2.1. China's pollution levy system

The pollution levy system has always been a centerpiece of China's pollution regulations. Article 18 of the EPL stated that ‘the levy should be imposed on pollution discharges which exceed national pollution discharge standards, based on quantity and concentration of discharges and levy fee schedules established by the State Council’. By the end of 1981, 27 out of a total of 29 provinces, autonomous regions and municipalities in China started to implement the levy system. In February 1982 China issued an ‘Interim Procedure on Pollution Charges’ (hereafter Procedure). The Procedure defined objectives, levy standards, collection methods and principles for use of the levy collected. The nationwide implementation of the levy system followed the passage of the Procedure – discharge of wastewater, waste gas and solid wastes that were higher than the standards were subject to the levy.

The evolution of the levy system has two distinct periods, pre- and post-2003 regimes. The pre-2003 levy system classifies firms into compliant and non-compliant firms. We denote an individual firm's concentration of pollutant i in category k (e.g., wastewater, air pollutants) by C _i ^k and the pre-specified standard by ${\bar C}_{i}^{k}$ . A firm is a compliant firm if the concentration of each pollutant in category k does not exceed the pre-specified national standard, that is, $C_{i}^{k} \le {\bar C}_{i}^{k}$ for all i in k; otherwise it is a non-compliant firm if $C_{i}^{k} \gt {\bar C}_{i}^{k}$ for at least one pollutant in category k.

The levy schedules differ between water pollution and air pollution and across different regimes. In the pre-2003 regime, as shown in the left part of figure 1, the levy on wastewater discharge for a compliant firm is based on the total volume of wastewater discharge (W) and the levy rate (R ₀ ^W ):

A uniform levy rate of 0.05 Chinese Yen (CNY) per ton of wastewater discharge was used in the pre-2003 regime. Figure 1 also illustrates the block levy rates paid by non-compliant firms in the pre-2003 regime. Formally, the potential levy on pollutant i for a non-compliant firm is:

Figure 1.

An illustration of the levy system for wastewater in the pre-2003 regime

where T _i ^W is the regulatory threshold separating the discharge higher than the standards from the within-standard discharge; R _i1 ^W and R _i2 ^W are the levy rates for the within discharge and discharge that exceed the standards where R _i1 ^W  > R _i2 ^W ; and p _i ^W is a discharge factor that depends on both the total wastewater discharge volume and the degree to which pollutant concentration exceeds the discharge standard, i.e., $p_{i}^{W} = \max \lcub 0\comma \; W\lpar C_{i}^{W} - {\bar C}_{i}^{W}\rpar /{\bar C}_{i}^{W}\rcub $ . The discharge factor varies by pollutants and industry sectors. The central government determines the regulatory threshold and levy rates for each pollutant, T _i ^W , R _i1 ^W and R _i2 ^W . All three parameters are invariant within each industry sector and across geographic regions.

The concentration standards vary by pollutants and industry sectors (see table A1 in the online appendix, available at http://journals.cambridge.org/EDE for more details). The pulp paper sector, for example, is subject to looser standards for three main water pollutants, namely, COD, BOD (biochemical oxygen demand) and TSS (total suspended solids), than the textile and beverage sectors. BOD has a lower concentration standard than COD and TSS. The levy rates for COD, SO₂ and TSS are shown to be decreasing block rates – CNY 0.18 per ton for within-standard discharge and CNY 0.05 for the discharge higher than the standards for both COD and BOD; the corresponding numbers are CNY 0.03 and CNY 0.01 for TSS. The amount of levy for water pollution that a non-compliant firm has to pay is the largest levy among all water pollutants:

where n is the total number of relevant water pollutants discharged by the firm.

In the case of air pollution, only non-compliant firms are subject to pollution levies. For air pollutant i, we denote the levy rate by R _i ^a , the pollutant concentration by C _i ^a , and the corresponding concentration standard by ${\bar C}_{i}^{a}$ . The potential levy of air pollutant i for a non-compliant firm is

where V is the total discharge of air pollutant i measured in cubic meters. Unlike the water pollution levy for non-compliant firms (see equation (2)), the air pollution levy is assessed on the absolute, rather than percentage, deviation from the concentration standard. Table A1 provides values of the key parameters for two common air pollutants, SO₂ and TSP (total suspended particles). A special regulation is imposed on SO₂ discharge among non-compliant firms in the acid rain and/or SO₂ control areas (including most coastal areas); however, a uniform levy rate, CNY 0.2 per kg of SO₂ emission, is applicable regardless of the concentration level.

The pre-2003 levy system has been greatly criticized. It lacks incentives or even provides disincentives to pollution control and abatement because of the decreasing block levy rates for water pollutants, the air pollution levy being not applicable to compliant firms, and constant rates that are not adjusted for inflation (the real value of the later years was substantially lower than in the early years). The levy system was significantly amended in 2003 and called for two steps to calculate the levy amount in the post-2003 regime. The first step is to convert discharge into either COD equivalent for water pollutants or SO₂ equivalent for air pollutants:

where the regulatory conversion parameters for pollutant i are denoted by e _i ^W for water pollution and e _i ^A for air pollution. The conversion parameter takes a value of 1 for COD and 0.95 for SO₂. Equation (5) suggests that both compliant and non-compliant firms need to pay for pollution discharge. The pollutants that are more likely to cause environmental damages are assigned with a smaller conversion parameter and have a greater amount of the equivalents. The equivalent is calculated for all pollutants in each pollution type k, but only the top three pollutants with the greatest equivalent matter in calculating the levy amount:

where the levy rate R equals CNY 0.7 per unit of COD equivalent and CNY 0.6 per unit of SO₂ equivalent for the within-standard discharge, but the rates are doubled for the discharge higher than the standards, i.e., CNY 1.4 and CNY 1.2 for COD and SO₂ equivalent, respectively. In the case of air pollution, all polluting firms, whether they are a compliant or non-compliant firm, need to pay the levy. The rates are the same for firms located in the acid rain and SO₂ control area or other areas.

To illustrate the differences between the pre- and post-2003 levy systems, we assume there are two firms, one compliant firm and one non-compliant firm. As shown in table 1, each emits a total of 500,000 tons of wastewater with three particular pollutants, COD, BOD and TSS. The levy for the compliant firm under the post-2003 levy regime totals up to CNY 35,875, increasing from the pre-2003 level of CNY 25,000 by 43.5 per cent. For the non-compliant firm, the levy under the post-2003 regime is almost six times the levy under the pre-2003 regime, changing from CNY 36,000 to CNY 207,625. The numerical comparisons show that the post-2003 levy system penalized all polluters more than the pre-2003 system did and penalized heavy polluters substantially more.

Table 1.

Numerical comparison of potential levy under different regimes (pre- and post-2003)

^a We assume the total amount of wastewater discharge is 500,000 tons in both examples. The pollutant discharge is the total wastewater discharge multiplied by the concentration of the corresponding pollutant. ^b The conversion parameters for COD, BOD and TSS are one, half and four, respectively.

2.2. Pollution quantity control

China first adopted pollution quantity control to regulate pollution in the environmental special zonesFootnote ³ during the 9th and 10th Five-Year Plan periods. The national pollution quantity control imposed on COD and SO₂ was elaborated in the 11th Five-Year Plan period (2005–2010), released in 2006. The goal was to reduce major pollutant discharges by 10 per cent relative to the 2005 levels by the end of the 11th Five-Year Plan period (2006–2010). In other words, the target was to reduce China's national emissions from 14.14 million tons in 2005 to 12.73 million tons in 2010 for COD and from 25.49 to 22.94 million tons for SO₂. The national pollution control target was further decomposed into the provincial and local levels. The greatest reductions required for COD were in Hebei, Jiangsu and Zhejian provinces – more than 15 per cent of their levels in 2005. Similarly, the great reductions required for SO₂ were in Beijing, Shanghai and Shandong – more than 20 per cent of their levels in 2005. Hainan, Qinghai, Xinjiang and Xizang (Tibet) provinces were not required to reduce their COD and SO₂ emissions, but were required to maintain their 2005 levels (see figure A1 in the online appendix for more details). The provincial control targets were then decomposed into the city/county level and finally to individual firms. For instance, a COD quota given to a firm was calculated by its total production in the base year of 2005 multiplied by the standard of wastewater discharged for each unit of production and then multiplied by the corresponding concentration standard of COD, where the standard of wastewater discharged per unit of production was designed to incorporate the local COD control target.

In order to achieve the quantity control targets for COD and SO₂ by the end of 2010, China implemented a number of policies. For example, whether the region had achieved the local target of pollution control became an important criterion to evaluate local officials' performance and promotion. SEPA was elevated to become the Ministry of Environmental Protection (MEP) in March 2008, which strengthened its position in ensuring compliance with environmental regulations. The minister of MEP claimed that ‘T[t]he total amount control targets have been successfully achieved and the environmental quality of China has been significantly improved over the past 5 years [2005–2010]’ as ‘during the 11th Five-Year Plan Period, China has conducted total amount control of major pollutants. Relative to the 2005 level, COD discharge went down by 12.45 per cent and SO₂ by 14.29 per cent, which exceeded the target level of 10 per cent’. However, rigorous research was not found to have examined the role of the pollution quantity control.

2.3. Other pollution control instruments

China has implemented different types of pollution control instruments including command-and-control, market incentives instruments, voluntary instruments and information disclosure (see details in online appendix table A2). Below we discuss specific instruments implemented or enforced in recent years.

Like other developing countries, China lacks the technical and administrative capacity for the effective monitoring and enforcement of environmental laws, regulations and standards to employ command-and-control instruments effectively. China has adopted pollution control instruments that provide economic incentives to firms. For example, emission permit trading of COD was first introduced in 2008 in the Tai Lake area, an environmental specific zone (Bi and Liu, Reference Bi, Liu, Bi, Otsuka, Ge and Wang2009).Footnote ⁴ Similar permit trading systems have been adopted in other geographical areas and applied to other pollutants. For instance, Chongqing, Shaanxi and Liaoning provinces established their own COD permit trading market in 2010; Chongqing and Shaanxi provinces started trading SO₂ emission permits in 2009 and 2010, respectively.

However, market failures and weak institutional support may limit the effectiveness of market-based instruments. China also used voluntary instruments to induce voluntary pollution control, such as state environmental friendly firm labeling, ecological friendly labeling and the voluntary reduction of greenhouse gas emissions program.Footnote ⁵ On the other hand, corporate performance rating and disclosure (PRD) reduces the information asymmetry between polluters and environmental stakeholders (e.g., consumers, communities, NGOs, investors), empowering these stakeholders to pressure polluters for improved environmental performance (Kennedy et al., Reference Kennedy, Laplante and Maxwell1994; Foulon et al., Reference Foulon, Lanoie and Laplante2002; Bui and Mayer, Reference Bui and Mayer2003; Oberholzer-Gee and Mitsunari, Reference Oberholzer-Gee and Mitsunari2006). The literature has documented the positive effects of PRD programs on regulatory compliance (Wang et al., Reference Wang, Bi, Wheeler, Wang, Cao, Lu and Wang2004; Dasgupta et al., Reference Dasgupta, Wang, Wheeler, Tietenberg and Folmer2007; García et al., Reference García, Sterner and Afsah2007, Reference García, Afsah and Sterner2009). In 1999 China launched its first pilot program of PRD called the Green Watch program in Zhenjiang City, Jiangsu province and Hohhot City, Inner Mongolia Autonomous District. The Green Watch Program rates firms' environmental performance from best to worst by five colors and discloses firms' color ratings to the public. It improved the compliance rates and decreased pollution discharge and pollution intensity (Wang et al., Reference Wang, Bi, Wheeler, Wang, Cao, Lu and Wang2004; Jin et al., Reference Jin, Wang and Wheeler2010). Given the success of its pilot program, the Green Watch program was extended to Jiangsu province in 2001, to eight other provinces during 2003–2005, and promoted to a nationwide implementation since 2005. The Chinese government adopted two critical disclosure measures to improve environmental transparency effectively on 1 May 2005, namely, Government Information Disclosure Regulations (hereafter Regulations) by China's State Council and provisional Environmental Information Disclosure Measures (hereafter Measures) by the MEP. The Regulations state that citizens, legal persona and other organizations have the right to obtain government information by lawful means. The Measures stipulate that: (1) firms listed for violating discharge standards or exceeding discharge quota limits are required to publish their discharge data within 30 days in local media and register the data with the local government agency; (2) non-compliant polluting firms are required to publish their discharge data by relevant local government agencies; (3) the environmental agencies will be legally bound to disclose the list of polluters within 20 days through different channels. If the government agencies fail to fulfill the requirements, the public is entitled to apply for the disclosure and the environmental agencies are required to respond within 15 days. If the agencies turn down the public application for disclosure, the public may report this to the superior environmental authority and/or apply for administrative review or file administrative suits.

To promote corporate social responsibility (CSR) and to foster environmentally and socially sustainable private sector development, the stock markets in China took initiatives. The Shenzen Stock Exchange and the Shanghai Stock Exchange issued voluntary CSR guidance for their listed companies in early 2006 and May 2008, respectively. The MEP and the China Securities Regulatory Commission (CSRC) jointly launched the ‘Green Securities’ policy in February 2008 and then the ‘Green IPO’ policy in June 2008. The Green Securities policy aims to make it harder for polluting firms to raise capital by requiring publically traded companies to disclose information about their environmental record. The Green IPO policy requires firms who want to apply for an initial IPO to include their environmental performance assessment evaluated by the MEP along with financial records for the 36 months in their application before initiating an IPO or receiving refinancing from banks. The MEP conducts assessments and calls for public opinions during a 10-day pre-IPO evaluation period. The applicant firm cannot further pursue an IPO without approval from the MEP. Between February 2008 when the policy was first implemented and September 2008, 20 out of 38 companies were rejected for further review of IPOs because of this procedure (Siddy, Reference Siddy2009). The CSRC and the MEP developed a guide on environmental information disclosure of publically listed companies in September 2010, which requires the listed companies to disclose their environmental performance regularly and to report environmental accidents within 24 hours of their occurrence.

We are not able to estimate the effectiveness of these most recently implemented policies because of the lack of available data. Our empirical analyses mainly focus on the pollution levy system and pollution quantity control.

4.1. Estimation of technical efficiency

Technology and environmental management is an important determinant of pollution (Managi and Kaneko, Reference Managi and Kaneko2009). Using time trend variables to proxy technology changes (a common practice in the existing literature) is likely to capture any time-varying effects that are not necessarily related to technology. Thus, it is important to explicitly estimate technical efficiency. There is an ongoing debate whether deterministic methodologies such as DEA or stochastic methodologies such as stochastic frontier analysis should be used (Bravo-Ureta et al., Reference Bravo-Ureta, Solis, Moreira López, Maripani, Thiam and Rivas2007). Using a meta-analysis to review empirical studies investigating agricultural technical efficiency in developing countries, Thiam et al. (Reference Thiam, Bravo-Ureta and Rivas2001) find no statistically significant differences in these estimates between using stochastic and deterministic approaches. Managi and Kaneko (Reference Managi and Kaneko2009) employ DEA to explicitly estimate technical efficiency and then incorporate it as one of the determinants of pollution abatement. We use a similar approach, but our focus is on the effectiveness of technical efficiency on pollution control.

DEA, developed by Charnes (Reference Charnes1978), is a non-parametric methodology for evaluating the relative efficiency of a set of comparable entities called decision making units (DMUs) with multiple inputs and outputs. It has gained popularity in environmental performance measurement by incorporating undesired output (pollution) along with the desirable output (production) in the traditional DEA framework (Zaim, Reference Zaim2004; Zhou et al., Reference Zhou, Poh and Ang2007, Reference Zhou, Ang and Poh2008a, Reference Zhou, Ang and Pohb; Li, Reference Li2010) for a detailed review of DEA in energy and environmental studies.

Suppose we observe a sample of I DMUs. Each of them use N different inputs X = (x ₁,x ₂,…,x _N ) ∈ R ₊ ^N to produce M different desirable outputs Y = (y ₁,y ₂, …,y _M ) ∈ R ₊ ^M , and K different undesirable outputs (pollution) Z = (z ₁, z ₂, …,z _K ) ∈ R ₊ ^K . The production technology can be characterized by T = {(X,Y,Z):X produce (Y, Z)}. Besides satisfying standard regularity conditions like convexity and closeness (see details at Fare and Primont, Reference Fare and Primont1995; Cooper et al., Reference Cooper, Seiford and Tone2006), the technology T(X, Y, Z) also satisfies the following two properties to account for negative production externalities:

P1 suggests that the proportional reduction in desirable and undesirable outputs is feasible. For example, to reduce pollution (undesirable output), a proportionate reduction in desirable output is needed. P2 implies that production with only desirable outputs is impossible, and the only way to eliminate all undesirable outputs is to stop production.

Technical efficiency is the ability of a decision maker to maximize output with given quantities of inputs and certain technology (output-oriented), or the ability to minimize input uses with a given output target (input-oriented). Output-oriented technical efficiency is commonly used. In particular, following Tyteca (Reference Tyteca1997) the technology efficiency for DMU i at time t is defined as

We define two types of outputs: the undesired environmental output that is denoted by Z and measured by COD- and SO₂-equivalent emissions, and the desired output that is denoted by Y and measured by provincial gross industry productivity. The production inputs denoted by X consist of labor and capital. Equation (7) suggests that an improvement in technology efficiency shifts the production possibility frontier (PPF) outwards since more can be produced from the same amount of inputs given the improvement. Following equation (7), the technical efficiency score (TES) of province i at time t is

A greater TES implies higher technical efficiency.

We then investigate provincial variations of technical efficiency by employing econometric models. We assume that technical efficiency in province i at time t denoted by TE _it is a function of input prices (PXit) consisting of labor wage and capital cost, pollution control variables (levy rates and quantity control targets) represented by Q _it , province-specific economic factors (M _it ), region-specific time trends (γ _i ), and province effects (θ _i ). The reduced structural equation for technical efficiency is

We expect β₁ to be positive because a higher input price is expected to provide an incentive for being more technically efficient. We also expect β₂ to be positive because a higher levy rate or a larger quantity control target also gives firms an incentive to improve technical efficiency so that the undesirable output (pollution) can be reduced. The province-specific economic factors include the percentage of the population aged 15 and above who have at least tertiary education, GDP per capita, and ownership structures (measured by share of GDP contributed by state-owned enterprises, collectively owned enterprises, foreign investment and private companies). We expect that technical efficiency relates to investment in research and technology (Nishimizu and Page, Reference Nishimizu and Page1982; Ahmad and Bravo-Ureta, Reference Ahmad and Bravo-Ureta1996), but such data is not available from the 1990s. We use the annual count of patent applications as a proxy for investment in research and technology.

4.2. Investigating the role of the pollution control system and technical efficiency

We use the following reduced form to investigate the effectiveness of technical efficiency, pollution control instruments and prices of production inputs on pollution intensity:

where S _ijt is the intensity of either COD or SO₂ equivalent discharge (index by j) in province i at time t and TE _it is the estimated technical efficiency. Equation (10) is consistent with the three emission determinants proposed by Grossman (Reference Grossman, Goldin and Winters1995) and Grossman and Krueger (Reference Grossman and Krueger1991): (a) the scale effect that measures change in economic activity represented by the provincial economic variables such as gross output; (b) the composition effect that measures the structural economic changes, substitutions between inputs and the tradeoff between inputs and undesirable output; (c) the technique effect that measures changes in energy/resource intensity, represented by technical efficiency. Equation (10) estimates the effect of key variables of interest on pollution intensity, including φ for technical efficiency; α's for effective levy rates of water and air pollution, the pollution quantity control targets, and pollution abatement costs.

6.1. Estimated technical efficiency

Figures 4(a) and 4(b) show significant regional differences of technical efficiency. On average, technical efficiency was much higher in the Northeast region (0.51), followed by the East region (0.44), the Central region (0.36), and lowest in the West region (0.31). Figure 4(a) shows that the East region was characterized by a few highly technically efficient provinces, while the West region had provinces with high and low technical efficiency.

Figure 4.

Estimated technical efficiency

Following equation (9) we explore the factors that affect technical efficiency and present the results of both fixed effect (FE) and random effects (RE) models in Table 2. Both RE and FE models are estimated with robust standard errors clustered on province to account for correlation within provinces. The standard Hausman test rejects the null hypothesis that an RE model provides consistent estimates as the test statistic overall equals 36.02, which exceeds the critical value at the 1 per cent significance level (p-value = 0.000) (see table 2). However, a serious shortcoming of the standard Hausman test is that it requires the RE estimator to be efficient. Wooldridge (Reference Wooldridge2002) proposes performing the Wald test using cluster-robust standard errors. The robust Hausman test also rejects the null hypothesis and favors the FE model – the Sargan–Hansen statistic (259.31) exceeds the critical value at the 1 per cent significance level (p-value = 0.000) (see Table 2). We therefore conclude that the FE model fits the data better and mainly discuss the estimation results of the FE model.

Table 2.

Panel estimation results of technical efficiency

Notes: ***, ** and * represent 1%, 5% and 10% of statistical significance, respectively.

^a The excluded instruments for provincial GDP per capita are national GDP per capita and opened to trade measured by the ratio of total imports and exports to the gross GDP.

GDP per capita is likely to be endogenous. For example, the possible omitted variable for research and development investment may affect GDP per capita. We use a panel IV fixed effects model to control for potential endogeneity of GDP per capita, where the instruments we propose are national GDP per capita and openness to trade measured by the ratio of total imports and exports to gross GDP. Several statistical tests are performed to validate the instruments (see Table 2). First, the LM version of the Anderson canonical correlation test (Anderson, Reference Anderson1951) rejects the null hypothesis of under-identification, which suggests that two instrument variables are correlated with GDP per capita. Second, the Sargan–Hansen test is conducted to test for over-identifying restrictions. The result provides strong evidence that we cannot reject the null hypothesis that the instruments are valid instruments, i.e., uncorrelated with the error term, and that the excluded instruments are correctly excluded from the estimated equation. The test results for under-identification and over-identification suggest the proposed instrument variables meet the relevance and exogeneity conditions. However, the Cragg–Donald Wald statistic (7.58) is smaller than the critical value at the 10 per cent significance level (19.93) and, thus, we fail to reject the null hypothesis of weak instruments.

The results suggest that wage has a positive effect, but capital cost has a negative effect on technical efficiency, while both effects are statistically significant at the 1 per cent level. The above findings suggest that an increase of wage or a decrease in capital cost is associated with an improvement in technical efficiency. The results may partially explain that the West region is found to be lower in technical efficiency, but more labor and pollution intensive than the East region. The total number of employees per unit of gross output value and pollution intensity of SO₂ equivalent in the West region was more than triple that in the East region, but technical efficiency was lower by 28 per cent on average. We find no statistically significant effect of the quantity control targets on technical efficiency. The effective levy rate for air pollution has a statistically significant effect on technical efficiency – the higher the effective levy rate, the higher the technical efficiency. However, the effective rate of water pollution is not statistically significant.

Table 2 also shows that GDP per capita has a positive, statistically significant effect on technical efficiency, which suggests that the more economically advanced provinces tend to have higher technical efficiency. We find that the percentage of GPD contributed by different ownership structures has no statistical difference in technical efficiency with only one exception for foreign investment. Technical efficiency is lower if a greater share of GDP is contributed by foreign investment compared with that by state-owned enterprises. Another surprising result is that the annual number of patent applications is statistically and negatively associated with technical efficiency at the 10 per cent level.

6.2. Estimation results on intensity of COD- and SO₂-equivalent

Based on the specification in equation (10), we conduct panel estimations of pollution intensity for COD and SO₂ equivalent. As reported in table 3, the robust Hausman test favors an FE model for both COD and SO₂ intensity as the Sargan–Hansen test statistics exceed the critical value at the 1 per cent significance level (χ²(16) = 239.15 for COD and χ²(16) = 659.39 for SO₂). Second, in both estimations we use the first lag of GDP per capita in logarithm to avoid endogeneity.

Table 3.

Estimation results of pollution intensity of COD- and SO2-equivalent

Notes: ***, ** and * represent 1%, 5% and 10% of statistical significance, respectively. Numbers in parentheses are standard deviations. ^a The levy and total amount control are specific for COD- or SO₂-equivalent. Similarly, pollution abatement costs refer to water pollution abatement costs for COD intensity, and air pollution abatement for SO₂ intensity.

We can draw the following conclusion based on the estimation results presented in table 3. First, pollution control instruments are found to be effective in reducing pollution intensity. Both the effective levy rate and the pollution quantity control targets have a negative, statistically significant effect on pollution intensity except for the effective levy rate for the COD intensity (negative but insignificant). Other than these two main pollution control instruments, the regime change does not lead to statistically significant differences in pollution intensity in the 2003–2006 regime compared with the pre-2003 regime for both COD and SO₂ intensity, but the post-2006 regime is characterized by a higher pollution intensity for both COD and SO₂ compared with the pre-2003 regime. Second, technical efficiency has a negative, statistically significant effect on pollution intensity for both COD and SO₂. Third, input prices play important roles. The higher the labor cost and/or the lower the capital cost, the higher the pollution intensity. However, pollution abatement cost does not have a statistically significant effect on pollution intensity. Table 3 also suggests that a high income level is associated with lower pollution intensity, as shown by the coefficient of the first lag of GDP per capita in logarithm.