What is Driving China’s Decline in Energy Intensity?*

Karen Fisher-Vanden**

Dartmouth College

Gary H. Jefferson Brandeis University

LIU Hongmei

National Bureau of Statistics

TAO Quan National Bureau of Statistics

April 10, 2002

Abstract In this paper, we employ a unique data set of approximately 2,500 large and medium-sized industrial enterprises in China for the years 1997-1999 to identify the factors driving the fall in total energy use and energy intensity. Using an econometric approach that identifies sources of variation in energy intensity, we find that changing energy prices and research and development expenditures are significant drivers of declining energy intensity. Changes in ownership, region, and industry composition are less important. By identifying the contribution of each of these five factors, we reduce the size of the contribution of sectoral shift and identify specific sources of productivity gain. The association between differences in relative energy prices and measured energy intensities indicates that Chinese firms are responding to prices—something not largely observed in the past. In addition, the impact of R&D spending on energy intensity suggests that firms are using resources for energy saving innovations. JEL codes: Q4, P2 * We would like to thank Su Jian for excellent research assistance. We also thank Messers Xu Jianyi , Ma Jingkui, and Liu Fujiang for making the Dartmouth-Brandeis -NBS collaboration possible, and Jim Feyrer, Mun Ho, Adam Jaffe and participants at the Rockefeller Center’s faculty seminar (February 2002) for helpful comments. This research was supported by the Rockefeller Center at Dartmouth College, the U.S. Department of Energy’s Biological and Environmental Research Program (contract # DE-FG02-00ER63030), and the National Science Foundation (project/grant #450823). **Corresponding author: Dartmouth College, 6182 Steele Hall, Hanover, NH 03755. Phone: 603-646-0213, fax: 603-646-1682, email: kfv@dartmouth.edu

2

1. Introduction

A significant proportion of the robust economic growth that China has enjoyed over the

past 20 years has accrued from the growth of total factor productivity. China’s rising levels of

TFP have, in part, been driven by rising energy productivity. In 1990, however, as we can see

from Figure 1, China’s overall energy efficiency lagged behind that of other countries of similar

levels of per capita income. By 1998, the use of similar data indicates that China had

substantially closed the gap between its predicted and actual level of energy use

This movement toward greater energy efficiency is reflected in China’s energy intensity

over time. As shown in Figure 2, since 1978, China’s energy intensity (defined as the ratio of

real energy consumption to real GDP) has fallen dramatically. Although China’s energy

intensity has fallen since 1978, primary energy consumption and production have generally risen.

As shown in Figure 3, beginning in 1996, this trend of rising consumption and production seems

to have reversed itself. Between 1996 and 2000, total primary energy consumption fell by eight

percent, driven mainly by a 17.4 percent decline in coal consumption. This downward trend in

energy consumption has tracked a similar trend in primary energy production. During 1996-

2000, primary energy production has fell by 17.8 percent, again driven by a 26.6 percent decline

in coal production. Based on trends from 1978 to 1996, the persistence of reported high levels of

economic growth, and increases in household demand for automobiles and modern appliances,

the fall in energy production and use is contrary to what most have forecast.

Figure 1

y = -6E-07x2 + 0.2895x + 29.738

100

1000

10000

100 1000 10000 100000

1990 GDP per capita (US$, PPP, log scale)

1990

en

erg

y u

se p

er c

apit

a (k

g o

f oil

equ

iv, l

og

sca

le)

China

3

Figure 2Energy-Output Ratios, 1978-2000

02468

1012141618

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

10,0

00 t

on

s S

CE

/100

mill

ion

19

78 Y

uan

Source: China Statistical Yearbook, 2001 (NBS, 2001).

Figure 3Primary energy consumption

0

20000

40000

60000

80000

100000

120000

140000

1978

1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

10,0

00 t

on

s S

CE

HydroN. gas

OilCoal

Source: China Statistical Yearbook, 2001 (NBS, 2001).

Explanations for China’s decline in energy use and intensity fall into three categories.

The first is gains in energy efficiency: at the firm level, energy productivity could be improving,

thereby lowering the amount of energy required to produce a unit of output. The second possible

explanation centers on structural change: sectoral output shares could be changing, so that the

share of the more energy intensive sectors in total output could be declining. Finally,

inaccuracies in China’s energy statistics could be underreporting the amount of energy

consumed, and/or overestimating the growth of output. Although previous work has also

attempted to measure the relative contribution of each of these factors, we have the advantage of

4

a data set, which includes industry detail at the 3 and 4-digit levels. In addition, we use a large

sample of data to estimate a range of contributions – prices, R&D expenditure, ownership

change, and other factors – to changes in energy productivity at the firm level. This level of

detail has not been previously possible.

A number of studies have attempted to measure the relative contributions of sectoral shift

and sub-sector productivity change. Using decomposition methods, Garbaccio, Ho and

Jorgenson (1999), Lin and Polenske (1995), and Sinton and Levine (1994) each find sub-sector

productivity change to be the principal contributor to the fall in China’s energy-output ratio.1

A shortcoming of these studies is the aggregate nature of the data used. Most of the

studies use industrial data, which are aggregated at the 2-digit industry level. As acknowledged

in previous studies and confirmed by our analysis, this level of aggregation can overstate the

contribution of sub-sector energy productivity improvements and understate the role of sectoral

shift. A second shortcoming of prior research is the inability of these studies to identify factors

that drive sub-sector energy productivity improvements. Such factors include changes in energy

prices, differences in energy efficiency across ownership types, and differences in R&D intensity

that may have important policy implications. A third drawback of prior studies is that the data

cover broad, heterogeneous segments of the Chinese economy. While these studies have the

advantage of comprehensive coverage, by including data collected from the non- industrial

sectors and smaller firms, these studies also run a greater risk of inaccurate data. Lastly, these

previous studies do not cover the period 1996-1999, during which we observe not only a

continuation of the decline in energy intensity, but also a fall in total energy consumption.

In this paper we employ a unique data set of approximately 2,500 large and medium-

sized Chinese industrial enterprises for the years 1997-1999. We use the data to identify the

factors that are driving the fall in total energy use and energy intensity, while hopefully resolving

the issues faced in previous studies. Similar to these previous studies, we find that

improvements in energy productivity play a larger role than sectoral shift in explaining the

decline in energy intensity. The magnitude of this contribution varies significantly, depending

on the level of industry aggregation used. In explaining improvements in firm-level energy

productivity, we find that in some measure shifts in output across ownership type, region, and

1 Garbaccio, Ho, and Jorgenson compare 1987 and 1995; Lin and Polenske compare 1981 and 1987, and Sinton and Levine compare different years between 1980 and 1990.

5

industry all contribute efficiency improvements. The largest contributions, however, originate

from energy prices and research and development activity within the firm. Price changes

account for the largest efficiency gains. This result suggests that Chinese firms are reacting to

changing energy prices—something not observed in the past.

We conduct an analysis of the level of price variation in energy markets across region,

ownership type, and industry. That prices tend to be the highest in the eastern coastal provinces

suggests that energy prices in this area have most fully adapted to world prices. If energy prices

in other regions of China rose to match those in coastal areas, then presumably the country could

achieve even further efficiency gains.

The paper is organized as follows. Section 2 describes the data set we use; in this

section, we place our sample within the context of China’s overall industry and all large and

medium enterprises. Section 3 compares trends in consumption and efficiency by energy type

between our sample, total industry, and the Chinese economy as a whole ; we discuss the possible

explanations for these trends. Section 4 applies the Divisia decomposition method to our sample

to measure the relative contribution of sectoral shift and productivity change. Section 5 provides

a closer examination of the improvements in firm-level energy productivity by conducting an

econometric analysis to identify the factors driving these improvements. Section 6 examines the

market efficiency of energy prices (a factor shown to be important in explaining firm-level

energy productivity improvements) by conducting an econometric analysis to test the importance

of region, ownership and industry differences on energy prices. Lastly, we present the

conclusions of our research in Section 7.

6

2. Data Description

The analyses presented in this paper were conducted using a dataset spanning

approximately 2,500 large and medium-size Chinese industrial enterprises for the years 1997-

1999. The energy data are combined with two other data sets of data that cover all of China’s

approximately 22,000 large and medium-size enterprises (LMEs) and are updated annually by

the National Bureau of Statistics (NBS) in China. The first is an economic and financial data set,

collected by the Bureau’s Department of Industrial and Transportation Statistics [NBS, 2001b];

the second data set, which consists of science and technology measures, including innovation

inputs and outputs, is maintained by the Bureau’s Department of Population, Social, and Science

and Technology Statistics [NBS, 2001c]. These two data sets include all of China’s

approximately 22,000 large and medium-size enterprises, far more than the 3,700-4,700

enterprises that have been covered by the energy survey in recent years.

The number of enterprises covered in the energy survey range from a low of 3,746

enterprises in 1998 to 4,687 enterprises in 1999. A part of this variation reflects changes in

capacity utilization and energy consumption over the business cycle. A total of 2,582 enterprises

appear in all three years. Table 1 compares levels of sales, employment, and fixed assets in our

sample with total industry and the full sample of total large and medium-size enterprises.

Table 1 Shares of LMEs and energy sample in aggregate industry

(% of total industry)

Measure 1999 All industry1 Of which: L&M

Enterprises2 Of which: energy

sample 3

Sales (100 million yuan) 69,851 (100%) 41,166 (59%) 13,159 (19%) Employment (10,000 persons) 4,428 (100%) 3,061 (69%) 1,031 (23%) Assets4 (100 million yuan) 71,847 (100%) 53,070 (74%) 21,810 (30%) No. of enterprises 162,033 (100%) 22,000 (14%) 2,582 (1.6%) 1 Industrial state owned and non-state owned enterprises with annual sales over 5 million Yuan. Source: China Statistical Yearbook, 2000 [NBS, 2000]. 2 Source: NBS (2001b). 3 Represents the largest number of enterprises reporting in each of three years. Source: NBS (2001b). 4 Original value fixed assets

7

The energy dataset includes 21 energy types and a measure of aggregate energy

consumption (Table 2). These data, collected annua lly by the NBS, are reported by the most

energy intensive enterprises among the population of large and medium-size enterprises. In this

paper, we concentrate on energy types that comprise a large share of total energy use and are

consumed for combustion purposes. Three energy categories (in addition to aggregate energy)

were created from the 21 energy types provided in the data: coal and coal products (01-06),

refined oil (12-17), and electricity (20). Crude oil (11) was not included in this classification

since it is almost entirely consumed in the refined oil sector. Coal gases (07-09) and natural gas

(10) were not included due to their small relative share of total energy and the presence of

relatively few observations in the data set. Due to the non-combustion functions, other

petroleum products (18), steam heat (19), and other energy types (21) were not included.

The NBS data set classifies enterprises into 37 industrial categories. For the purposes of

this analysis, we group the 37 industrial classifications into12 industry categories. Table 3 shows

the distribution of the enterprises in our sample over these 12 industrial sectors. Not

surprisingly, relative to the distribution of total industry and LMEs, the energy sample includes a

higher incidence of enterprises in the more energy- intensive industries. The chemical and

electric power industries are among those that are over-represented relative to the total industry

and LME populations.

8

Table 2

Energy Classifications

01 Unprocessed coal 02 Cleaned coal 03 Other coal 04 Coke 05 Other coking products 06 Briquettes 07 Coke oven gas 08 Smelting iron gas 09 Other coal gas 10 Natural gas 11 Crude oil 12 Gasoline 13 Kerosene 14 Diesel oil 15 Fuel oil 16 Liquefied petroleum gas 17 Refinery gas 18 Other petroleum products 19 Steam heat 20 Electric power 21 Other energy types 22 Aggregate energy

The NBS data set also classifies enterprises into the seven ownership classifications

shown in Table 4. As shown in this table, our sample is largely concentrated in the state-owned

sector, which reflects the fact that energy-intensive firms are largely state-owned. Lastly, the

NBS dataset includes a firm’s geographic location. Our six regional groupings are shown in

Table 5. The firms in our sample are largely concentrated in the north (e.g., Beijing), east (e.g.,

Shanghai), and south (e.g., Guangzhou).

9

Table 3 Industry distribution, 1999 [%]

Industry classification

(2-digit SIC) Total industry1 LMEs Energy sample only2

Mining (06-10,12) 7,257 [4%] 829 [4%] 220 [9%] Food and Beverage (13-16) 20,125 [12%] 2,593 [11%] 239 [9%] Textile, apparel, and leather products (17-19) 20,784 [13%] 2,637 [12%] 169 [7%]

Timber, furniture, and paper products (20-24) 12,374 [8%] 1,332 [6%] 130 [5%]

Petroleum processing and coking (25) 988 [1%] 120 [1%] 59 [2%]

Chemicals (26-28) 15,412 [10%] 2,760 [12%] 501 [19%] Rubber and plastic products (29-30) 7,852 [5%] 893 [4%] 57 [2%]

Non-metal products (31) 14,366 [9%] 1,699 [8%] 395 [15%] Metal processing and products (32-34) 13,644 [8%] 1,429 [6%] 188 [7%]

Machinery, equipment, and instruments (35-37,39-42) 29,955 [18%] 6,287 [28%] 270 [10%]

Electric power (44) 4,941 [3%] 1,039 [5%] 286 [11%] Other industry (43,45,46) 14,335 [9%] 971 [4%] 68 [3%] Total 162,033 [100%] 22,589 [100%] 2,582 [100%]

1 Includes all state and non-state enterprises with annual sales above 5 million yuan. Source: NBS (2000). 2 Represents the largest number of enterprises reporting in each of three years.

Table 4 Ownership distribution, 1999 [%]

Ownership type Total industry1

LMEs Energy sample only2

State-owned 61,301 [38%] 10,451 [46%] 1,795 [70%] Collective-owned 42,585 [26%] 3,381 [15%] 111 [4%] Hong-Kong, Macao, Taiwan 15,783 [10%] 1,567 [7%] 91 [4%] Foreign 11,054 [7%] 1,966 [9%] 88 [3%] Shareholding 4,480 [3%] 4120 [18%] 468 [18%] Private 316 [1%] 6 [0%] Other domestic

26,830 [17%] 792 [4%] 23 [1%]

Total 162,033 [100%] 22,111 [100%] 2,582 [100%] 1 Includes all state and non-state enterprises with annual sales above 5 million yuan. 2 Represents the largest number of enterprises reporting in each of three years.

10

Table 5 Regional distribution, 1999 [%]

Ownership type Total industry1 LMEs Energy sample

only2

Huabei—North 22,593 [14%] 753 [4%] 432 [17%] Dongbei—Northeast 11,565 [7%] 1,795 [9%] 341 [13%] Huadong—East 65,082 [40%] 9,151 [47%] 715 [28%] Huanan—South 44,186 [27%] 5,237 [27%] 697 [27%] Xinan—Southwest 11,094 [7%] 1,764 [9%] 174 [7%] Xibei—Northwest 7,515 [5%] 973 [5%] 223 [9%] Total 162,033 [100.0] 19,673 [100%] 2,582 [100%]

1 Includes all state and non-state enterprises with annual sales above 5 million yuan. 2 Represents the largest number of enterprises reporting in each of three years.

11

3. Energy Consumption and Intensity, 1997-1999

How do the consumption and intensity figures in our survey data compare with the

national trends shown above in Figures 2 and 3? Tables 6a-e compare consumption and intensity

figures for total energy, coal, oil and electricity in our sample with total industry and economy-

wide figures from the national statistics. The trends in our sample mirror those of the total

economy and industrial sector—namely, consumption levels of total energy, coal, and oil have

fallen since 1996 while electricity has risen. In addition, the ratios of consumption of total

energy, coal, oil, and electricity to output have all decreased, implying a fall in the intensity of

each energy type.

Table 6a Overall energy consumption and efficiency

Year Consumption

(10,000 tons of standard coal (SCE))

Intensity

Total1 Of which: industry1

(share of total)

Of which: sample

(share of industry)

Total (tons/

constant GDP)

Of which: industry

(tons/ constant GVIO)

Of which: sample (tons/

constant GVIO)

1980 60,275 38,986 (65%) - 14.34 19.82 - 1985 76,682 51,068 (67%) - 10.97 16.20 - 1990 98,703 67,578 (68%) - 9.67 13.79 - 1995 131,176 96,191 (73%) - 7.29 8.70 - 1996 138,948 100,322 (72%) - 7.05 8.06 - 1997 137,798 100,080 (73%) 56,009 (56%) 6.42 7.23 6.51 1998 132,214 94,409 (77%) 54,646 (58%) 5.71 6.26 6.19 1999 130,119 90,797 (70%) 56,475 (62%) 5.25 5.55 6.00

1 NBS (2000,2001a).

12

Table 6b Energy consumption and efficiency: coal and coal products

year Consumption

(10,000 tons) Intensity

Total1 Of which: industy1

(share of total)

Of which: energy sample

(share of industry)

Total (tons/

constant GDP)

Of which: industy (tons/

constant GVIO)

Of which: sample (tons/

constant GVIO)

1985 81,603 58,613 (72%) - 11.67 18.59 - 1990 105,523 81,090 (77%) - 10.34 16.55 - 1995 137,676 117,571 (85%) - 7.65 10.63 - 1997 139,248 121,671 (87%) 44,966 (37%) 6.49 11.38 5.23 1998 129,492 114,952 (89%) 42,960 (37%) 5.60 9.86 4.87 1999 126,365 112,757 (89%) 41,634 (37%) 5.10 9.28 4.42

1 NBS[2000, 2001a]

Table 6c Energy consumption and efficiency: Refined Oil

year Consumption

(10,000 tons) Intensity

Total1 Of which: industy1

(share of total)

Of which: sample

(share of industry)

Total (tons/

constant GDP)

Of which: industy (tons/

constant GVIO)

Of which: sample (tons/

constant GVIO)

1985 9,169 6,171 (67%) - 1.31 1.96 - 1990 11,485 7,322 (64%) - 1.13 1.49 - 1995 16,064 9,349 (58%) - 0.89 0.85 - 1997 19,691 11,304 (57%) 1,364 (12%) 0.92 1.06 0.16 1998 19,817 10,870 (55%) 1,250 (11%) 0.86 0.93 0.14 1999 21,072 10,852 (51%) 1,172 (11%) 0.85 0.89 0.12

1 NBS[2000, 2001a]

13

Table 6d Energy consumption and efficiency: electricity

year Consumption

(100M KWh) Intensity

Total1 Of which: industy1

Of which: sample

Total Of which: industy

Of which: sample

1985 4,117 3,283 (80%) - 0.59 1.04 - 1990 6,230 4,873 (78%) - 0.61 0.99 - 1995 10,023 7,659 (76%) - 0.56 0.69 - 1997 11,284 8,395 (74%) 1,943 (23%) 0.53 0.79 0.23 1998 11,598 8,406 (72%) 1,845 (22%) 0.50 0.72 0.21 1999 12,305 8,832 (72%) 1,911 (22%) 0.50 0.73 0.20

1 NBS[2000, 2001a]

Explanations for the declines in aggregate energy consumption and intensity fall into

three general categories: (1) sectoral shift; (2) subsector energy productivity change; and (3)

inaccurate statistics. We examine each of these candidate explanations below.

(1) Sectoral shift

The sectoral shift approach measures the implications of industry share in total output for

overall energy intensity. For example, China’s gradual move away from heavy industry, which

is generally energy intensive, has contributed to China’s declining energy intensity. Over the

past 20 years, market reforms have affected China’s industry mix through multiple channels. In

particular, the opening to international trade and the phasing out of state-set prices and state-

directed investment have altered the relative profitability of certain industries – in particular,

heavy industry, which had most enjoyed trade protection and government subsidies in the past.

Rising incomes and changing relative prices have also spurred changes in patterns of consumer

demand, notably increases in a wide range of consumer and electronic goods.

Most previous studies have found sectoral shift to be a relatively small contributor to the

decline in China’s energy intensity [e.g., Garbaccio, Ho and Jorgenson, 1999; Lin and Polenske,

1995; and Sinton and Levine, 1994]. A World Bank study (1997), however, identifies sectoral

shift as the largest contributor. As suggested by Garbaccio, Ho and Jorgenson (1999) and

14

discussed later in this paper, this mismatch may result from the use of different levels of industry

aggregation.

(2) Subsector productivity change

We measure subsector energy productivity change as change in the ratio of energy use to

real output within an industry. For instance, Hicks-neutral technical advance within an industry

would result in lower energy intensity.

A number of factors can affect firm-level energy efficiency. The elimination of state-set

prices, for example, has led to rising relative energy prices, thereby inducing energy saving

innovations. In addition, enterprise reform has increased the incentives of managers to utilize

resources to implement cost-saving innovations [Jefferson, Bai, Guan, and Yu, 2002]. Sinton

and Levine [1994] also attribute improvements in subsector energy productivity to conservation

policies imposed by the Chinese central government. Government policies are targeting

inefficient production, including closing down small generators in the power sector so as to

improve efficiency in electricity generation and the closing of small, inefficient mines that slows

the growth in coal production [Sinton and Fridley, 2000].

(3) Inaccurate statistics

As shown in the introduction, aggregate energy intensity in China has fallen dramatically

since the introduction of market reforms in 1978. Additionally, aggregate energy consumption

has fallen since 1996. Numerous observers argue that these results are based on reported data

that are inaccurate. Some question whether the energy data may be underreported. On the

consumption side, when household purchases of air conditioners and modern appliances have

risen so rapidly, the possibility that household energy consumption could be falling seems

implausible.

On the production side, Sinton and Fridley (2000) believe that energy statistics may be

understated as a result of the omission of coal produced by small coal mines that have been

officially shutdown and imported fuels, such as diesel, that have officially been banned. As

elaborated in Sinton [2001], notwithstanding the order from the central government in 1999 to

shutdown small coalmines, these mines have continued their operation. In addition, to protect

15

domestic fuel producers that were being undercut by cheaper imported fuel, the Chinese central

government ordered a ban on imported fuel (mainly diesel) in 1998. This has led to illegal

smuggling of diesel fuel that does not show up in the import statistics. In both cases, this

undercounting affects the reported production of fuels and not consumption.

Others argue that China’s statistical authorities have over reported output. Rawski (2001)

and Sinton and Fridley (2000) have both suggested that measures of lower energy intensity may

result from inflated GDP rather than real gains in energy efficiency. While many observers take

two percent as an appropriate downward adjustment figure (See Sinton and Fridley, 2000),

Rawski (2001) believes this number should be closer to 9 percent. As discussed in Garbaccio,

Ho and Jorgenson (1999) and Sinton and Levine (1994), falsification of output figures is more

prominent among non-state and rural firms.

Garbaccio, Ho and Jorgenson (1999) and Sinton and Levine (1994) raise the issue of

deverticalization as another potential reporting problem that could overstate the decline in energy

intensity. Deverticalization occurs when firms outsource certain parts of their production

process that were previously performed in-house. Deverticalization raises an industry’s value of

output (due to double counting), without increasing the level of energy consumed. The

consequence of deverticalization is that energy intensity measured in terms of energy use per unit

of GNP or value added should be unaffected, but when the denominator of the energy intensity

measure is measured in terms of the gross output of firms or sectors, deverticalization will lead

to the appearance of declining energy intensity.

Our sample avoids many of the reporting problems raised regarding the aggregate data.

Output numbers are also more reliable in these survey data because they come from large

established firms, which tend to provide more reliable numbers than smaller firms. Since we are

concentrating on energy consumption figures in these survey data, the problems with production

estimates are less relevant. These firms tend to buy from large established coalmines, and

therefore avoid the problem of underreporting purchases from small illegal mines.

16

4. Sectoral shift vs. subsector productivity change

As shown in the previous section, energy use in our sample follows a similar decline to

that seen in both the Chinese economy as a whole and total industry. We use our sample to

decompose the contributions of sectoral shift (i.e., changes in the sectoral composition of total

industrial output) and subsector productivity change (i.e., changes in the energy intensity within

an industrial sector). We then compare our results with previous decomposition studies of China

that have been conducted for years prior to 1995 [e.g., Garbaccio, Ho, and Jorgenson, 1999; Lin

and Polenske, 1995; and Sinton and Levine, 1994].

Relative to these studies, our decomposition has two advantages. First, we are able to

examine the years after 1996 when China experienced a dramatic fall in energy use and intensity.

In addition, these studies apply decomposition methods to a 2-digit industrial classification,

which, due to the fact that each 2-digit sector is an aggregation of many separate industries, can

underestimate the role of sectoral shift in the decline of energy intensity. Although previous

studies typically use industry classifications at the 2-digit aggregation level, Sinton and Levine

(1994) present results from a highly disaggregated industry classification (i.e., 267 subsectors),

but are unable to adequately compare the results with their other datasets due to differences in

coverage and the energy variable used (i.e., end-use energy versus gross energy consumption).

Our sample includes 4-digit industrial classification detail, which allows us to substantially avoid

the aggregation problem faced in previous studies.

We apply to our sample the following multiplicative form of the Divisia decomposition

method:2

2 A derivation of this equation can be found in Ang and Zhang [2000].

17

residual. and t;at timeoutput totalof share si'sector s t;at time isector ofintensity energy total

t;at time (E/Q)intensity energy total t;at time isector in energy totalGVIO);(constant t at timeoutput total

t;at timeenergy total

(1) )ln()()ln()()ln()ln()ln(

ti

100

0

21

100

0

21

00

≡≡≡≡≡≡≡

∑ ++∑ ++++===

R

IIEQEwhere

Rss

EE

EE

II

EE

EE

QQ

EE

ti

t

ti

t

t

N

i i

ti

t

tii

N

i i

ti

t

tii

tt

The first two terms, which can be combined as )ln(0

0tQ

QE

, represent energy use at time t,

conditional on output at time t being produced at the same energy intensity as in period 0. The

third term captures changes in total sample energy intensity between periods 0 and t due to

changes in industrial subsector energy productivity. The fourth term represents changes in total

sample energy intensity due to changes in the sectoral composition of total industrial output (i.e.,

sectoral shift). Since this decomposition method is an approximation, the last term captures the

residual change.

We apply the Divisa method to aggregate energy use in our sample for the period 1997-

1999. In Table 7, we report five sets of results using different levels of industry aggregation.

These are: (a) the 1-digit industry classification, (b) our 12- industry aggregation, (c) the 2-digit

industry classification, (d) the 3-digit industry classification, and (e) the 4-digit industry

classification. Figure 4 shows graphically the percent contribution of intensity change versus

sectoral shift. These results show that the level of industry classification matters significantly

when measuring the relative contributions of sectoral shift and subsector productivity change.

Using all of these industry classifications, we find that changes in subsector energy intensity

account for the largest share of the overall decline in energy intensity. However, as we move

from the 1-digit to the 12- industry or 2-digit classification, the sectoral shift contribution rises

substantially. While we find a further increase in the sectoral shift contribution as we move from

the 2-digit to the 3-digit classification, our results show no further advantage in incorporating the

additional sectoral detail associated with the 4-digit classification.

18

Table 7 Divisia Decomposition Results, 1997-1999

Total Energy

Terms in equation (1)

(a) 1-digit

(b) 12-industry

(c) 2-digit

(d) 3-digit

(e) 4-digit

First term 10.9333 10.9333 10.9333 10.9333 10.9333 Second term 0.0912 0.0912 0.0912 0.0912 0.0912

Third term (subsector energy

productivity -0.1205 -0.0607 -0.0585 -0.0498 -0.0512

Fourth term (sectoral shift)

0.0376 -0.0222 -0.0244 -0.0340 -0.0335

Sum 10.9416 10.9416 10.9416 10.9407 10.9398 Compare 10.9416 10.9416 10.9416 10.9416 10.9416 Residual 0.0000 0.0000 0.0000 0.0008 0.0018

Figure 4Intensity change vs. Sectoral Shift

-100%

-50%

0%

50%

100%

150%

200%

1-digit 12-industry 2-digit 3-digit 4-digit

Aggregation

% C

on

trib

utio

n

Intensity changeSectoral Shift

Using the 3-digit industry classification, Table 8 provides decomposition results for three

types of energy – coal, refined oil, and electricity. While subsector productivity change emerges

as the dominant factor driving declines in oil and electricity intensities, sectoral shift plays a

more equal role in the decline of coal intensity.

19

Table 8 Divisia Decomposition Results, 1997-1999

By Energy Type

Terms in equation (1)

Coal Refined Oil Electricity

First term 10.7137 7.2182 7.5722 Second term 0.0912 0.0912 0.0912 Third term -0.0947 -0.2004 -0.1019

Fourth term -0.0723 -0.0521 0.0065 Sum 10.6378 7.0570 7.5681

Compare 10.6367 7.0659 7.5559 Residual -0.0011 0.0089 -0.0122

% Contribution Subsector

productivity change 57% 79% 107%

Sectoral shift 43% 21% -7%

20

5. Factors affecting firm-level energy productivity

Although the magnitude of the contribution of intensity change depends on the sectoral

aggregation used and type of energy examined, the results from the Divisia exercise tells us that

during the critical period 1997-1999, subsector productivity change contributed more to the

decline in energy intensity than sectoral shift. Given this result, we are interested in identifying

the factors driving the decline in subsector energy intensity.

Among the factors that may be contributing to rising firm-level energy efficiency are the

following:

(1) Price reform: With the initiation of China’s “two-tiered pricing system” in 1984, prices set

by the central plan have been slowly replaced with prices set by the market. By 1990, most

goods had substantial outside plan supplies that were priced by the market, causing firms to

face market prices for factor inputs at the margin [Byrd, 1989, 1991]. In 1990,

approximately 46 percent of coal was plan allocated, leading to a ratio of market-plan price

of 2.70; for crude oil, the figures were 80 percent and 5.92 [Garbaccio, 1995]. By 1999, plan

allocations had been virtually eliminated. As we will see, by 1999, there were no significant

differences in prices faced by state-owned enterprises and foreign- invested firms. These

rising prices for energy inputs, we expect, have induced firms to seek ways to improve

energy efficiency.

(2) Ownership reform: A significant component of the China’s market reforms have been a

gradual clarification and decentralization of property rights within the enterprise sector.

Enterprise reform in China has included both the reform of managerial control rights within

firms that have maintained their SOE designation and a rapid expansion of various ownership

classifications outside the state sector, both through new entry and through the conversion of

SOEs particularly to domestic and foreign joint ventures and to shareholding companies.

[e.g. Li, 1997; Jefferson and Rawski, 1994; Jefferson, Rawski, and Zheng, 2000] .

(3) Deliberate innovation: Many firms maintain formal research and development operations

whose scale can be measured by R&D expenditures or by R&D personnel. Whether these

innovation activities lead to product or process innovation, they may have measurable affects

21

on energy intensity. Jefferson, Bai, Guan, and Yu (2002) find that R&D inputs – both

personnel and expenditures – generate returns that exceed the returns to production workers

and investment in physical capital. With rising relative energy prices, we should expect that

R&D would at least have a neutral effect in promoting gains in energy efficiency and may

even have an energy saving effect.

(4) Changing industry composition: We anticipate that different industries use technologies that

exhibit different levels of energy intensity. By their nature, electric power generation and

petroleum processing and coking are likely to be significantly more energy intensive than the

food processing or apparel industries. For the same reason that we used the Divisa approach

to identify the contribution of changing industry composition to energy intensity, we wish to

include this factor in our regression analysis.

(5) Changing regional composition: Across regions there may exist significant regional

variation associated with different supply-demand relationships that may not be equalized

through low-cost transportation choices. Also, provincial policies relating to pricing or

regulation may vary across regions and result in differences in levels of energy intensity

across regions. Change in the regional distribution of output may therefore lead to changes

in overall energy intensity.

In order to investigate the sources of subsector productivity gains in energy consumption,

we constructed a balanced sample of 2,582 enterprises that we used in the construction of Tables

6a-d. With 2,582 firms and three years, the sample included 7,746 observations. Because our

regression includes variables for prices, R&D expenditure, ownership, and location, as well as

industry classification, the number of observations for which data are available for all of the

relevant variables shrinks to a total of 2,818 observations over the 3-year period. Using our

limited regression sample, we first want to test how closely this sample tracks the changes in

energy intensity shown for our full sample in Tables 6a-d. We do this by regressing the log of

the energy/GVIO ratio on the 1998 and 1999 time dummies. The results in Table 9 show both

for the full sample and the regression subsample similar patterns of decline in intensity. For each

22

of the four energy categories, intensities decline by significant amounts in both 1997-98 and

1998-99.

Table 9 % Change in Energy/GVIO

Coal Ref. Oil Electricity Total Energy

Regression subsample: 1997-1998 -10.1% -14.7% -3.8% -5.6% 1997-1999 -18.0% -27.0% -8.4% -9.4%

Obs. = 2,818 Full sample:

1997-1998 -6.9% -10.7% -7.5% -5.0% 1997-1999 -15.5% -21.6% -10.2% -8.0%

Obs. = 7,746 Because firm-level energy intensities in the regression subsample follow the trends identified in

the full sample, we proceed to use the subsample to identify the contribution to declining energy

intensity of each of the factors identified above.

The estimation equation used in this analysis was derived from profit maximization,

assuming the following translog production function:

lnQ = α0 + ∑= MELKX

X X,,,

lnα + αRlnRDE + γ1T97 + γ2T98 + γ3T99 + ∑=

J

jj

P INSTj

1

α

+ ∑∑∑===

++MELKX

XT

MELKX

XT

MELKX

XT XTXTXT,,,

99,,,

98,,,

97 lnlnln 999897 βββ

+ ∑ ∑= =

⋅MELKX

J

j

XP XINSTj

,,, 1

lnβ + ∑=

⋅MELKX

RDEX,,,

lnln

+ βRT1lnRDE⋅T97 + βRT2lnRDE⋅T98 + βRT3lnRDE⋅T99 + lnRDE∑=

J

jj

RPjINST1

β

where

Q ≡ quantity of output; K ≡ quantity of capital input; L ≡ quantity of labor input; E ≡ quantity of energy input; M ≡ quantity of material input; RDE ≡ quantity of R&D activities;

23

T97, T98, T99 ≡ time dummies; and INSTi ≡ institutional variables such as ownership type, region, industry.

This formulation implies that institutional factors (such as ownership type, industry, and

region-specific factors), prices, time, and level of R&D activities each enter into the input

demand equations. Given this, we derive the following first-order condition with respect to

energy:

PEE / PQQ = αE + βERlnRDE + βET1T97 + βET2T98 + βET3T99 + ∑=

J

jj

EPjINST1

β

where PE ≡ price of energy and PQ ≡ price of output. This condition implies the following

relationship: E/Q = f(Prices, RDE, time, institutional variables).

In our estimation, we use the three specific aggregations described earlier - coal, refined

oil, and electricity - and estimate the following system as a system of seemingly unrelated

regressions (SUR):

∑∑∑

∑∑∑

∑∑∑

===

===

===

++++

++++++=

++++

++++++=

++++

++++++=

7

1

12

1

6

1997986

9754321

7

1

12

1

6

1997986

9754321

7

1

12

1

6

1997986

9754321

)ln()ln()ln()ln()ln( )3(

)ln()ln()ln()ln()ln( )2(

)ln()ln()ln()ln()ln( )1(

kikk

jijj

rirr

iiiioi

i

kikk

jijj

rirr

iiiioi

i

kikk

jijj

rirr

iiiioi

i

OWNINDREGIONTT

TRDEPelePoilPcoalGVIOQele

OWNINDREGIONTT

TRDEPelePoilPcoalGVIOQoil

OWNINDREGIONTT

TRDEPelePoilPcoalGVIOQcoal

γγβαα

αααααα

γγβαα

αααααα

γγβαα

αααααα

where

Qcoali ≡ quantity of coal purchased by firm i; GVIOi ≡ gross value of industrial output of firm I in constant Yuan; Pcoali ≡ average price of coal paid by firm i; Poili ≡ average price of refined oil paid by firm i; Pelei ≡ average price of electricity paid by firm i;

24

RDEi ≡ research and development expenditures by firm i;3 T97, T98, T99 ≡ dummy variables associated with years 1997-1999; REGIONir ≡ dummy variables associated with 6 regions (refer to table 5 for list of regions); INDij ≡ dummy variables associated with 12 industry categories (refer to table 3 for list of

industries); and OWN ik ≡ dummy variables associated with 7 ownership types (refer to table 4 for list of

ownership types).

In each set of estimates, the reference intercept includes state-owned enterprises, the

machinery industry, the Xibei (northwest) region, and the year 1997. The results from the SUR

are shown in columns (a)-(c) of Table 10. In addition, we estimated a demand equation for

aggregate energy. These results are shown in column (d) of Table 10. In all four cases, prices

are significant and the sign of the coefficients are consistent with our prior expectations – own-

price elasticities are negative implying that an increase in the price of the energy type will

decrease intensity (and increase efficiency defined as GVIO/E) and cross-prices are positive

implying that the three energy types are substitutes. Expenditure on research and development is

also significant for coal, electricity, and aggregate energy. The negative sign on the coefficient

implies an increase in R&D activities increases energy efficiency and decrease energy intensity.

Firm ownership matters only in certain cases. In particular, foreign- invested firms are

more efficient than state-owned firms in their use of all four energy types, as reflected in the

negative coefficient. Collectives, foreign, and Hong-Kong-Macao-Taiwan (HKMT) firms all

generally exhibit greater energy efficiency than their SOE counterparts. Industry-specific factors

are also significant in explaining efficiency levels. Mining, petroleum processing, non-metal

mineral products, metal processing and products, electric power, and other all report high levels

of energy intensity across all four categories relative to machinery. Even after controlling for

differences in prices, ownership, and industry mix, we find significant regional differences.

Huadong (East) and Huanan (South) report consistently lower levels of energy intensity than the

rest of the country. These differences may reflect generally higher levels of firm-level

efficiency in these coastal regions. Lastly, year-specific factors seem to matter largely in the

case of refined oil. This result indicates that while the price and institutional factors included in

the regressions are largely able to explain the declining energy intensity for the oil, electricity,

3 We use current R&D expenditures as a proxy for the accumulation of past R&D expenditures. Regressing lagged R&D expenditures on current R&D expenditures, we find current R&D expenditures to be a good predictor of past

25

and aggregate energy sectors, our regression analysis is generally unable to identify the specific

factors that explain the decline in the intensity of use of refined oil.

In Table 11, we decompose the relative contribution of each of the factors to changes in

the dependent variable (i.e., the log of energy intensity) from 1997 to 1999. Relative

contributions are calculated by multiplying the change in the within sample mean of each of the

independent variable by the relevant coefficient.4

Prices of each of the three energy types increased from 1997 to 1999. In the case of coal,

the increase in coal and electricity prices led to a decrease in coal intensity, but this fall was

offset by the increase in oil prices, which led to a substitution toward coal. For the other energy

categories, however, the decline in energy intensity resulting from own-price increases

outweighed the increase in energy intensity from substitution.

R&D expenditures. Using current R&D expenditures (rather than lagged) allows us to retain one year of observations. 4 Because the firms in our sample do not change industry or region over time, we use coefficients from a regression that uses share of total output in each dummy category rather than zero or one.

26

Table 10

Determinants of intensity by energy type1

Dependent variable = ln(Quantity of energy/constant GVIO)

Coal (a)

Refined Oil (b)

Electricity (c)

Aggregate Energy

(d)

Constant2 -2.763*** (0.239)

-4.541*** (0.208)

-3.564*** (0.151)

-1.360*** (0.141)

Ln(price of coal) -0.513*** (0.057)

0.406*** (0.049)

0.066* (0.036)

-0.023 (0.033)

Ln(price of refined oil) 0.408*** (0.060)

-1.070*** (0.052)

0.092** (0.038)

0.012 (0.035)

Ln(price of electricity) -0.069 (0.045)

0.301*** (0.040)

-0.261*** (0.029)

-0.120*** (0.027)

Ln(R&D expenditures) -0.125*** (0.017)

-0.005 (0.015)

-0.070*** (0.011)

-0.085*** (0.010)

Collectives -0.715*** (0.180)

-0.249 (0.156)

-0.522*** (0.113)

-0.571*** (0.106)

Foreign -0.501*** (0.197)

-0.490*** (0.171)

-0.434*** (0.124)

-0.363*** (0.116)

Hong Kong, Macao, Taiwan

-0.702*** (0.243)

-0.223 (0.221)

-0.302** (0.153)

-0.514*** (0.143)

Shareholding -0.003 (0.089)

-0.057 (0.077)

-0.118** (0.056)

-0.032 (0.052)

Private 0.512 (0.943)

0.245 (0.820)

0.858 (0.594)

0.100 (0.555)

Other 0.011 (0.410)

0.888*** (0.356)

-0.205 (0.258)

-0.095 (0.241)

Mining (06-10, 12) 0.721*** (0.128)

1.529*** (0.111)

1.300*** (0.080)

1.324*** (0.075)

Food and beverage (13-16)

0.157 (0.138)

-0.917*** (0.120)

-0.841*** (0.087)

-0.253*** (0.082)

Textile, apparel, and leather products (17-19)

-0.449*** (0.140)

-1.395*** (0.122)

0.338*** (0.088)

0.142* (0.082)

Timber, furniture, and paper (20-24)

1.435*** (0.170)

-0.511*** (0.148)

0.706*** (0.107)

1.004*** (0.100)

Petroleum processing and coking (25)

2.589*** (0.278)

0.555** (0.242)

0.281* (0.175)

3.320*** (0.164)

Chemicals (26-28) 1.486*** (0.102)

-0.253*** (0.089)

1.074*** (0.065)

1.319*** (0.061)

Rubber and plastic products (29-30)

0.466*** (0.186)

-0.438*** (0.161)

-0.053 (0.117)

0.043 (0.109)

Non-metal mineral products (31)

2.058*** (0.119)

1.053*** (0.103)

1.235*** (0.075)

1.783*** (0.070)

Metal processing and products (32-34)

1.189*** (0.122)

0.826*** (0.106)

1.297*** (0.077)

1.379*** (0.072)

Electric power (44) 3.303*** (0.230)

1.413*** (0.200)

1.046*** (0.145)

3.219*** (0.135)

27

Other (43,45,46) 0.751*** (0.233)

0.990*** (0.203)

2.053*** (0.147)

2.191*** (0.137)

Huabei—North -0.060 (0.122)

-0.391*** (0.106)

-0.236*** (0.077)

-0.048 (0.072)

Dongbei—Northeast 0.132 (0.129)

-0.071 (0.112)

-0.355*** (0.081)

-0.134* (0.076)

Huadong—East -0.024 (0.123)

-0.767*** (0.107)

-0.177** (0.078)

-0.188*** (0.073)

Huanan—South -0.378*** (0.122)

-0.687*** (0.106)

-0.136* (0.077)

-0.204*** (0.072)

Xinan—Southwest -0.216 (0.157)

-0.246* (0.136)

-0.132 (0.099)

0.016 (0.092)

1998 -0.070 (0.076)

-0.206*** (0.066)

-0.019 (0.048)

-0.035 (0.044)

1999 -0.130* (0.077)

-0.194*** (0.067)

-0.028 (0.048)

-0.038 (0.045)

R sq. (obs.) 0.277 (2818)

0.373 (2818)

0.3656 (2818)

0.4752 (2813)

1 Standard errors in parentheses. 2 The constant term include state-owned enterprises, the machinery, equipment and instrument industries (35-37, 39-42), the Xibei (Northwest) region, and the year 1997. * Significant at the 10% level, ** Significant at the 5% level, *** Significant at the 1% level.

In all cases, increases in research and development activities resulted in significant

decreases in energy intensity. For coal, R&D activity was the factor that contributed the most to

the fall in energy intensity. Change in the share of output by ownership type also contributed

significantly to declines in energy intensity. In particular, the declining share of output from

state-owned enterprises significantly reduced the intensity of each of the energy types including

aggregate energy. A substantia l portion of these reallocations of output was made to the

shareholding sector, which exhibits a lower level of energy intensity than state industry. In terms

of region, for coal, refined oil and aggregate energy an increase in the share of output in the east

coast region (in which more efficient firms are typically found) led to a decline in energy

intensity. Changes in the distribution of output across out 12 industry classifications made a

small contribution to declining energy intensity. With the exception of coal, shifts in the

composition of industry output led to small improvements in energy efficiency in each of the

other energy aggregates.

28

Table 11

Decomposition of Relative Contributions

Contribution to Change in Dependent Variable (ln(E/Q)

Coal Ref. Oil Electricity Total Energy

Ln(price of coal) -.0260 .0204 .0033 -.0013 Ln(price of ref. Oil) .0446 -.1158 .0101 .0017 Ln(price of electricity) -.0138 .0595 -.0517 -.0239

Total price contribution .0049 -.0358 -.0383 -.0235 Ln(R&D expenditures) -.0539 -.0018 -.0302 -.0366

State-owned -.0676 -.0134 -.0786 -.0516 Collectives -.0065 -.0027 -.0035 -.0053 Foreign-owned .0002 .0003 .0001 .0001 Hong-Kong, Macao, Taiwan -.0030 -.0011 -.0002 -.0021 Shareholding .0458 -.0039 .0300 -.0279 Private .0000 -.0000 .0001 -.0000 Other .0013 .0044 .0005 -.0006 Total ownership contribution -.0297 -.0165 -.0517 -.0304 Huabei—North .0233 .0024 .0000 .0186 Dongbei—Northeast .0107 -.0019 .0011 .0109 Huadong—East -.0533 -.0222 .0028 -.0499 Huanan—South .0033 .0009 -.0002 .0024 Xinan—Southwest .0091 -.0001 -.0007 .0060 Xibei—Northwest .0044 -.0010 -.0009 .0035

Total region contribution -.0026 -.0218 .0021 -.0084 Mining .0138 .0009 -.0149 .0039 Food & beverage .0019 .0023 .0004 .0018 Textiles .0285 .0319 -.0071 .0171 Timber, paper .0047 .0123 -.0060 .0043 Refined oil .0003 -.0008 .0005 .0012 Chemicals -.0109 -.0270 .0204 -.0063 Refined products .0019 .0022 -.0003 .0018 Non-metals -.0021 -.0065 .0181 .0003 Metals .0007 .0006 .0012 .0003 Machinery -.0372 -.0269 .0056 -.0292 Electric power -.0079 .0016 -.0106 -.0111 Other .0135 .0057 -.0215 -.0039 Total Industry contribution .0070 -.0039 -.0167 -.0197

1997-1999 (residual) -.1017 -.1579 .0163 -.0019 SUM -.1761 -.2376 -.1184 -.1204

Change in E/Q (dependent variable)

-.1751 -.2372 -.1169 -.1201

residual .0010 .0004 .0015 .0002

29

We note several interesting implications of this regression analysis. First, we see that

price and quantity adjustments played an important role in the decline in energy intensity, as

demand seemed to have gravitated toward energy types that posted smaller price increases.

Also, it appears that innovation associated with the R&D process is resulting in considerable

energy savings.

We should also note that, due to the short time series in our data set, the price elasticities

we estimate are principally long-run elasticities since they are based on the cross-sections. From

the magnitude of the “1997-1999 (residual)” term, we see that these long-run elasticities

significantly underestimate the short-run decline in the intensities of coal and refined oil. On

average, the price of coal increased less than the price of refined oil, and the price of electricity

increased significantly more than the prices of coal and refined oil from 1997 to 1999. Thus, in

the case of coal, we see a significant substitution effect from the fall in the relative price of coal.

That is, the total price contribution actually leads to an increase in coal intensity. This

substitution effect is much larger in the long-run and therefore results in an underestimation of

the decline in coal intensity in the short-run.

Our final observation concerns a comparison of the role of sectoral shift. Our Divisia

analysis applied to the 12-sector classification concluded that 26.8 percent of the decline in

industry intensity could be attributed to changes in the composition of industry output. The

decomposition based on the regression analysis attributes a smaller portion – 16.4 percent (i.e.,

“total industry contribution” divided by “sum”) – to changes in the distribution of output across

the same 12 sectors. Apparently some portion of the sectoral shift contribution found in the

Divisia analysis is attributable to systematic differences in price, R&D intensity, changes in

ownership composition, and shifts in regional allocations across the 12 sectors.

30

6. Efficiency of prices

From our previous analysis, we found price changes to be a large contributor to energy

efficiency improvements. This finding raises the question, how efficient are energy markets in

China? That is, how fragmented are energy markets across region, ownership, and industry? If

prices are highly uneven and, in general, higher in coastal areas than in interior areas where we

might expect prices to be more consistent with world prices, then we might surmise that

significant potential exists for efficiency gains by “getting prices right.”

In Table 12, we examine the extent of market efficiency by energy type. To test for

market efficiency, we regress energy prices on ownership, industry, and regional differences. To

the extent that these regressors serve to explain differences in energy prices, we find evidence of

market fragmentation. If these conditions are not significant determinants of energy prices, we

postulate the presence of efficient markets. That is, the closer the R-square is to zero, the more

efficient the market.

Comparing R2’s and coefficients across energy types, we find the market for refined oil

to be the least fragmented and the market for coal to be the most fragmented. Given the

problems associated with transporting coal across China, it is not surprising that we should find

significant fragmentation across regions. This is also true with electricity due to limited linkages

across regions of distribution grids.

Using the coefficient on the Eastern region as an indicator of world prices, we see that

prices in other regions tend to be lower. These results, in combination with the results shown in

Table 11, suggest that the potential for efficiency gains could be realized if regions, ownership

types, and industries all faced uniform (world) energy prices.

31

Table 12 Determinants of energy prices1

Dependent variable = ln(price of energy)

Coal (a)

Refined Oil (b)

Electricity (c)

Aggregate Energy

(d)

Constant2 -1.827*** (0.044)

0.759*** (0.041)

1.273*** (0.052)

-0.475*** (0.044)

Collectives -0.018 (0.049)

0.104** (0.046)

0.158*** (0.058)

0.115** (0.045)

Foreign 0.006 (0.061)

-0.087 (0.058)

0.051 (0.072)

0.207*** (0.052)

Hong Kong, Macao, Taiwan

0.239*** (0.072)

0.023 (0.068)

0.093 (0.085)

0.381*** (0.054)

Shareholding -0.035 (0.030)

-0.039 (0.028)

-0.064* (0.035)

-0.011 (0.028)

Private 0.294 (0.300)

0.162 (0.282)

-0.068 (0.354)

0.461** (0.224)

Other 0.147 (0.121)

0.080 (0.114)

0.199 (0.143)

0.326*** (0.107)

Mining (06-10, 12) -0.365*** (0.044)

-0.031 (0.041)

-0.304*** (0.052)

-0.188*** (0.044)

Food and beverage (13-16)

-0.300*** (0.044)

0.031 (0.042)

-0.066 (0.053)

-0.381*** (0.043)

Textile, apparel, and leather products (17-19)

-0.110** (0.048)

0.033 (0.045)

-0.045 (0.057)

-0.045 (0.047)

Timber, furniture, and paper (20-24)

-0.283*** (0.054)

-0.062 (0.050)

-0.108* (0.063)

-0.414*** (0.051)

Petroleum processing and coking (25)

0.064 (0.084)

-0.164** (0.079)

-0.201** (0.099)

-0.393*** (0.066)

Chemicals (26-28) -0.104*** (0.037)

-0.079** (0.034)

-0.366*** (0.043)

-0.369*** (0.037)

Rubber and plastic products (29-30)

-0.314*** (0.072)

-0.064 (0.067)

-0.128 (0.085)

-0.292*** (0.069)

Non-metal mineral products (31)

-0.260*** (0.038)

-0.120*** (0.035)

-0.196*** (0.044)

-0.489*** (0.038)

Metal processing and products (32-34)

0.063 (0.044)

-0.273*** (0.041)

-0.307*** (0.051)

-0.237*** (0.046)

Electric power (44) -0.489*** (0.064)

-0.147*** (0.060)

-0.592*** (0.075)

-0.991*** (0.040)

Other (43,45,46)2 0.085 (0.074)

0.089 (.070)

-0.146* (0.088)

0.137** (0.064)

Huabei—North 0.172*** (0.043)

0.040 (0.040)

0.193*** (0.050)

-0.002 (0.040)

Dongbei—Northeast 0.205*** (0.044)

-0.154*** (0.042)

0.121*** (0.052)

0.019 (0.042)

Huadong—East 0.590*** (0.042)

-0.028 (0.039)

0.307*** (0.049)

0.285*** (0.040)

32

Huanan—South 0.373*** (0.041)

-0.022 (0.039)

0.252*** (0.049)

0.224*** (0.039)

Xinan—Southwest 0.229*** (0.053)

0.192*** (0.050)

0.157** (0.063)

0.086* (0.052)

1998 -0.139*** (0.025)

-0.160*** (0.023)

-0.135*** (0.029)

-0.145*** (0.023)

1999 0.051** (0.025)

0.137*** (0.023)

0.194*** (0.029)

0.109*** (0.023)

R sq. (obs.) 0.121 (4420)

0.069 (4420)

0.072 (4420)

0.1597 (6309)

1 Standard errors in parentheses. 2 The constant term include state-owned enterprises, the machinery, equipment and instrument industries (35-37, 39-42), the Xibei (Northwest) region, and the year 1997. *Significant at the 10% level, ** Significant at the 5% level, *** Significant at the 1% level.

33

7. Conclusions

In this paper, we have attempted to identify and measure the factors driving the fall in

China’s energy intensity and use during the period 1997-1999. According to our Divisia

calculations, sectoral shifts account for the smaller part of this decline, although the contribution

of sectoral shifts increases as we move to the three-digit level of industry detail. Subsector

energy productivity gains account for most of the decline. Unlike previous studies, we are able to

identify the factors behind improvements in firm-level energy productivity. We find that energy

prices and R&D activities are important contributors to the decline in firm-level energy intensity.

We also find that changes in the share of output by region, ownership type, and industry

contribute to declines in measured energy intensity. The fragmentation of energy markets and

tendency for energy prices to have risen more in coastal areas, where relative prices are more

likely to reflect world prices, suggest the potential for further efficiency gains as energy markets

become more integrated.

A few caveats are appropriate. First, this analysis includes only the industrial sector;

therefore, the effects of changes in final demand sectors such as consumer demand and imports

cannot be directly measured. In addition, our analysis only includes large and medium-size

enterprises and therefore omits small and rural enterprises. Although this omission allows us to

avoid certain data problems that have affected previous studies, it also does not allow us to

assess the complete situation in China. Yet, as shown, notwithstanding the limited scope of our

sample, the patterns of energy consumption exhibited by our sample are consistent with those

found in nationwide data.

34

REFERENCES Ang, B.W. and F.W. Zhang. 2000. “A Survey of Index Decomposition Analysis in Energy and Environmental Studies.” Energy, 25 (2000), 1149-1176. Byrd, W.A. 1989. “Plan and Market in the Chinese Economy: A Simple General Equilibrium Model.” Journal of Comparative Economics 13(2), pp. 177-204. Byrd, W.A. 1991. The Market Mechanism and Economic Reforms in China. M.E. Sharpe, Inc.: Armonk, NY. Garbaccio, R.F., M.S. Ho, and D.W. Jorgenson. 1999. “Why has the Energy-Output Ratio Fallen in China?” Energy Journal, 20 (3), 63-91. Jefferson, G.H. and T.G. Rawski. 1994. “Enterprise Reform in Chinese Industry.” Journal of Economic Perspectives, Vol. 8, no. 2 (Spring), pp. 47-70. Jefferson, G.H., T.G. Rawski, and Y. Zheng. 1996. “Chinese Industrial Productivity: Trends, Measurement Issues, and Recent Developments.” Journal of Comparative Economics, vol. 23, pp. 146-180. Jefferson, G.H., Bai Huamao, Guan Xiaojing, and Yu Xiaoyun. 2002, “R and D Performance in Chinese Industry,” manuscript, Brandeis University, GSIEF. Li, Wei. 1997. “The Impact of Economic Reform on the Performance of Chinese State Enterprises, 1980-1989.” Journal of Political Economy, vol. 105, no. 5, pp. 1080-1106. Lin, X. and K.R. Polenske. 1995. “Input-Output Anatomy of China’s Energy use Changes in the 1980’s.” Economic Systems Research, 7(1), 67-84. National Bureau of Statistics. 2000, 2001a. China Statistical Yearbook (2000, 2001 eds). Beijing: China Statistics Press. National Bureau of Statistics. 2001b. Department of Industrial and Transportation Statistics, annual LME economic and fiscal survey data and annual energy survey. National Bureau of Statistics. 2001c. Department of Population, Social, and Science and Technology Statistics, annual LME science and technology survey. Rawski, T.G. 2001. “What’s Happening to China’s GDP Statistics?” China Economic Review, 12(4), 347-354. Sinton, J.E. and D.G. Fridley. 2000. “What Goes Up: Recent Trends in China’s Energy Consumption.” Energy Policy, 28 (2000), 671-687.

35

Sinton, J.E. and M.D. Levine. 1994. “Changing Energy Intensity in Chinese Industry.” Energy Policy. 17 (March 1994), 239-255. World Bank. 1997. Clear Water, Blue Skies: China’s Environment in the new Century. World Bank, Washington, DC.