Does Environmental Regulation Drive away Inbound Foreign Direct Investment? Evidence from a Quasi-Natural Experiment in China

Does Environmental Regulation Drive away Inbound Foreign Direct Investment? Evidence from a Quasi-Natural Experiment in China Yi Lu Mingqin Wu Linhui Yu First Draft: May 2012 Revised Version: November 2013 Abstract In this paper, we investigate whether environmental regulation a ects inbound foreign direct investment. Our identi cation uses the Two Control Zones policy implemented by the Chinese government in 1998. The di erence-in-di erences and di erence-in-di erence-in-di erences estimations show that cities with tougher environmental regulation attract less foreign direct investment. Speci cally, toughening environmental regulation causes the amount of FDI to drop by 31:9%. We also nd that the negative e ect of environmental regulation is stronger for polluting industries than non-polluting industries. Keywords: Environmental regulation; Foreign direct investment; Pollution haven e ect; Di erence-in-di erences estimation; Di erence-in-di erence-in-di erences estimation; Two control zones JEL Codes: R11; L25; D22 1

1 Introduction Concerned about further deterioration of living environments, governments across the world are toughening their regulations on pollution with the hope that rms will develop greener technologies and produce more environmentally responsible goods. An unintended consequence, however, is that rms may respond by reallocating production to places with less stringent environmental regulations, a phenomenon known as the pollution haven hypothesis. This may not only counteract the e ects of environmental policies, but also worsen the overall scenario. For example, developing countries may manipulate their environmental policies to attract more foreign direct investment (FDI), which could lead to an increase in the overall pollution levels. Despite of much anecdotal evidence, however, empirical studies fail to provide conclusive ndings on the e ect of environmental regulation, with some nding no such e ect 1 while others documenting signi cant e ects. 2 As a result, the investigation on the pollution haven hypothesis is considered to be one of the most contentious issues in the debate regarding international trade, foreign investment, and the environment (Kellenberg, 2009). An inherent empirical challenge to nd an e ect of environmental regulation on rm location choice is how to deal with the potential endogeneity of environmental regulation. 3 Much of the existing literature treats environmental regulations as exogenous (see Levinson, 2008 for a survey). Recent studies start to tackle the potential endogeneity of environmental regulations, for example, by using either the instrumental variable approach (see Millimet and Roy, 2011, for a survey) or the di erence-in-di erences semi-nonparametric propensity score matching method (List, Millimet, Fredriksson, and McHone, 2003; List, McHone, and Millimet, 2004; Millimet and List, 2004). We examine whether environmental regulation a ects inbound FDI by using a change in environmental policy in China, i.e., the implementation of the Two Control Zones (TCZ) policy (for details about environmental regulations in China, see Section 2). Speci cally, we explore two variations, time (before and after the policy change) and cross-sectional (some cities had the new environmental policy treatment group, and others did not control group), to conduct a di erence-in-di erences (DID) analysis. Our DID estimation shows that cities with tougher environmental regulations attract less FDI, which con rms the pollution haven hypothesis. Meanwhile, the magnitude of the 1 For example, Friedman, Gerlowski, and Silberman (1992); Levinson (1996); Eskeland and Harrison (2003); Javorcik and Wei (2004). In a related study, List (1999) show that air pollution emissions in the U.S. converged during the 1929-1994 period, suggesting that states in the U.S. did not compete for industries by loosening the environmental regulations. 2 For example, Henderson (1996); Becker and Henderson (2000); List and Co (2000); Keller and Levinson (2002); List, Millimet, Fredriksson, and McHone (2003); Kellenberg (2009). 3 Jeppesen, List, and Folmer (2002) conduct a meta-analysis and conclude that di erences in methodological considerations explain much of variations in the ndings of the e ect of environmental regulation. 2

regulatory e ect is found to be large but comparable with those in the literature (e.g., Becker and Henderson, 2000; Kellenberg, 2009): toughening environmental regulations causes the amount of FDI to drop by around 31:9%. The validity of our DID estimation hinges on the condition that the treatment group would have followed the trend of the control group in attracting FDI if they had not implemented the new environmental policy. To verify the satisfaction of this identifying assumption, we conduct a series of sensitivity analyses. First, we control for post-treatment FDI trend generated by the determinants of TCZ designation. We then experiment with saturating the model with city-speci c time trend, checking any expectation e ect, controlling for other policy reform taking place around the same period, and conducting a placebo test, a series of falsi cation tests, and an instrumental variable estimation. Our ndings on the negative e ect of environmental regulation on FDI remain robust to all of these validity checks. While the aforementioned estimations calculate the average e ect of environmental regulation over all industries, the e ect could vary across industries, especially polluting versus non-polluting industries. By using another data source allowing the calculation of FDI at the city-industry-year level, we nd that the e ect of environmental regulation on FDI is negative and statistically signi cant for polluting industries, but statistically insigni cant for non-polluting industries. Furthermore, the city-industry-year level data enables us to conduct a di erence-in-di erence-in-di erences (triple di erence) analysis, which improves on the aforementioned DID estimations by including a whole set of city-year xed e ect, city-industry xed e ect, and industry-year xed e ect. Our triple di erence estimation reinforces the DID estimation results, that is, environmental regulation casts a negative and statistically signi cant e ect on inbound FDI. In addition to the change in the environmental policy, China provides an ideal setting for investigating the pollution haven hypothesis. On the one hand, since it adopted the open and reform policy in 1978, Chinese governments have been aggressively attracting FDI, which has made China the second-largest FDI (stock) recipient country in the world. 4 On the other hand, China s fast economic growth in recent decades has been accompanied by severe environmental degeneration, such as over-exploration and mass industrial pollution, which are typical problems in developing countries. Meanwhile, China is a large country with substantial di erences in the FDI distribution and environmental quality, which provides us with enough variations to identify the e ect of environmental regulation. Our study is similar to and complements other studies using the change of environmental regulations, speci cally, the Clean Air Act and its amendment in the U.S. For example, Becker and Henderson (2000) use a panel data for the 1963-1992 period and nd that rm birth rate for polluting industries in non-attainment areas (that have tougher 4 Based on the statistics by the CIA World Factbook (accessed on August 15, 2013). 3

regulations) is reduced by 26 45%. A recent work by Hanna (2011) uses a DID analysis to investigate how the Clean Air Act Amendments in the U.S. a ects its out ow FDI. Whereas these studies look at the U.S., the largest developed country in the world, we use data from China, the largest developing country in the world. Meanwhile, we investigate whether environmental regulation a ects inbound FDI, whereas Hanna (2011) examines whether U.S. multinationals reallocate their production to foreign countries in response to domestic environmental regulations (or outbound FDI). In the context of China, Dean, Lovely, and Wang (2009), using a sample of equity joint ventures (EJVs) projects, nd that levies on water pollution only a ects the location choice by EJVs in highly-polluting industries and from Hong Kong, Macao, and Taiwan. Our study di ers from theirs by using a di erent identi cation strategy (i.e., DID versus discrete choice model), an investigation of a di erent environmental regulation (i.e., air versus water pollution), and di erent sample period (i.e., 1992-2009 versus 1993-1996). Meanwhile, Hering and Poncet (2011) also explore the change in pollution regulations like us to investigate how environmental regulations a ect exporting activities of rms. The remainder of this paper is organized as follows. The institutional background of environmental regulations in China is described in Section 2. Section 3 discusses the estimation framework, along with a number of robustness checks on the identifying assumption. Data and variables are described in Section 4, and empirical ndings are presented in Section 5. The paper concludes with Section 6. 2 Environmental Regulations in China Timeline: The SO2 emissions generated by coal combustion have increased substantially alongside the fast economic growth in China in the past decades. National coal consumption in 1990 was 1.05 billion tons and increased to 1.28 billion in 1995. In 1993, 62.3% cities in China had annual average ambient SO2 concentration values above the national Class II standard. In Chongqing, the annual ambient SO2 concentration reached 270 or 4.5 times the national Class II standard. Around the same period, 40% of the national territory reported acid rain with average PH value lower than 5.6. SO2 and acid rain may hurt human health and destroy ecosystems, which may consequently impede economic growth. Concerned with its long-term sustainable economic development, Chinese governments started to tackle air pollution issues in the mid 1980s by implementing a series of regulatory policies. The Air Pollution Prevention and Control Law of the People s Republic of China (APPCL) was enacted in 1987 and executed in 1988. This new environmental law provided general principles of air pollution regulation for local governments and related agencies. However, the APPCL was considered very sketchy. For example, it did not present any concrete policies on how to control SO2 emissions and specify which government body should be responsible for enforcing the 4

policies. As a result, the e ect of the regulation on air pollution was limited, with SO2 emissions and acid rain continuing to increase in the late 1980s and early 1990s. With a growing concern over the air pollution problem, Chinese governments decided to take stricter measures. In 1995, the 1987 APPCL was amended, and one chapter about the regulation on air pollution and SO2 emissions was included. More importantly, a new policy, namely the Two Control Zones (TCZ) policy, was proposed to prevent the air quality of those heavily-polluted areas from deteriorating further. In 1997, The Request for Approval of the Proposal of Designation for Acid Rain Control Areas and SO2 Pollution Control Areas was issued by National Environmental Protection Bureau (NEPB) and sent to State Council for approval. In January 1998, the proposal was approved by the State Council (or the cabinet) in the document The O cial Reply of the State Council Concerning Acid Rain Control Areas and SO2 Pollution Control Areas. It was then put into e ect. Among a total of 380 prefecture-cities, 175 were designated as TCZ cities, accounting for 11.4% of the nation s territory, 40.6% of population, 62.4% of GDP, and 58.9% of total SO2 emissions in 1995 (Hao, Wang, Liu, and He, 2001). Figure 1 shows the geographic distribution of TCZ cities in China. In general, SO2 pollution control zones are located in northern China because of the reliance on the thermal energy for heating, whereas acid rain control zones are located in southern China where the climate is relatively more humid. Criteria: The two control zones comprise SO2 pollution control zones and acid rain control zones. The NEPB began designating cities as TCZ cities in late 1995, based on several criteria. Speci cally, a city was designated as a SO2 pollution control zone if: (1) its average annual ambient SO2 concentration was larger than the national Class II standard (i.e., 60 ug/m3) in recent years; (2) its daily average ambient SO2 concentrations exceeded the national Class III standard (i.e., 250 ug/m3); or (3) its SO2 emissions were signi cant. And a city was designated as an acid rain control zone if: (1) its average PH value of precipitation was equal or smaller than 4.5; (2) its sulfate deposition was above the critical load; or (3) its SO2 emissions were large. New Policies: Once a city was designated as a TCZ city, tougher regulatory policies were implemented. For example: 1. New collieries based on coal with a sulfur content of 3% and above were prohibited, and existing collieries using similar quality of coal must gradually reduce the production or be shut down; 2. New coal-burning thermal power plants were prohibited in city propers and suburbs of larger or medium cities, except for cogeneration plants with primary purpose of supplying heat. Meanwhile, newly constructed or renovated coal-burning thermal 5

power plants using coal with a sulfur content of 1.5% and above must install sulfurscrubbers, whereas existing power plants using similar quality of coal must adopt SO2 emission-reduction measures by 2000; 3. In polluting industries such as chemical engineering, metallurgy, nonferrous metals and building materials, production technologies and equipment generating severe air pollution must be phased out; 4. Local governments must strengthen the collection, administration and use of SO2 emission fees. Enforcement: In the 1998 reply, the State Council also laid out the targets for environmental controls in TCZ cities in the short run (by 2000) and in the long run (by 2010). Speci cally, by the end of 2000, the sources of industrial SO2 pollution should achieve the national standard of SO2 emission. The total amount of SO2 emission should be within the required amount. Ambient SO2 concentrations in important cities should achieve the national standards. The acid rain in the acid rain control zones should be alleviated. By the end of 2010, the total amount of SO2 emission should be lower than that in 2000. Ambient SO2 concentrations in all cities should achieve the national standards. The number of acid rain areas with average PH value of precipitation equal or smaller than 4.5 should be reduced signi cantly. These new environmental regulations have generated signi cant improvement in air pollution control. In 2000, 102 TCZ cities achieved the national Class II standard of average ambient SO2 concentrations and 84.3% of severely-polluted rms achieved the target level of SO2 emissions (China Environment Yearbook, 2001). The average growth rate of SO2 emissions from industries and livelihood in TCZ cities from 2001 to 2006 was -6.5% (Annual Statistic Report on Environment in China, 2007). In 2010, 94.9% of TCZ cities had achieved the national Class II standard of average ambient SO2 concentrations, with no city reporting values above the national Class III standard (Report of Ministry of Environmental Protection of the People s Republic of China, 2011). For more discussion on the e ectiveness of the TCZ policies, see He, Huo, and Zhang (2002), Yang, Cao, Ge, and Gao (2002), etc. In Figure 2, we report the annual average ambient SO2 concentrations for TCZ and non-tcz cities from 1992 to 2008 (China Environment Yearbook, various years). There is a clear pattern of annual average ambient SO2 concentrations in TCZ cities decreasing substantially over this period. By 2008, no city reported number above 100 ug/m3 and the SO2 emission in TCZ cities became similar to those in non-tcz cities. Figure 2 also shows that TCZ and non-tcz cities may follow di erent trends of SO2 emission: there was a common U-shape trend for both cities between the late 1990s and the early 2000s; whereas after 2004, SO2 emission continued to decline in TCZ cities but was stabilized 6

in non-tcz cities. To control for the common time trend, we control for year xed e ect in the regressions; and to further control for di erential trends across TCZ and non-tcz cities, we further include city-speci c linear time trend in the analysis. 3 Estimation Strategy To identify the e ect of environmental regulation on FDI, we exploit the TCZ policy that was put into e ect in 1998 in China as a quasi-natural experiment to conduct a DID analysis. Speci cally, there are two groups of cities, the treatment group comprising cities designated as TCZ in 1998 (or TCZ cities), and the control group including cities not designated as TCZ in 1998 (or non-tcz cities). We compare the amount of total FDI in TCZ cities before and after the adoption of the TCZ policy in 1998 with the amount of total FDI in non-tcz cities during the same period. The simplest (but most restrictive) DID estimation speci cation is as follows: Y ct = + T CZ c + P ost t + T CZ c P ost t +X 0 ct+" ct ; (1) where Y ct is the logarithm of FDI in city c at year t; T CZ c indicates city c s TCZ status in 1998, i.e., T CZ c = 1 if city c is a TCZ city; and = 0 if city c is a non-tcz city; P ost c indicates the post-treatment period, i.e., P ost c = 18t 1998 and = 0 otherwise; and " ct is the error term. X ct is a vector of many other potential determinants of FDI, included to isolate the e ect of environmental policy. These controls include agglomeration economies (i.e., domestic output), market size (i.e., retail consumption), economic growth, education (i.e., number of college students and number of high school students), and infrastructure (i.e., electricity consumption, number of telephone, and road area). See Table 2 for the detailed construction and summary statistics of these variables. As an illustration of this DID regression, consider a case of two periods, one period before the TCZ policy (say 1995), and the period after the TCZ policy became e ective (say 2000). The DID estimator is DID = [Y ct (T CZ c = 1; t = 2000jX ct ) Y ct (T CZ c = 1; t = 1995jX ct )] [Y ct (T CZ c = 0; t = 2000jX ct ) Y ct (T CZ c = 0; t = 1995jX ct )] : (2) And the associated identifying assumption is [" ct (T CZ c = 1; t = 2000jX ct ) " ct (T CZ c = 1; t = 1995jX ct )] [" ct (T CZ c = 0; t = 2000jX ct ) " ct (T CZ c = 0; t = 1995jX ct )] = 0: (3) In other words, TCZ cities would have followed the same trend of inbound FDI as non- 7

TCZ cities if there were no TCZ policy implemented in 1998. However, speci cation (1) is restrictive: it assumes all cities behave the same in the same group (i.e., TCZ cities versus non-tcz cities) and the post-tcz yearly uctuations are the same. To relax such restriction, we estimate, in the benchmark case, the following DID speci cation Y ct = c + T CZ c P ost t + t + X 0 ct+" ct ; (4) where c is the city dummy, capturing city c s all time-invariant characteristics such as geographic features, climate, natural endowment, etc; and t is the time dummy, capturing all yearly factors common to all cities such as business cycle, monetary policy, macro shocks, etc. To deal with potential heteroskedasticity and serial correlation, we cluster the standard errors at the city level, following Bertrand, Du o, and Mullainathan (2004). As Chinese provinces usually have di erent regional policies and guidelines for policy enforcement, we further replace year dummies with province-year dummies ( pt ) to control for any arbitrary (time-varying or time-invariant) provincial compounding factors. The inclusion of province-year dummies also provides us with a control for the spatial correlation issues pointed out by Drukker and Millimet (2008). The estimator of interest is, which is expected to be negative. Firstly, the TCZ policies request the phase-out of outdated, dirty production technologies and equipment and strengthen the collection of SO2 emission fees, which increase the production costs especially for polluting industries in the TCZ cities. Secondly, the TCZ policies also prohibit the establishment of new collieries and new coal-burning thermal power plants using low quality of coal and require the installation of desulfurization equipment in existing ones, which lead to an increase in the electricity costs faced by rms in the TCZ cities given that coal is still the main resource for power in China. Ideally, we can test the di erential e ects of environmental regulation on FDI for polluting industries and non-polluting industries, with the former being stronger and the latter being smaller and even potentially insigni cant. However, the city-level data used in the main analysis does not have information about inward FDI disaggregated at sector level; hence, our estimation of re ects the average e ect of environmental regulation (over industries with di erent polluting degrees). Later, we supplement our city-level data with a data of foreign-invested enterprises in 2001, which allows us to check whether the e ect of environmental regulation on FDI is di erent across di erent industries. However, due to some data issues (i.e., a sample selection issue), such analysis is mainly used as a robustness check. See Section 4 for details. The identifying assumption associated with the DID estimation speci cation (4) is that conditional on a whole list of controls ( c ; pt ; X ct ), our regressor of interest, T CZ c 8

P ost t, is uncorrelated with the error term, " ct, i.e., 5 E [" ct jt CZ c P ost t ; c ; pt ; X ct ] = E [" ct j c ; pt ; X ct ] : (5) In the remaining part of this section, we discuss potential violations of our identifying assumption (5) and our remedies, as well as several robustness checks. 3.1 Designation of TCZ Cities A key challenge of using TCZ policy as a quasi-natural experiment to investigate the e ect of environmental regulation of FDI is that the designation of TCZ cities may be correlated with some unobserved determinants of FDI. Note that we employ city xed e ect in all the following regressions; hence, the relevant concern is whether the designation of TCZ cities is correlated with any pre-existing city trends, in particular, the time trend of FDI. Two facts presented in Section 2 may help relieve such a concern. First, the initiation of the TCZ policy and the designation of TCZ status were conducted by the central government and largely exogenous to lower-level (such as city) governments. Second, the designation of TCZ cities was based on several criteria, in particular past pollution levels (i.e., ambient SO2 concentration value or the PH value of precipitation) and speci c threshold levels, both of which could not be manipulated by city governments retrospectively. Nonetheless, we conduct a quantitative analysis to better understand the determinants of TCZ designation. Speci cally, we look at geographic factors (i.e., northern areas, coastal areas, and mountainous areas), political factors (i.e., municipality city or provincial capital city, special economic zones, old industrial cities), economic factors (i.e., economic development, industrial production, number of rms, share of polluting industries, population), as well as the amount of FDI and past pollution levels. See Table 2 for the details of the construction and summary statistics of these variables. The Probit regression results are reported in Table 1. We start with the investigation of geographic factors in Column 1. It is found that cities located in northern and coastal areas are less likely to be designated as TCZ cities, while those located in mountainous areas are more likely to be (albeit, statistically insigni cant). These are consistent with our expectation, as inland cities are more likely to retain SO2 concentrations over long periods of time due to not being reached by prevailing winds, and cities surrounded by 5 Note that the identi cation does not require our control variables to be exogeneous, i.e., E [" ct j c ; pt ; X ct ] = 0: In other words, for these control variables, their estimated coe cients may not have causal interpretation. See Stock and Watson (2012, p274) for more discussion on this point. 9

mountains may take longer time for SO2 emission to disperse. Meanwhile, northern cities are less likely to be designated as TCZ cities because of their heating systems and heavily reliance on thermal instead of hydro powers due to lack of rainfall. In Column 2, we add political factors. While municipalities, provincial capital cities and old industrial cities are found to more likely become TCZ cities initially, they become highly insigni cant once conditional on economic factors. Cities in special economic zones are consistently found to be more likely designated as TCZ cities, suggesting that these special economic zones may be on the top of government agenda. We further include economic factors in Column 3 and nd that cities with larger industrial production and larger share of polluting industries are more likely to become TCZ cities. These are intuitive and consistent with casual observations that the fast economic growth in the past decades in China is accomplished with severe damages to its environment. In Column 4, we investigate whether conditional on these important potential determinants, the designation of TCZ was caused by the past FDI, or a concern of reverse causality. Evidently, the estimated coe cient of FDI is not only statistically insigni cant, but also very small in magnitude, implying that the concern of reverse causality is not relevant in our research setting. Lastly, we in Columns 5-6 add past pollution levels and the speci c threshold levels (the criteria used by the central government to designate TCZ cities) to investigate the e ectiveness of the government policy. It should be pointed out that the construction of these two pollution-related explanatory variables face two data challenges. First, we do not have information about the PH values of precipitation (the assignment variable for TCZ cities in southern China). Our remedy is to use the average annual ambient SO2 concentration to replace the PH value for southern cities, because the dissolution of SO2 in water reduces the PH value and generates acid rain, and the assignment should be comparable across northern and southern cities. Second, information of average annual ambient SO2 concentrations is only available for 76 cities, about 30% of the whole sample. Nonetheless, the regression results are consistent with the government policy: rst, as shown in Column 5, the past pollution level (i.e., average annual ambient SO2 concentrations in 1995) 6 is positively and statistically signi cantly correlated with the TCZ status; second, cities with SO2 concentrations above the speci c threshold level (i.e., 60 ug/m3) set by the central government to designate TCZ cities are more likely to become TCZ cities (see Column 5); and third, conditional on the speci c threshold level, SO2 concentrations level does not have any independent e ects on the TCZ status. In summary, our quantitative analysis shows that the location of the city, industrial production, and share of polluting industries are important and signi cant factors in determining whether a city was designated as a TCZ city. Meanwhile, in a sub-sample of 6 Using average of 1993-1995 or 1994-1995 produces similar results (available upon request). 10

cities with ambient SO2 concentration information, we nd strong evidence of compliance with the government policy: the designation of TCZ status is determined by whether cities past pollution levels are beyond the speci c threshold level set by the central government. Moreover, conditional on these determinants, we do not nd evidence of the TCZ status being determined by the past FDI. 3.2 Augmented Estimation Speci cation In the previous sub-section, we have established that the designation of TCZ cities was largely complied with the government policy (the criteria listed in Section 2). However, there is evidence that TCZ and non-tcz cities ex ante di ered signi cantly (i.e., location, industrial production, share of polluting industries). To alleviate the concern that these pre-existing di erences between treatment and control groups may generate the di erential patterns of inbound FDI over time between these two groups, we control for the post-treatment time trend in inbound FDI generated by these pre-existing di erences. Speci cally, we add into our baseline speci cation the interactions of the post-treatment period indicator (P ost t ) with whether a city is located in northern areas, with whether a city is located in coastal areas, with whether a city is municipality or provincial capital city, with whether a city is located in special economic zones, with whether a city is an old industrial city, with industrial production (1992-1997), and with share of polluting industries (1995). Hence, our augmented DID estimation speci cation becomes Y ct = c + T CZ c P ost t + pt + X 0 ct + Z 0 c P ost t + " ct ; (6) where Z c is a vector of determinants; and the associated identifying assumption is E [" ct jt CZ c P ost t ; c ; pt ; X ct ; Z 0 c P ost t ] = E [" ct j c ; pt ; X ct ; Z 0 c P ost t ] : (7) As further checks on the identifying assumption (7), we conduct several sensitivity analysis, including the control for city-speci c linear time trend, the check on the expectation e ect, a exible estimation, a placebo test using 1996 as the arti cial year of treatment, the control for other two events (the on-going Asian nancial crisis, stateowned-enterprises (SOEs) reform launched by Premier Zhu Rongji, and the relaxation of FDI regulations) happened around the same time as the TCZ policy, a series of falsi- cation tests, an instrumental variable estimation based on the regression discontinuity design, and a triple di erence estimation. For details, see Sections 5.2-5.4. 11

4 Data and Variables Our analysis draws on data from the following four sources: 1. Chinese City Statistical Yearbook for the period 1992 (the earliest) - 2009 (the most recent) 2. Chinese Environment Yearbook for the period 1992 (the earliest) - 2008 (the most recent) 3. The State Council s o cial document, The O cial Reply of the State Council Concerning Acid Rain Control Areas and SO2 Pollution Control Areas 4. A survey of foreign-invested enterprises in 2001 From the rst data source, we collect yearly data about our outcome variable, the amount of (realized) FDI, for each city during the 1992-2009 period, as well as our control variables X ct, determinants Z c used in the TCZ designation regressions, and outcome variables W ct used in the falsi cation tests. Summary statistics of key variables are presented in Table 2. From the second data source, we obtain information about the annual average ambient SO2 concentrations. The SO2 concentration statistics come from the records of many monitoring stations in a few cities, the number of which has steadily increased over time. For example, there were only 65 cities with records of pollution in 1992, whereas in 2003 that number rose to 113. To construct the instrumental variable, we use information from 1995, which contains information on the value of the annual average ambient SO2 concentrations for 80 cities. The third data source provides us with a detailed name list of cities designated as the TCZ. During our sample period (1992-2009), the composition of this list remained unchanged. Appendix Table 1 supplies this list of these TCZ cities. Among a total of 280 cities for which the Chinese City Statistical Yearbook has information, 158 are TCZ cities. One shortcoming of the Chinese City Statistical Yearbook is that it does not have information about FDI for di erent industries or sectors, which forces us to estimate the average e ect of environmental regulation across all industries and prohibits the examination of di erential e ects for polluting and non-polluting industries. To overcome such drawback, we supplement our main analysis based on the Chinese City Statistical Yearbook with the fourth data source. Speci cally, the survey of foreign-invested enterprises contains information about establishment date, registered capital, industrial a liation, and location for around 150,000 foreign-invested enterprises in 2001 (or ~75% of total foreign-invested enterprises). We then aggregate the amount of FDI (regardless 12

of whether the rm is an equity joint venture, contractual joint venture or wholly-owned foreign-invested rm) from rm-level to city-level, and estimate the speci cation (6) separately for polluting and non-polluting industries. One caveat of conducting this analysis is that given the access to only one year survey of foreign-invested enterprises (i.e., 2001), we have to use the location in 2001 to extrapolate location information at the time of entry during the 1992-2000 period. If foreign-invested enterprises relocated or exited during this period, however, we may face a sample selection bias. In awareness of this data issue, we mainly use the analysis based on the survey of foreign-invested enterprises in 2001 as one robustness check. 5 Empirical Findings 5.1 Graphical Results A virtue of the DID analysis is that it can use gures to show transparently how the pollution-regulation e ect is identi ed. To this end, we present in Figures 3-4 the time trends of the di erence in the level and growth rate of FDI between TCZ and non-tcz cities, respectively. Figure 3 shows that the di erence in the amount of FDI between TCZ and non-tcz cities continued to grow during the pre-treatment period (i.e., 1992-1997) but started to narrow down right after the implementation of the pollution regulation in 1998. Meanwhile, Figure 4 shows that the di erence in the growth rate of FDI between TCZ and non-tcz cities was quite stable in the pre-treatment period and started to decline right after 1998, the year of new pollution regulation being e ective. These two gures imply that tougher environmental regulations (i.e., the implementation of the TCZ policy) drive away inbound FDI, a support of the pollution haven hypothesis. 5.2 Main Results Regression results of the speci cation (1) are presented in Column 1 of Table 3. It is found that the interaction between the indicator of the treatment status T CZ c and that of the post-treatment period P ost t is negative and statistically signi cant. This result implies that cities with tougher environmental regulations (i.e., the implementation of the TCZ policy) attracted fewer FDI, con rming the pollution haven hypothesis. In Column 2, we estimate the more exible DID speci cation (4), in which city dummy c is used to replace treatment status indicator T CZ c and year dummy t is used to replace post-tcz period indicator P ost t. In Column 3, we further replace year xed e ect with province-year xed e ect, making the comparison between TCZ and non- 13

TCZ cities in the same province. Evidently, our nding of the negative and statistically signi cant e ect of environmental regulation on FDI remains robust. To address the concerns that pre-existing di erences between TCZ and non-tcz cities may compound our estimates, we control for post-treatment FDI trend generated by these pre-existing di erences in Column 4. For example, we add the interaction between the post-treatment period indicator (P ost t ) and the indicator of northern areas, assuring that our estimation is not driven by either northern cities or southern cities. Consistently, we nd that environmental regulation casts a negative and statistically signi cant e ect on the amount of FDI. The economic magnitude of the e ect of environmental regulation is found to be signi cant. Using our most conservative estimate (i.e., the one in Column 4), we nd that the implementation of the TCZ policy caused the amount of FDI to drop by 33:3%. Such magnitude is comparable to those found in the literature. 7 For example, Becker and Henderson (2000) nd that tougher environment regulations cause rm birth rate in polluting industries to drop by 26 45% in the U.S.. Kellenberg (2009) estimates that during 1999-2003, the failing environmental policy caused the value added of U.S. a liates located in the top 20th percentile countries to grow by approximately 8:6% while the corresponding number for the top 20th percentile developing and transition economies was 32%. Hanna (2011) nds that the Clean Air Act Amendments in the U.S. over the 1966-1999 period increased U.S. multinationals foreign assets by 5:3% and foreign output by 9%. Finally, estimated coe cients of other determinants of FDI also make economic sense, though one should be cautious in interpreting them as causal. For example, FDI are attracted to cities with stronger agglomeration economies, larger market size, higher education levels, and better telecommunication infrastructure. Investment De ection. From a policy viewpoint, it is curious to know whether the signi cant, negative e ect of environmental regulation is due to the withdrawal of FDI into China or the de ection of FDI from TCZ to non-tcz cities. If the latter channel plays a signi cant role, then the magnitude of the 31:9% drop in FDI found in the aforementioned analysis could be exaggerated. While it is di cult to distinguish these two channels, we provide two additional tests to shed light on the extent of investment de ection, based on the premise that investment de ection is relatively easier among neighboring places. First, when we replace year xed e ect with province-year xed e ect, the comparison moves from between any TCZ and non-tcz cities to between TCZ and non-tcz cities located in the same province. Hence, if there were a de ection e ect, we should nd a much smaller e ect of environmental regulation in the regression with year xed e ect than in the one with province-year xed e ect. Clearly, we obtain the opposite e ect, 7 Note that these studies are conducted using di erent methods, di erent data, and at di erent time; hence, the magnitude comparison should be interpreted with caution. 14

i.e., the coe cient drop from 0:330 in the regression with year xed e ect to 0:279 in the regression with province-year xed e ect. Second, for each TCZ city, we construct an alternative control group comprised of all its neighboring non-tcz cities. Speci cally, if a TCZ city and a non-tcz city share a border, then the non-tcz city is de ned as a neighboring city of the TCZ city. If there were a de ection e ect, we should detect a larger e ect of environmental regulation using this alternative control group. 8 However, as shown in Column 4 of Table 3, the estimated coe cient from the regression using this alternative control group is almost identical to the one obtained before. Combined, these two exercises suggest that the tougher environmental regulation may cause foreign multinationals to seek other places in the world to invest. 5.3 Checks on the Identifying Assumption In this sub-section, we conduct a battery of robustness checks on our identifying assumption (7). City-speci c linear time trend. While we have controlled for post-treatment FDI trend generated by those signi cant determinants of TCZ designation, one may still be concerned about some other unidenti ed (or unobserved) characteristics that may generate di erential trajectories of ex post FDI between TCZ and non-tcz cities. As a robustness check, we saturate our DID estimation speci cation (6) with the inclusion of city-speci c linear time trend, which e ectively controls for the FDI time trend generated by all city characteristics in a restricted way (i.e., linear time trend). Regression results are reported in Column 1 of Table 4. It is found that the e ect of environmental regulation on FDI remains statistically signi cant. Meanwhile, despite an increase in the estimated magnitude, the Hausman test shows that the estimate with the inclusion of the cityspeci c time trend is statistically indi erent from those in Table 3. Expectation e ect. Another potential concern regards the timing of the change in environmental policy. Speci cally, as the NEPB began compiling the TCZ list in late 1995 and took two years to get approval from the State Council, one may be concerned whether there is any expectation e ect, that is, the e ect of environmental regulation on FDI happened before the e ective date of the policy (i.e., 1998). To address this concern, we conduct two robustness checks. Firstly, we include in the regression (6) an additional control, indicating the implementation of TCZ policies next year. If there were an expectation e ect, we should nd the comparison between TCZ and non-tcz cities to be signi cantly di erent before the adoption of TCZ policies. Clearly, we nd no signi cant e ect of adopting TCZ policies next year in Column 2 of Table 4, and our 8 Note that the number of observation increases due to the fact that a non-tcz city may be used multiple times in constructing control group for di erent TCZ cities. 15

main results barely change. Secondly, we conduct a placebo test, that is, using 1996 as the time of treatment instead of the real e ective date, 1998. If there is no expectation e ect and the treatment and control groups are comparable before the treatment, then the estimation using 1996 as the time of treatment should not produce any statistical signi cance. Indeed, we nd that it is statistically insigni cant (Column 3 of Table 4). Flexible estimation. To further check the comparability between our treatment and control groups as well as the expectation e ect, we conduct a exible estimation speci cation, that is, replacing the post-treatment period indicator (P ost t ) with a series of time dummies indicating two year ahead to eleven year behind the policy implementation. Such exercise can allow us to check whether treatment and control groups are comparable till the time of the implementation of tougher pollution regulations and become di erent after that. Meanwhile, it can also allow us to check whether the e ect identi ed is due to the pollution regulations or driven by other events happened many years later. Regression results are reported in Column 4 of Table 4. It is found that TCZ and non-tcz cities had similar level of FDI up to two years before the policy change, leading support to our identifying assumption. Meanwhile, the di erence in FDI among these group became signi cantly di erent right after the implementation of the new pollution regulations in 1998, and the magnitude of the rst year e ect (i.e., larger than the benchmark estimate (i.e., 0:421) is comparably 0:333 in Column 4 of Table 3), suggesting that the e ect identi ed in the aforementioned analysis is caused by the change in the pollution regulations. Other policy shocks. If there are other policies adopted at the same time, any ndings of the treatment e ect cannot be attributed only to the e ect of environmental regulation. There are three important events happened around the same time of TCZ reform, the on-going Asian nancial crisis, the SOEs reform launched by Premier Zhu Rongji, and the relaxation of FDI regulations (which allows more wholly-owned FDI instead of equity joint ventures). If the Asian nancial crisis hit TCZ cities more badly, if SOEs reform was carried out more prominently in TCZ cities than non-tcz cities and if equity joint ventures were more concentrated in TCZ cities, our aforementioned estimates of the e ect of environmental regulation could be contaminated. To address this concern, we construct three variables, T rade Exposure c (the ratio of total trade over GDP) 9, SOE P resence c (the percentage of SOEs among all rms), and W F DI P resence c (the percentage of wholly-owned FDI among all foreign-invested rms), all measured as the average of pre-treatment period, to capture potential impacts of the Asian nancial crisis, SOEs reform and FDI policy reform across cities, respectively. We then interact these three variables with the indicator of the post-treatment period (P ost t ) and include the 9 Unfortunately, there is no information regarding the trade exposure to those nancial-crisis-a ected Asian countries like Tailand. Hence, we resort to the general level of trade exposure. 16

three interaction terms (i.e., T rade Exposure c P ost t and SOE P resence c P ost t ) in the regression (6) to isolate the e ect of environmental regulation. Regression results are reported in Columns 5-7 of Table 4. Clearly, none of these three interaction terms has any statistical power. More importantly, our estimated e ect of environmental regulation remains nearly unchanged in terms of both statistical signi cance and economic magnitude. Constructed treatment group. One potential placebo test is to make some non- TCZ cities as the arti cial treatment group and compare them to the remaining non-tcz cities (see, for example, Belot and James, 2011). Given no TCZ policies were actually implemented in all these non-tcz cities, the comparison should not generate signi cant di erence. Speci cally, we use all non-tcz cities that adjacent to one or more TCZ cities (i.e., sharing borders) as our treatment group, and those non-tcz cities that are not adjacent to any of TCZ cities are regarded as the control group. Regression results are reported in Column 8 of Table 4. As expected, the estimated coe cient of T CZ c P ost t is not only statistically insigni cant (and even positive). Falsi cation tests. Instead of looking at FDI as the outcome variable, we examine other outcome variables W ct that are supposed to be una ected by the change in environmental regulation. Any ndings of insigni cant e ects of the TCZ policy on these outcome variables W ct may indicate that there are no other compounding factors taking place in the same time. In Table 5, we report regression results with ve alternative outcome variables, i.e., the number of buses, the number of buses passengers, the number of primary schools, the number of middle schools, and the number of colleges. It is found that none of these ve estimates produce any statistical signi cance and many of the estimated magnitudes are quite close to zero. 10 However, one may be concerned whether the lack of signi cance is due to the fact that an important determinant of these outcome variables is the amount of FDI, which is in turn substantially a ected by the change in environment regulations. In an unreported table (available upon request), we add the logarithm of FDI in all these ve regressions and nd that the estimates barely change in both statistical signi cance and magnitude. In summary, despite the fact that it is impossible to fully verify the identifying assumption (7), the analysis conducted in this sub-section boosts our con dence that our aforementioned estimates may not su er from severe estimation bias. 10 One may be concerned that the statistical insigni ance is due to the lack of time variations for these ve outcome variables. In Appendix Table 2, for each of these ve outcome variables, we report the mean value and standard deviation of the coe cient of variation CV (a standard measure of the degree of dispersion in the literature, de ned as the standard deviation of the outcome variable for an individual city over time divided by the corresponding mean value of all cities). We nd signi cant time variations in these outome variables among all cities during 1992-2009. 17

5.4 Other Robustness Checks In this sub-section, we conduct additional robustness checks on our aforementioned ndings. First, to control for the potential agglomeration determinant of FDI, we include domestic output in each city in the aforementioned regressions. However, an alternative measure used in the literature (e.g., Wagner and Timmins, 2009) is to include historical stock of FDI. As a robustness check, we, in Column 1 of Table 6, include such alternative measure of agglomeration e ect, and nd our results remain robust. Second, a number of studies point out the importance of neighboring attributes in determining the FDI in ows (e.g., Blonigen, Davies, Waddell, and Naughton, 2007). To control for such spatial correlation issue, we include province-year dummies in the aforementioned regressions, making the comparison between treatment and control groups within the same province. Admittedly, however, we cannot control for those neighboring cities located in di erent provinces. As a robustness check, we follow the literature in including a spatially lagged FDI measure (i.e., the inverse-distance-weighted average of the FDI received by all other cities) as an additional control variable in the regression. As shown in Column 2 of Table 6, our results remain robust to this additional control, implying that our results are not driven by any spatial spillover e ects. Third, we exclude four municipalities (Beijing, Chongqing, Shanghai, and Tianjin), which have higher administrative levels and hence potentially di erent government policies. Estimation results are reported in Column 3 of Table 6. It is found that our estimate barely changes with the exclusion of these four municipalities. Fourth, we exclude cities without information about the amount of FDI in 1998 because they do not have immediate post-treatment values. As shown in Column 4 of Table 6, our main ndings on the e ect of environmental regulation continue to hold in this sub-sample. Fifth, we exclude cities without information about the amount of FDI in the 1995-1997 period because they do not have enough pre-treatment values. As shown in Column 5 of Table 6, our main ndings on the e ect of environmental regulation continue to hold in this sub-sample. Finally, we exploit the discontinuity in the designation of TCZ cities to construct a plausibly exogenous instrument for TCZ status. Speci cally, the instrumental variable is constructed as T CZ IV = I SO2 c95 60ug=m 3 ; where I [:] is an indicator function that takes a value of 1 if the argument in the bracket is true and 0 if false; and SO2 c95 is the average annual ambient SO2 concentration in 1995. Regression results are reported in Appendix Table 3. As shown in Column 1, the instrumental variable is found to be positive and statistically signi cantly correlated with 18

our regressor of interest. With respect to our central issue, the e ect of environmental regulation, after being instrumented, remains negative, and its magnitude is statistically indi erent from the estimate without being instrumented, according to the insigni cant Dubin-Wu-Hausman test. However, as expected, due to the severe sample attrition problem (i.e., 30% of the total sample), the standard error of the instrumented estimate is quite large. 11 5.5 Heterogenous E ects and Triple-Di erence Estimation As discussed in Section 3, the e ect of environmental regulation on FDI could be stronger for polluting industries than non-polluting industries. To check such di erential e ects, we use the survey of foreign-invested enterprises in 2001. Speci cally, based on the SO2 emission data in 1995-2004, we rst classify industries with average SO2 emission above 300,000 tons as polluting industries, and the remaining as non-polluting industries (for the similar practise, see Keller and Levinson, 2002; Dean and Lovely, 2010; etc). 12 Then, by assuming no location changes since the establishment and ignoring the exits, we calculate respectively the amount of FDI in polluting and non-polluting industries for each city from 1992-2001 (through the aggregation from rm-level to city-level). Regression results for these two sub-samples are reported in Columns 1-2 of Table 7. Interestingly, we nd that the e ect of environmental regulation on FDI is negative and statistically signi cant for polluting industries, but statistically insigni cant for non-polluting industries, consistent with our argument. With FDI decomposed at the city-industry-year level, we are able to further improve our aforementioned DID estimation by conducting a di erence-in-di erence-in-di erences (triple-di erence) estimation. Speci cally, the triple-di erence estimation speci cation is Y ict = T CZ c P ost t P olluting i + ct + ic + ' it + " ict ; (8) where P olluting i is an indicator of polluting industries. The triple-di erence estimation allows us to control a whole set of industry-year xed e ect, industry-city xed e ect and city-year xed e ect. In other words, all time-varying and time-invariant city characteristics have been controlled for, as well as time-varying and time-invariant industry characteristics and city-industry characteristics. 13 Regression results are reported in Column 11 Another possible explanation for the statistical insigni cance is that our instrumental variable may be weak, as the weak identi cation statistic is below the conventional value for the safety zone of strong instrument (i.e., 10; see Straiger and Stock, 1997). 12 Polluting industries include Firepower Electricity, Production and Distribution of Heat Power, Smelting and Pressing of Non-Ferrous Metals, Smelting and Processing of Ferrous Metals, Manufacture of Non- Metallic Mineral Products, Manufacture of Raw Chemical Materials and Chemical Products, Processing of Petroleum, Coking and Nuclear Fuel, and Manufacture of Paper and Paper Products. 13 A drawback in this triple di erence estimation is that the e ect of TCZ status on FDI in non-polluting industries may not be zero, and hence the TCZ e ect may be underestimated. 19

3 of Table 7. It is found that the triple interaction term has a negative and statistically signi cant e ect, implying that environmental regulation causes the FDI to drop more in polluting industries. These results reinforce our aforementioned DID estimation results on the negative e ect of environmental regulation on FDI. 6 Conclusion In this paper, we investigate whether rms respond to environmental regulations by reallocating their production to places with less stringent regulations. To control for the potential endogeneity of environmental regulations, we use a change in environmental policy, namely China s 1998 TCZ policy. Our identi cation of the e ect of environmental regulation comes from a comparison of the outcome variable for TCZ cities with that for non-tcz cities before and after the policy change, or the DID estimation. By using the amount of FDI for 280 cities over the 1992-2009 period, we nd that cities designated as TCZ attract around 31:9% less FDI than their non-tcz counterparts. The results are robust to a series of robustness checks on the identifying assumption, along with other econometric concerns. Meanwhile, we nd that environmental regulation signi cantly drives away FDI in polluting industries but barely a ects FDI in non-polluting industries. Our paper contributes to the literature on the pollution haven hypothesis by carefully addressing the endogeneity problem associated with environmental regulations. Meanwhile, our use of data from a developing country complements existing studies that focus more on developed countries, particularly the U.S. 20

References [1] Becker, Randy and Vernon Henderson. 2000. "E ects of Air Quality Regulations on Polluting Industries", Journal of Political Economy 108, 379-421 [2] Belot, Michèle and Jonathan Jamesb. 2011. "Healthy School Meals and Educational Outcomes", Journal of Health Economics 30, 489-504 [3] Bertrand, Marianne, Esther Du o, and Sendhil Mullainathan. 2004. "How Much Should We Trust Di erences-in-di erences Estimates?", Quarterly Journal of Economics 119, 249-275 [4] Blonigen, Bruce A., Ronald B. Davies, Glen R. Waddell and Helen Naughton. 2007. "FDI in Space: Spatial Autoregressive Relationships in Foreign Direct Investment", European Economic Review 51, 1303-1325 [5] Dean, Judith M. and Mary E. Lovely. 2010. "Trade Growth, Production Fragmentation, and China s Environment", in Robert Feenstra and Shang-jin Wei ed., China s Growing Role in World Trade, Chicago: NBER and University of Chicago Press [6] Dean, Judith M., Mary E. Lovely, and Hua Wang. 2009. "Are Foreign Investors Attracted to Weak Environmental Regulations? Evaluating the Evidence from China", Journal of Development Economics 90, 1-13 [7] Drukker, David and Daniel L. Millimet. 2008. "Assessing the Pollution Haven Hypothesis in an Interdependent World", working paper [8] Eskeland, Gunnar S. and Ann E. Harrison. 2003. "Moving to Greener Pastures? Multinationals and the Pollution Haven Hypothesis", Journal of Development Economics 70, 1-23 [9] Friedman, Joseph, Daniel A. Gerlowski, and Johnathan Silberman. 1992. "What Attracts Foreign Multinational Corporations? Evidence from Branch Plant Location in the United States", Journal of Regional Science 32, 403-418 [10] Hanna, Rema. 2011. "US Environmental Regulation and FDI: Evidence from a Panel of US-Based Multinational Firms", American Economic Journal: Applied Economics 2, 158-189 [11] Hao, Jiming, Shuxiao Wang, Bingjiang Liu, and Kebin He. 2001. "Plotting of Acid Rain and Sulfur Dioxide Pollution Control Zones and Integrated Control Planning in China", Water, Air, and Soil Pollution 230, 259-264 21

[12] He, Kebin, Hong Huo, and Qiang Zhang. 2002. "Urban Air Pollution in China: Current Status, Characteristics, and Progress", Annual Review of Energy and the Environment 27, 397-431 [13] Henderson, Vernon. 1996. "E ects of Air Quality Regulation", American Economic Review 86, 789-813 [14] Hering, Laura and Sandra Poncet. 2011. "Environmental Policy and Trade Performance: Evidence from China", working paper [15] Javorcik, Beata S. and Shang-jin Wei. 2004. "Pollution Havens and Foreign Direct Investment: Dirty Secret or Popular Myth?", Contributions to Economic Analysis & Policy 3, Article 8 [16] Jeppesen, Tim, John A. List, and Henk Folmer. 2002. "Environmental Regulations and New Plant Location Decisions: Evidence from a Meta-Analysis", Journal of Regional Science 42, 19-49 [17] Kellenberg, Derek K. 2009. "An Empirical Investigation of the Pollution Haven E ect with Strategic Environment and Trade Policy", Journal of International Economics 78, 242-255 [18] Keller, Wolfgang and Arik Levinson. 2002. "Pollution Abatement Costs and Foreign Direct Investment In ows to U.S. States", Review of Economics and Statistics 84, 691-703 [19] Levinson, Arik. 1996. "Environmental Regulations and Manufacturer s Location Choices: Evidence from the Census of Manufactures", Journal of Public Economics 62, 5-29 [20] Levinson, Arik. 2008. "Pollution Haven Hypothesis", New Palgrave Dictionary of Economics, 2nd edition [21] List, John A. 1999. "Have Air Pollutant Emissions Converged among U.S. Regions? Evidence from Unit Root Tests", Southern Economic Journal 66, 144-155 [22] List, John A. and Catherine Y. Co. 2000. "The E ects of Environmental Regulations on Foreign Direct Investment", Journal of Environmental Economics and Management 40, 1-20 [23] List, John A., W. Warren McHone, and Daniel L. Millimet. 2004. "E ects of Environmental Regulation on Foreign and Domestic Plant Births: Is There a Home Field Advantage?", Journal of Urban Economics 56, 303 26 22

[24] List, John A., Daniel L. Millimet, Per G. Fredriksson, and W. Warren McHone. 2003. "E ects of Environmental Regulations on Manufacturing Plant Births: Evidence from A Propensity Score Matching Estimator", Review of Economics and Statistics 85, 944-952 [25] Millimet, Daniel L. and John A. List. 2004. "The Case of the Missing Pollution Haven Hypothesis", Journal of Regulatory Economics 26, 239 262 [26] Millimet, Daniel L. and Joyjit Roy. 2011. "Three New Empirical Tests of the Pollution Haven Hypothesis When Environmental Regulation is Endogenous", working paper [27] Staiger, Douglas and James H. Stock. 1997. "Instrumental Variables Regression with Weak Instruments", Econometrics 65, 557-586 [28] Stock, James H. and Mark W. Watson. 2012. Introduction to Econometrics, 3rd Edition, Essex, England: Pearson Education Limited [29] Wagner, Ulrich J. and Christopher Timmins. 2009. "Agglomeration E ects in Foreign Direct Investment and the Pollution Haven Hypothesis", Environmental and Resource Economics 43, 231-256 [30] Yang, Jintian, Dong Cao, Chazhong Ge, and Shuting Gao. 2002. "Air Pollution Control Strategy for China s Power Sector", Chinese Academy for Environmental Planning, Beijing, China 23

(Source: The national Environmental Protection Bureau, The Proposal of Designation for Acid Rain Control Areas and SO2 Pollution Control Areas ) Figure 1: Distribution of TCZ cities

Non TCZ cities TCZ cities Figure 2: Plot of the mean and standard deviations for ambient SO2 concentrations (ug/m3) in TCZ cities and non-tcz cities