A Dynamic and Structural Analysis of Sectoral Energy Intensities in China

A dynamic and structural analysis of sectoral energy intensities in China

Baiding Hu

Department of Accounting,Economics and Finance, Faculty of Commerce

Lincoln University, Christchurch, New Zealand, phone +64 3 325 2811 extn 8069

Overview

Various studies have been carried out to explain factors driving China’s energy consumption, only a few studies have focused on modelling the causes of China’s decline in energy intensity. The study of Fisher-Vanden et al (2004) used the regression approach to modelling energy intensity based on data on industrial enterprises. Garbaccio et al (1999) employed the input-output approach to constructing an index on energy intensity. Employing Schnabl’s (2003) Elasticity Coefficient Analysis method, Hu (2007) used the 1987 and 1997 Chinese input-output tables to studyChina’s energy intensities in intermediate production and final consumption.

A 20 per cent reduction in energy intensity by 2010, measured as energy consumption per unit of gross domestic product, is one of the targets set by the Chinese government to achieve the twin goals of saving energy and reducing emmissions. The aformentioned definition of energy intensity is typically used in the literature (for example, Welsch and Ochsen (2005), Fisher-Vanden et al (2004), Garbaccio et al (1999)). Using this definition, the current paper presents a dynamic and structural analysis of sectoral energy intensitiesusing sectoral level production and energy consumptiondata and fourrecent input-output tables in China. The sectoral level production data form a panel of fifteensectors for the period 1985-2006 with the four Chinese input-output tables beingcompiled for 1987, 1992, 1997 and 2002, respectively. The analysis amounts to describing the dynamic behaviours of sectoral energy intensities, their relationships and the role of sectoral fuel-using (dis)similarity in dictating such relationships.

The analysis begins with testing the stationarity of sectoral energy intensities. A non-stationary energy intensity impliesthat either the growth rate of energy consumption has outpacedthat of output, which ought to be a cause of concern, or the other way around. Then, it is investigated whether there exists a long-run equilibrium for the sectoral energy intensities. An existence would suggest that comovements in output as well as energy consumption among the sectors.

The literature on the comovements of sectoral output and productivity suggests that sectoral input-output relations play a role in generating such comovements. The four Chinese input-output tables are used to characterise the sectoral input-output relations. Since our focus is on energy intensity, the sectors were classified in terms of sectoral distances in energy consumption, which is similar to the sectoral Buy distances proposed by Conley and Dupor (2003). The industry definitions used in compiling the input-output tables do not exactly conform to the sectors in the sector panel data set, thus interpolations are implemented to match the two sources of data.

Section 1 of the paper contains introduction and an overview of current sectoral energy intensities in China. Section 2 describes the data and methodology used for the study. Section 3 presents empirical time series regression results with some concluding remarks in Section 4.

1. An overview of energy intensities in China

The Chinese government started to address the energy intensity problem in 1980, by aiming at decoupling energy consumption and economic growth. As Sinton, Levine and Wang (1998) pointed out, the government policies and implementation measures to curb energy intensity “have been exceptionally successful. Chinais one of the few countries at a relative early stage of industrialisation in which energy demand has consistenly – and over many years – grown significantly less rapidly than gross domestic product. China’s primary energy consumption in 1995 was 1250 million metric tons of standard coal equivalent (Mtce). If enery intensity has remained at the 1977 level, China would have consumed 2700 Mtce in 1996, 2.2 times the actual level.”

In fact, the energy intensity in China over the period 1985-2006 have significantly declined as shown in Figure 1, where “Total” denotes the economy-wide energy intensity. The energy intensities were measured as the amount of standard coal equivalent (SCE) energy in 10,000 tons per 100 million Yuan of value-added in the 1990 prices. Previous studies, such as Zhang (2003) and Ma and Stern (2008), have depicted a similar picture. It is evident that the energy intensity in the secondary industry has recorded the largest fall over the period. Those for the primary and tertiary industries have been generally stable albeit exhibiting a decreasing tendency. It is, therefore, interesting to focus on studying the energy intensity for the secondary industry which is comprised of the manufacuring and construction industries.

Figure 1. Movement of energy intensity: total, primary, secondary and tertiary

2. Data and methodology

This study disaggregates the manufacturing industry into twelve sectors presented in Table 1. Energy Production includes coal mining, crude oil production, electricity generation and petroleum refinery. The first two components are classified,by the National Bureau of Statistics of China, as Mining and Quarrying which is part of the primary industry. In calculating the sectors’ shares,the outputs of the two components were added to the total manufacuring output since the energy production sector is regarded as part of the manufacturing industry rather than the primary industry in the present study. In addition to Construction, AgricultureandMining and Quarrying were also included for comparison purposes.Within the manufacturing industry, Metal productshad producedjust over 14 per cent of the total manufacuring output, on average, over the sample period 1985-2006, the largest share compared to the other sectors in the table. This is closely followed by Energy production and Non-metalic products, each of which accounted for just over 13 per cent of the total manufacuring output. The smallest was recorded for Other manufacturingwith a share of 1 per cent.

Table 2 presents the energy intensities of the 15 sectors, with the energy intensity being calculated as the ratio of energy consumption, measured ten thousand tons of SCE, to total output, measured in 100 million Yuan in the 1990 prices. Three sectors, namely, the Energy production, Chemical industrial products and Non-metalic products and smelting, accounted for 73 per cent of the total energy consumption of the 15 sectors.

The table shows that the energy intensity had been generally declining across the board until 2006, when every sector recorded a significant rise in energy intensity. This is consistent with Zhang (2003) who found that “the trend of real energy intensity declines in the 1980s at the 2-digit level was still maintained in the 1990s”, and but is different to Ma and Stern (2008) who found that China’s energy intensity rose after 2000. The table only shows that Agriculture and Metal Products started to experience an increase in energy intensity from 2001. The evidence in the table, however, is in contrast to Ma, Oxley, Gibson and Kim (MOGK) (2008) who found that China’s energy intensity increased in 2004 on the 1995 level. It is worthwhile pointing out that the energy intensity in MOGK’s study was inferred in the context of the translog cost function, hence it is based on the assumptions underneath the cost function. Rather,the sectoral energy intensities in the table were calculatedusing the definition of energy intensity.

The energy statistics prior to 1994 were incomplete at the 2-digit level. For example, in the 1992 China Statistical Yearbook, while the energy consumption for the manufacturing industry was given those for the Non-ferrous and Ferous Mining, Timber, Transportatio equipment and Electronics were unavailable. In such cases, the 1994 composition of energy consumption was used to impute the energy consumption for those sectors for the period 1985-1992. The National Bureau of Statistics of China did not produce the 1993 energy consumption data, thus the 1993 energy consumption in the present study is the average of the 1992 and 1994 figures.

The output data for the 12 manufacturing sectorsfor the period 1985-1989 were unavailable and therefore were imputed as well. In doing so, it is assumed that the ratios of sectoral output to total industrial output for the 4 years were similar to that in 1987. Using the 1987 input-output table and the total industrial output data for the period, the sectoral output figures were imputed and then inflated in the 1990 prices.

Table 1 Sectors included in the present study

1

1

Table 2 Sectoral energy intensities (10,000 tons of SCE/100million Yuan in the 1990 prices)

Table 3 Sectoral proportions of fuel expenditure in total expenditure: 1987

Table 4 Sectoral proportions of fuel expenditure in total expenditure: 1992

Table 5 Sectoral proportions of fuel expenditure in total expenditure: 1997

Table 6 Sectoral proportions of fuel expenditure in total expenditure: 2002

Table 7. Sectoral distancec in fuel intensities: 1987

Table 8. Sectoral distancec in fuel intensities: 1992

Table 9. Sectoral distancec in fuel intensities: 1997

Table 10. Sectoral distancec in fuel intensities: 2002

1

Apart from the time series data on the sectoral energy intensity, the study also makes use of the latest four Chinese input-output tables that provide structural information about the national economy in terms of sectoral relationship. These are the 1987, 1992, 1997 and 2002 input-output tables, which cover a period of 15 years of the 1985-2006 period. Each of the input-output tables contains 40 sectors/products, of which five are energy products, namely, coal, crude oil (CO), petroleum products (PP), electricity and gas in value terms.

The five fuels are taken into accountin assessing sectoral (dis)similarity in fuel utilisation If the input of each of the five fuels for two sectors accounts for the same proportion of total input, then the two sectors are regarded the same in fuelutilisation. The fuel expenditure for each of the five fuels are divided by the sector’s total expenditure (input)to obtain the proportions for the sector. These proportions, which are, in fact, energy intensities in value terms, are presented in Tables 3-6. The tables show that electricity input was the highest among the five fuels for all the 15 sectors over the period 1987-2002 except for Construction wherein petroleum products were the highest.

If we index the sectors by their vectors of the five fuels (), then the Euclidean distances between the vectors, that is,

measure the sectoral (dis)similarity in fuel utilisation – the shorter the distance between two sectors the more similar the two sectors are. With the four input-output tables, we calculated such distances for the four different years, which are contained in Tables 7-10. These distances form a diagnoal matrix with 0 on the main diagonal since the distance between a sector and itself is zero.

A more visually intuitive way of presenting these distances is to use the multidimensional scaling of Mardia, Kent and Bibby (1979) which, in the present case, amounts to projecting the fifteen dimensional vectors onto the two dimensional plane. Figures 2-5 display the results, where the sectors are labeled by their numbers as indicated in Table 1 above, that is, 1 for AG, 2 for EP and so on, to minimise the overlapping of the sectors’ positions on the plane and their names.

It is evident that over the period 1987-2002 the Energy Production (2) sector is very dissimilar to the rest of the sectors. If the value shares of the five fuels in Tables 3-6 are added up, one will find that more than 25 per cent of the Energy Production’s total input were fuel consumption in 1987, which rose to 31 per cent in 2002. This compares to 12 or so percent for the Non-metalic products and Smelting, the next largest sector in terms of the share of energy expenditure. So, there is the intuition for the Energy Production sector to be unique in fuel utilisation, which, in a way, suggests that the sector’s production technology is different to the non-energy sectors.

The rest of the 14 non-energy sectors seemed very similar as they clustered together. However, the picture may be created due to the large difference between the Energy Production sector and the non-energy sectors. The difference may be so large that to display it has necessarily made the non-energy sectors look more similar than they actually were. Thus, we removed the Energy Production sector from the graph so that the distances between the non-energy sectors can be plotted clearly.

Figures 6-9 show the distances between the non-energy sectors. The Non-metalic Products and Smelting sector became outstanding in 1987. This can be explained similarly to the above, that is, its share of energy expenditure (energy intensity in the value form) was much higher than those of the other non-energy sectors. However, the Minging and Quarrying sector became more and more similar to the Non-metalic Products and Smelting sector.

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

These graphs seem to suggest that four groups had emerged over the period 1987-2002 in terms of similarity. One group may containthe Agriculture, Fodd and Tobacco, Textile, Transportation Equipment, Electronics and Instruments sectors, which are the six sectors clustered in the north-east of the scatter digrams. A second group may be formed by the Timber, Paper and Printing, Machinery,and Construction sectors. A third group may comprise the Chemical Industrial Products, Metal Products,Other Manufacturingsectors, Mining and Quarrying and Non-metalic Products and Smeltingsectors. Since the sectors are grouped based on the five fuel intensities of theirs, these groups provide a basis on which cointegration analysis canbe conducted to study comovements in sectoral energy intensities.

3 Modelling the sectoral energy intensities

Most studies that model China’s energy intensities take a structural approach, that is, they model energy intensity as a function of other factors, such as (and obviously), output, energy consumption, technical progress, and so on. For example, Fisher-Vanden et al (2004)modelled energy intensity as a function of the prices of energy and output, R&D expenditure and a set of dummy variables capturing the impact on energy intensity of geographical, industry and ownership difference. In MOGB’s study, energy intensity is a function of, apart fromthe prices of output and energy, the substitutability between energy and capital and labour. Similarly, Ma and Stern modelled economy-wide energy intensity as a function of sector-specific energy intensities and sector-specific output shares. Thus, if there is a structural change in the form of changing sector output shares, the economy-wide energy intensity will change as a result.

The present study takes a different approach in that energy intensity is not directly modelled as a function of output and prices. Instead, sectoral energy intensities are thought to influence each other because of sectoral relationships and the past path of energy intensities is at least as important as the output shares of the sectors in shaping future energy intensity. This gives rise to the time series approach. We first examine the stationarity of the sectoral energy intensities, which entails unit root testing. Table 11 presents the testing results based on the Augmented Dicky-Fuller (ADF) test (Dickey and Fuller, 1979), using equation. Clearly, all of the sectoral energy intensities are I(1) series. The estimates of the time trend coefficient, , were negative for all the sectors, indicating that there was a deterministic downward trend in the sectoral energy intensities over the period.

Table 11. Unit root testing for sectoral energy intensities

Table 12 Cointegration test: Johansen ML procedure

Next we examine the comovements of the sectoral energy intensities. Sectors that are similar in fuel utilisation, which was defined in Section 2, are likely to exhibit similar patterns of energy intensity movements. Based on the analysis in Section 2, we envisage a cointegration relationship (Engle and Granger, 1987) may exist between the sectors within each of the 4 groups. The cointegrating relationships between the sectoral energy intensities are estimated using the Johansen methodology (Johansen, 1988) and are presented in Table 12.

The computations of the Johansen tests were conducted in Eviews. The first panel of Table 12 indicates that there are five cointegrating relationships for the six sectors, and the third shows that there is one cointegrating relationship the three sectors. It is puzzling that the second panel indicates that there may be five cointegrating relationships among the five sectors, which suggests that the rank of the “π” matrix will be zero, which, in turn, suggests the variables are I(0) (Enders, p333). This would be contradictory to the testing results in Table 11, therefore, we treat this as a Type I error of the cointegration test.