2.1 Scale, Composition, Technique Effects and the Decomposition Methodology

A change in energy consumption, pollution and carbon dioxide emissions can be decomposed into three channels: the scale effect, the composition effect and the technique effect (see Copeland and Taylor, 1994, 2003; Grossman and Krueger, 1995; Antweiler et al., 2001; Managi et al., 2009). The scale effect measures the effect on pollution of an increase in the economy size that results from income-driven growth in production. Other things being equal, a positive scale effect means that rising industrial output will drive up CO₂ emissions. Managi et al. (2009) find a positive effect of trade openness on CO₂ emissions for non-OECD countries, providing suggestive evidence of the scale effect from trade-driven increases in industrial production. The second is the composition effect, which measures the impacts of a change in industrial composition. This effect could be positive or negative, depending on the abundance of resources and the strength of environmental policy of the economy. For instance, a shift in an economy from relatively clean service industries towards relatively dirty ones such as steel and cement production is considered a composition effect and will lead to an increase in energy consumption and CO₂ emissions. Moreover, the composition effect may also be positive if more stringent environmental regulations increase costs and drive out polluting firms. A recent study by Zheng and Kahn (2013) shows that the rising costs and tighter environmental standards – especially for carbon dioxide and sulfur dioxide emissions – in the large Chinese cities have pushed those heavily polluting manufacturers to shut down their factories or to relocate them to other places with laxer environmental regulations.

The third one is the technique effect, which relates to the change in the production technique of a given industry. All else being equal, if the manufacturers in an industry adopt more efficient environmentally friendly production methods and improve management quality, it will induce a negative technique effect, thus reducing energy consumption and CO₂ emissions per unit of economic activity within this industry. Shapiro and Walker (2015) find that the observed decrease in nitrogen oxide emissions (NO_x) in the USA during the period of 1990–2008 is more attributable to falling pollution per unit of output within industries at a more disaggregated level, suggesting the existence of a technique effect in those industries. A study focusing on China by Zhang (2012) finds that the changes in input mix, sector energy intensity, fuel mix and carbon intensity of fuels can offset the increasing trade-induced carbon emissions in twenty-six sectors including agriculture, mining, manufacturing and service industries. His analysis provides some evidence that the technique effect contributes to mitigating pollution emissions arising from the energy consumption by trade-oriented sectors.

The decomposition of energy consumption, pollution and CO₂ emission changes into scale, composition and technique effects can be found in several studies. Some of them rely on the Divisia index method (Lin and Chang, 1996; Viguier, 1999). Another avenue makes use of structural decomposition analysis (SDA) based on input-output tables. This line of decomposition methods is based on regression analysis performed on aggregate data, and is often referred to as a ‘top-down’ approach. Twenty years ago, Grossman and Krueger (1995) introduced scale, composition and technique effects into the trade and the environment literature by developing a bottom-up approach, which relies on disaggregated emission and economic activity data.

Many studies have applied this ‘bottom-up’ approach to investigate the air-pollution dynamics using both cross-country and within-country data. For example, using data of sulfur dioxide concentrations across 293 cities in 44 countries from the Global Environment Monitoring Project over the period 1971–1996, Antweiler et al. (2001) decompose the pollution impacts of free trade into scale (GDP), composition (capital-labor endowment ratios), and technique effects. They regress pollution concentrations on representative variables of the above three effects. They attempt to distinguish between the pollution impacts of income changes brought on by international trade from those created by factor accumulation. Their empirical results show a positive scale effect, a negative technique effect and a negative composition effect. The composition effect caused by trade is found to vary across countries depending on relative income and factor endowments. Dean (2002) applies this decomposition framework to estimate the effect of trade openness on water pollution at the provincial level in China from 1987 to 1995. She estimates a two-equation model in which trade has both a direct effect on environmental quality through a composition effect and an indirect effect through an induced technique effect. She finds that the inflow of pollution-intensive industries aggravates environmental damage in China, but openness tends to reduce the environmental costs through stronger regulation as trade-induced income increases. Shapiro and Walker (2015) recently explained the decline of air pollution in the USA between 1990 and 2008 by utilizing this bottom-up decomposition approach. Following this decomposition methodology, this paper examines the change in CO₂ emissions in the context of China for a large number of cities and with more disaggregated sub-sectors in Section 3.2.1.3.

2.2 The Impact of FDI and Environmental Regulation on Industrial CO₂ Emissions

The decomposition framework allows scholars to better understand the dynamics and the underlying mechanisms of industrial CO₂ emissions at a geographic level (national, state/provincial, or city). In this paper we focus on two underlying forces, FDI and environmental regulation, and examine how they affect the three decomposition effects, thus shaping CO₂ emissions dynamics. Existing studies show mixed empirical findings on their effects since they examine different samples in different study periods. Moreover, current studies lack of a clean identification strategy to separate the different channels. Our analysis attempts to improve upon the literature by explicitly looking at their impacts on all three decomposition effects. We first survey the related literature.

3.1 Decomposition Framework

Following the recent literature (e.g. Shapiro and Walker, 2015), we decompose the change in a city's industrial CO₂ emissions into scale, composition and technique effects. Let x_kit be the output measure (we use value-added in this paper) for industry i in city k in year t, e_kit be the CO₂ emission intensity per unit of output. Then, CO₂ emissions for city k in year t, P_kt, is given by

(1)

Let X_kt be the total output for all industries in city k in year t and θ_kit the output share of industry i in city k in year t. Then Equation (1) can be rewritten as follows:

(2)

or in vector notation:

(3)

where θ and e are M × 1 vectors that include the each industrial activity's output share and its pollution intensity, respectively, and M is the number of industries in a given city.

By total differentiation of Equation (4), the change in CO₂ emissions can be decomposed into the following expression:

(4)

In Equation (4), the first term of RHS is the scale effect, measuring the impact of the increased size of total output on CO₂ emissions, holding industrial composition and emission intensities of all industries constant. The second term is the composition effect, measuring how the change in the industrial composition affects CO₂ emissions, holding the scale and emission intensities constant. The last term is the technique effect, capturing the effect of the changes in emission intensities on CO₂ emissions, holding the scale and composition effects constant.

3.2 Data Issues

To conduct the decomposition analysis, we need to econometrically estimate three key parameters in Equation (4): X_kt, θ_kit, e_kit. In this paper, we consider 44 sub-sectors (See Appendix A) across 287 prefecture-level-and-above cities in China during the time period of 1998–2009.⁵ The finer level of industrial disaggregation can help us to better understand the changes in industrial composition and how such changes matter for our decomposition results (Sinton and Levine, 1994; Fisher-Vanden et al., 2004). The terms X_kt and θ_kit can be estimated by using the industrial production data from the database of Annual Survey of Industrial Firms (ASIFs) (for the manufacturing sector)⁶ and from the China's Urban Statistical Yearbooks (for construction and service sectors).

3.3 Decomposition Patterns

Based on the estimates of X_kt, θ_kit, e_kit, we employ Equation (4) to conduct the decomposition. We are able to calculate the CO₂ emission numbers (in levels), and its growth rates (in differences), as well as the three decomposed effects within the differences, by sub-sector by city by year. We are also able to aggregate those numbers to provincial and national levels.

Figure 1(a) illustrates the average annual industrial CO₂ emissions in 287 cities during this 12-year period. We can see significant spatial variation. Large cities, especially the four first-tier cities, emit the most CO₂. Figure 1(b) shows the average annual CO₂ emissions per value added (in Yuan). This CO₂ emission intensity also varies a lot across different cities. However, Figure 2 clearly shows that this intensity has a clear declining trend over time in most cities, attributed to both composition and technique effects. According to our calculation, two first-tier cities, Shanghai and Beijing, rank at the very top places in terms of total CO₂ emissions because of the size of their economies; Urumqi and Xining are among the highest CO₂ emission intensity cities due to the large share of heavy industries in their economies. The annual industrial CO₂ emissions per capita for all cities during the sample period is 2.5 tons, which is 3.4 times as many as the annual residential CO₂ emissions per capita (Zheng et al., 2010). The industrial sector dominates the urban CO₂ emissions in China.

Figure 2 depicts the positions of the 287 cities in terms of their annual growth rates in both total industrial CO₂ emissions (X axis) and CO₂ emission intensities (Y axis). The annual growth rate of industrial CO₂ emissions during this 12-year period is 9.3% for all cities, and 7.4%, 8.9% and 9.4% for the first-tier, second-tier and third-tier cities, respectively. These figures are lower than a recent all sector forecasting exercise by Auffhammer and Carson (2008) based on population projections, which did not account for recent policy intervention and other potential local determinants. Our estimates at the disaggregated city level show that, a large portion of cities locate in the fourth quadrant, enjoying a decline in CO₂ emission intensity (composition and technique effects) but experiencing rising industrial CO₂ emissions due to the sizable scale effect. The first-tier cities are all in the fourth quadrant, and they are among the cities with the fastest declining CO₂ emission intensity. Most of the second-tier cities are also in the fourth quadrant, and they locate at the upper right of first-tier cities, which indicates a relatively higher CO₂ emission growth rate but a slower CO₂ emission intensity decline compared to the first-tier cities. A large share of the third-tier cities are also in the fourth quadrant, while a small number of cities locate in the first quadrant with growth in both CO₂ emission and its intensity. Shizuishan and Yinchuan (both in Ningxia Province) are the top two cities in terms of total CO₂ emission growth and CO₂ emission intensity growth because of their large share of energy-intensive industries; Beijing ranks at the top in terms of total CO₂ emissions decline, thanks to its successful and ongoing transformation to service industries and the vast technical innovations.

Now we turn to our decomposition results. We first look at the national level numbers. Figure 3(a) illustrates the decomposition results for total industrial CO₂ emissions at the national level. We first look at the changes between the initial and end years. Arrow ① shows that, if we had held the composition of industries and the production technology constant (Δθ = 0, Δe = 0) during this 12-year period, what the size of the total industrial CO₂ emissions would be (the column length between C and F). This scale effect equals to a 221% increase (37 billion tons). Arrow ② measures the composition effect. It is positive, but the size is small (1.3%, 22 million tons in total and 3.03 tons per value-added million Yuan) due to the mix of negative composition effects in some cities but positive effects in other cities. The column length between C and E is what the CO₂ emissions would be if every industrial activity still used the production techniques from 1998, which combines the scale and composition effects. Arrow ③ indicates the technique effect, which reduces CO₂ emissions. Its size is much larger than that of the composition effect, with contributing to a 41% decrease (690 million tons in total and 97.5 tons per value-added million Yuan) in CO₂ emissions from China's urban industrial sector.

Figure 3(b) plots these three effects at the national level in each of the 12 years. The solid line depicts the pure scale effect (the would-be CO₂ emissions in each year if all industries keep the same shares and the same production technology as those in 1998). The middle dashed line plots what CO₂ emissions would be in each year if every industrial activity used the original production technique in 1998 (the combination of both scale and composition effects). The gap between the middle dashed and the solid lines shows that the composition effect has led to a smaller change in CO₂ emissions between 1998 and 2007 with somewhat rising CO₂ emissions in the last two years. The bottom dashed line depicts CO₂ emissions when considering all three decomposed effects, so this measures the real CO₂ emissions in every year. The gap between this line and the middle dashed line reveals that the technique effect always works to reduce carbon dioxide emissions over time.

For the purposes of this paper we are more interested in the city-level industrial CO₂ emissions dynamics. Figure 4(a) shows some interesting patterns in these dynamics. We find the three effects play different roles at the three tiers of cities. The scale effect contributes to the rise of CO₂ emissions in all three tiers of cities. The absolute value of this effect is the largest for the first-tier cities and the smallest for the third-tier cities. The technique effect universally leads to the reduction of CO₂ emissions for all three tiers, indicating that all cities have enjoyed the improvements in both production technology and management quality. At the same time, the composition effect on CO₂ emissions is mild. We find that the composition effect contributes to the rising CO₂ emissions in the third-tier cities, while it reduces CO₂ emissions in the first-tier and second-tier cities. Figure 4(b) shows the dynamic change in CO₂ emission per capita across different tiers of cities between 1998 and 2009. We find that second-tier cities have higher levels of CO₂ emissions per capita than the other two. The decomposition patterns using CO₂ emissions per capita across these three groups of cities are quite similar to those using total CO₂ emission illustrated in Figure 4(a). Our decomposition results suggest that there may be a relocation of those energy-intensive industries from the first-tier and second-tier cities to those third-tier medium- and small-sized cities. In this sense, larger cities enjoy less CO₂ emissions at the expense of the increasing emissions in medium- and small-sized cities, so the domestic ‘pollution haven’ does exist. Yet this effect appears to be small.

We further examine CO₂ emissions changes over time across these three tiers of cities, respectively (See Appendix B). The results show that the increased CO₂ emissions in the first-tier cities over the past decade is more attributable to the scale effect, but the composition change towards cleaner industries and the technique improvement helps to offset the otherwise much higher CO₂ emissions. The second-tier cities have experienced similar patterns of these three decomposed effects on the change in CO₂ emissions but the contribution of composition effect on emission reduction is smaller than in the first-tier cities. In the third-tier cities only the technique effect works to reduce CO₂ emissions, but both the scale and composition effects work to increases emissions.

4.1 FDI Measures and Proxies for Environmental Regulation Intensity

4.1.2 Proxies for Environmental Regulations

To measure local officials’ effort in regulating pollution associated with CO₂ emissions, we develop an Environmental Regulation Index (ERI) by combining two observable regulation tools at the city level. One is related to the differentiated industrial electricity pricing across cities. In China, as a legacy of communism, the central government has used its power to control energy prices (Tan and Frank, 2009). However, recently the central government is increasingly willing to allow energy prices to fluctuate and decentralize the energy pricing power to the local level, which enables local governments to use energy prices as a policy tool to attract or regulate energy- and pollution-intensive industries. We collect the price of electricity for industrial usage across provinces from China's Price Yearbooks in the corresponding years.¹¹ As shown in Column (3) of Table 1, there is big variation in the electricity price across the three tiers of cities. The other is the industrial sewage treatment fee that firms have to pay. Lan et al. (2011) use this measure to proxy for the stringency of local governments’ pollution regulation. The data on industrial sewage investment is collected from China's Urban Statistical Yearbooks in the corresponding years. Column (4) in Table 1 illustrates the industrial sewage treatment fees in the unit of RMB/ton. The data reveals that first-tier cities in China are likely to put more investment on industrial sewage treatment than the other two groups of cities.

Calculating ERI begins with transforming two city-level environmental regulation indicators, electricity price changes (EPC) and industrial waste water treatment fee (IWWTF), to two standardized and comparable performance scores ranging from 0 to 1 (SCORE_EPC and SCORE_IWWTF), respectively. ERI for city k is then calculated by averaging the two standardized scores of these two indicators in city k:

(5)

In Equation (5), SCORE_k = (Z_k – L)/(H – L), in which H is the highest level of the indicator, L is the lowest level of the indicator and Z_k is the value of the indicator for city k.

Column (5) in Table 1 presents the ERI across cities. It is found that first-tier cities have higher ERI values than that of the other two groups of cities, which is consistent with the statistics of the two individual indicators in Column (3) and (4).

4.2 Estimation Results

We examine the effects of these local factors on the change in carbon dioxide emissions and its three decomposed components using China's city-level data between 1998 and 2009 via a reduced form approach. Equation (6) is in the form of first-difference because the three decomposed effects are with respect to the changes in CO₂ emissions. The first-difference specification also enables us to drop out city-level time-invariant unobservables. We also include regional fixed effects to control for region-specific trend in CO₂ emissions changes.

(6)

We have four versions of the dependent variable ΔY_k, including the total changes in CO₂ emissions from 1998 to 2009, as well as three decomposed components in city k; ΔFDI_k is the change in the share of total GDP for city k between 1998 and 2009; ERI_k is city k’s environmental regulation index; X is vector of control variables including the changes in GDP, share of secondary industry in GDP and share of tertiary industry in GDP; r_k is regional fixed effects¹² and ε_k is the error term.

Table 2 reports the OLS estimation results and Table 3 provides those with FDI being instrumented with the distance to the closest port. Firstly, we look at the estimated coefficients of the change in our FDI measure (FDI to GDP ratio) using OLS. The coefficient is negative and significant at the 1% level, suggesting that the inflow of FDI pulls down energy intensity and thus CO₂ emissions. Specifically, a 10% increase in the FDI to GDP ratio contributes to a 1.74 percentage point decrease in CO₂ emissions. As for the three decomposed effects, the estimated coefficient of the change in the FDI measure the technique effect is significantly negative, while the estimates on the other two effects are insignificant. These suggest that the inflow of FDI works on pollution emission more through the technique effect than the scale effect. This justifies why we find the optimistic result that FDI helps a city to reduce CO₂ emissions (and corresponding energy consumption). The underlying reasons may be that FDI brings in cleaner capital, technology and better management skills. As the IV estimation results suggest (see Table 3), the inflow of FDI reduces CO₂ emissions, but with larger magnitudes in terms of its impact on the total CO₂ emission change and the change of the technique component compared with OLS estimates. A 10% increase in the FDI to GDP ratio leads to a 6.89 percentage point decrease in total CO₂ emissions and a 4.31 percentage point decrease in CO₂ emissions due to the technique effect. This finding is consistent with our statistical analysis in Table 1 that the cities with more FDI also impose more regulation on the environment.

In Table 3, as expected, the Environmental Regulation Index has a negative impact on industrial carbon emissions. If we look into the three decomposed effects, we can see that stringent environmental regulation works through all three channels – the scale of production drops, the industrial composition shifts towards cleaner industries, and the production technology and management quality also increase (a significant negative technique effect). The magnitudes of the three coefficients suggest that the scale effect dominates this process. The observed composition effect due to the enforcement of environmental regulations is consistent with those in previous studies in the USA such as Berman and Bui (2001a, 2001b) and Henderson (1996). As the instrumental variable strategy here is less than perfect due to the lack of time series variation in the instrument, these results should be taken as suggestive.

The estimated coefficients on the other control variables are also consistent with our expectations. Cities with higher GDP growth experience a larger increase in CO₂ emission mainly due to the increases in economic scale, which dominates the changes from cleaner industry composition and more efficient production technology. The growth in the share of secondary industry is found to increase the cities’ CO₂ emission through scale and composition effect, while the growth of tertiary industry helps to decrease CO₂ emission by improving production technology.

We are also interested in whether the effect of FDI and environment regulation on CO₂ emissions varies across these three tiers of cities. We estimate Equation (6) separately for the first, second-tier cities and third-tier cities, and the regression results are provided in Table 4. It shows that the change in FDI had insignificant effect on CO₂ emissions in the first and second-tier cities, while it tends to decrease CO₂ emissions in the third tier cities mainly through the technique effect. A possible explanation is that the first- and second-tier cities already have advanced technologies (in our study period) so the marginal effect of FDI is weak; while its marginal effect is much larger in the third tier cities where production technologies stand at a lower level. Stringent environmental regulation is found to contribute to the declining CO₂ emission in all three groups of cities. This mitigation effect is mostly attributed to the technique effect in the first- and second-tier cities, while to the scale and composition effects in the third-tier cities. Our results indicate that the environmental regulation in the first- and second-tier cities helps to promote the adoption of greener production technology, but it constrains the scale expansion and the receipt of dirty industries (relocating from the first and- second-tier cities) in the third-tier cities.