas), which are each used in quantity Emeasured in 10,000 tons (coal and petroleum) or 100million cubic me-tres (natural gas). In order to avoid double counting, the secondarysource of electricity is not included in our estimates of direct energy-induced emissions. NCV is the net calorific value provided by theChina Energy Statistical Yearbook in 2007, CEF is the carbon emissionfactor provided by the 2006 National Greenhouse Gas Inventories inIntergovernmental Panel on Climate Change (IPCC, 2006), and COF isthe carbon oxidization factor set to be one for both petroleumandnaturalgas and 0.99 for coal in this study. The values of 44 and 12 are themolec-ular weights of carbon dioxide and carbon, respectively. The calculatedCO2 emission coefficients for China are 1.973 (Kg CO2/Kg) for coal,3.065 (Kg CO2/Kg) for petroleum and 2.184 (Kg CO2/m3) for natural gas.

References

Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation oferror-components models. J. Econ. 68, 29–51.

Boyd, G.A., Tolley, G., Pang, J., 2002. Plant level productivity, efficiency, and environmentalperformance of the container glass industry. Environ. Resour. Econ. 23, 29–43.

Camioto, Flávia de Castro, Mariano, Enzo Barberio, Aparecida do Nascimento Rebelatto,Daisy, 2014. Efficiency in Brazil's industrial sectors in terms of energy and sustainabledevelopment. Environmental Science & Policy 37, 50–60.

Charnes, A., Cooper, W., Rhodes, E., 1978. Measuring the efficiency of decision makingunits. Eur. J. Oper. Res. 2, 429–444.

Chen, Shiyi, 2011. Estimates of sub-industrial statistical data in China (1980–2008). ChinaEcon. Q. 10 (3), 735–776 (Jing-Ji-XueJi-Kan).

Chen, S., Jefferson, G.H., Zhang, J., 2011. Structural change, productivity growth and industrialtransformation in China. China Econ. Rev. 22 (1), 133–150.

Chung, Y.H., Färe, R., Grosskopf, S., 1997. Productivity and undesirable outputs: a direc-tional distance function approach. J. Environ. Manag. 51, 229–240.

Färe, R., Grosskopf, S., 2004. Modeling undesirable factors in efficiency evaluation:comment. Eur. J. Oper. Res. 157, 242–245.

Färe, R., Grosskopf, S., Pasurka Jr., C.A., 2001. Accounting for air pollution emissions inmeasures of state manufacturing productivity growth. J. Reg. Sci. 41 (3), 381–409.

Färe, Rolf, Grosskopf, Shawna, Pasurka, Carl, 2007. Pollution abatement activities and tra-ditional productivity. Ecol. Econ. 62 (3–4), 673–682 (15 May).

Färe, Rolf, Grosskopf, Shawna, Pasurka, Carl, 2014. “Potential grains from trading bad out-puts: the case of U.S. electric power plants”. Resour. Energy Econ. 36 (1), 99–112(January).

Farrell, M., 1957. The measurement of productive efficiency. J. R. Stat. Soc. Ser. A 120 (3),253–281 (General).

Fisher-Vanden, Karen, Jefferson, Gary, Jingkui, Ma, Jianyi, Xu, 2006. Technology develop-ment and energy productivity in China. Energy Econ. 28, 690–705.

Fujii, Hidemichi, Kaneko, Shinji, Managi, Shunseke, 2010. Changes in environmentallysensitive productivity and technological modernization in China's iron and steelindustry in the 1990s. Environ. Dev. Econ. 15, 485–504.

Golley, Jane, Meng, Xin, 2012. Income inequality and carbon dioxide emissions: the caseof Chinese urban households. Energy Econ. 34, 1864–1872.

98 S. Chen, J. Golley / Energy Economics 44 (2014) 89–98

Hong, Eunsuk, Sun, Laixiang, 2011. Foreign direct investment and total factor productivityin China: a spatial dynamic panel analysis. Oxf. Bull. Econ. Stat. 73 (6), 771–791.

Hu, A., Zheng, J., Gao, Y., Zhang, N., Xu, H., 2008. Provincial technology efficiencyranking with environment factors (1999–2005). China Econ. Q. 7 (3), 933–960(Jing-Ji-XueJi-Kan).

Jefferson, G.H., Su, J., 2006. Privatization and restructuring in China: evidence from share-holding ownership, 1995–2001. J. Comp. Econ. 34 (1), 146–166.

Jefferson, G.H., Rawski, T.G., Zhang, Y., 2008. Productivity growth and convergence acrossChina's industrial economy. J. Chin. Econ. Bus. Stud. 6 (2), 121–140.

Jeon, B.M., Sickles, R.C., 2004. The role of environmental factors in growth accounting. J.Appl. Econ. 19 (5), 567–591.

Krugman, Paul, 1994. The myth of Asia's miracle. Foreign Aff. 73 (6), 62–78.Kumar, Surender, 2006. Environmentally sensitive productivity growth: a global analysis

using Malmquist–Luenberger index. Ecol. Econ. 56 (2), 280–293.Olanrewaju, O.A., Jimoh, A.A., Kholopane, P.A., 2012. Integrated IDA–ANN–DEA for assess-

ment and optimization of energy consumption in industrial sectors. Energy 46 (1),629–635.

Perkins, D.H., Rawski, T.G., 2008. Forecasting China's economic growth over the next twodecades. Chapter 20. In: Brandt, Loren, Rawski, Thomas G. (Eds.), China's Great Eco-nomic Transformation. Cambridge University Press, Cambridge and New York.

Picazo-Tadeo, A.J., Reig-Martínez, E., Hernández-Sancho, F., 2005. Directional distancefunctions and environmental regulation. Resour. Energy Econ. 27, 131–142.

Sheng, Hong, Zhao, Nong (Eds.), 2013. China's State-owned Enterprises: Nature, Perfor-mance and Reform. World Scientific, Singapore.

Shestalova, Victoria, 2003. Sequential Malmquist indices of productivity growth: an appli-cation to OECD industrial activities. J. Prod. Anal. 19, 211–226.

Solow, R.M., 1957. Technical change and the aggregate production function. Rev. Econ.Stat. 39, 312–320.

Song, Zheng, Storesletten, Kjetil, Zilibotti, Fabrizio, 2011. Growing like China. Am. Econ.Rev. 101, 196–233.

Tu, Z., Xiao, G., 2005. China's industrial productivity revolution: a stochastic frontierproduction function analysis of TFP growth in China's large and medium industrialenterprises. Econ. Res. J. 3, 4–15 (Jing-Ji Yan-jiu).

Tuan, C., Ng, L.F.Y., Zhao, B., 2009. China's post-economic reform growth: the role of FDIand productivity progress. J. Asian Econ. 20 (3), 280–293.

Wang, B., Wu, Y., Yan, P., 2010. Environmental efficiency and environmental total factor pro-ductivity growth in China's regional economies. Econ. Res. J. 5, 95–109 (Jing-Ji Yan-Jiu).

Watanabe, Michio, Tanaka, Katsuya, 2007. Efficiency analysis of Chinese industry: a direc-tional distance function approach. Energ Policy 35 (12), 6323–6331.

Wong, Edward, 2013. ‘Aspollutionworsens, solutions succumb to infighting’, 21March 2013,The New York Times, online at: http://www.nytimes.com/2013/03/22/world/asia/as-chinas-environmental-woes-worsen-infighting-emerges-as-biggest-obstacle.html.

Wu, Yanrui, 2008. The role of productivity in China's growth: new estimates. China Econ.Q. 7 (3), 827–842 (Jing-Ji-XueJi-Kan).

Young, A., 1995. The tyranny of numbers: confronting the statistical realities of the EastAsian growth experience. Q. J. Econ. 110, 641–680.

Zhang, Chunhong, Liu, Haiying, Bressers, Hans, Buchanan, Karen, 2011. Productivitygrowth and environmental regulations – accounting for undesirable outputs: analysisof China's thirty provincial regions using the Malmquist–Luenberger Index. Ecol.Econ. 70, 2369–2379.

Zheng, J., Hu, A., 2005. An empirical analysis of provincial productivity in China(1979–2001). China Econ. Q. 4 (2), 263–296 (Jing-Ji-XueJi-Kan).

Zheng, J., Liu, X., Bigsten, A., 2003. Efficiency, technical progress, and best practice inChinese state enterprises (1980–1994). J. Comp. Econ. 31, 134–152.

Zheng, J., Bigsten, A., Hu, A., 2009. Can China's growth be sustained? A productivity per-spective. World Dev. 37 (4), 874–888.

Zhou, Liqun, Xia, Liangke, 2010. How R&D investments influence TFP growth: evidencefrom China's large and medium-sized industrial enterprises. Front. Econ. China 5(4), 537–558.

Zhou, P., Ang, B.W., Poh, K.L., 2008. A survey of data envelope analysis in energy andenvironmental studies. Eur. J. Oper. Res. 189, 1–18.

Chen, Shiyi: Professor of Economics at Fudan University, and Research Fellow at ChinaCenter for Economic Studies (CCES), Fudan Development Institute (FDDI) and Fudan Tyn-dall Centre. He is the Director of Leading Group of Ecology, Environment, Humanities andSocial Sciences Research at Fudan University, Director of Policy Lab for Sustainable Devel-opment at Fudan University, and Co-Director of Shanghai-Hong Kong Development Insti-tute (CUHK-Fudan). His research interests are in economic transformation anddevelopment in China, energy, environment and sustainability, applied econometricsand so on. His currentworks are published in English journals such asQuantitative Finance,Journal of Forecasting, The World Economy, China Economic Review, Review of DevelopmentEconomics, Economic Systems and top Chinese journals. His new English monograph, enti-tled Energy, Environment and Economic Transformation in China, was published byRoutledge Taylor & Francis Group in 2013.

Golley, Jane: She has recently joined the Australian Centre on China in theWorld as an As-sociate Director. She has a DPhil and MPhil in Economics from Oxford University, and aFirst Class Honours degree in Economics from The Australian National University. She be-gan her career in the Asia Section of the Australian Commonwealth Treasury, before un-dertaking postgraduate research on Chinese regional development and teaching inOxford for 7 years. She returned to the School of Economics at The Australian NationalUniversity in 2003 and thenmoved to the China Economy Program at the Crawford Schoolof Economics and Government in 2008. Jane has published a book on Chinese regional de-velopment and articles on Chinese industrial agglomeration and regional policy; Chinesedemographic change, economic growth and the real exchange rate; Chinese householdconsumption, energy requirements and carbon emissions; and cross-country comparisonsof trade openness and growth.