Pollution Haven Hypothesis - A Meta-analysis

Pollution Haven Hypothesis – A Meta-analysis Abstract In this paper the Pollution Haven Hypothesis (PHH) has been taken for quantitative literature review. An attempt has been made to isolate the systematic factors from the specific factors present in different published empirical works which are conducted to verify the presence of PHH in different contexts, with varied degree of success. According to the PHH, with trade liberalization the rich countries will specialize in cleaner industries since they are expected to have stricter environmental regulations while the poor or low income countries with a laxer environmental regulation will specialize in pollution intensive industries. The channel of pollution movement and the role of domestic policies as well as local characteristics are found to be quite distinct for different studies and no conclusive evidence in this context has been accepted in the literature. After conducting the meta-analysis, it is observed that the systematic factors contributing favorably to the PHH are related to the level of industry- disaggregation at which the study is being conducted, the nature of trade relation being considered (i.e. bilateral or multilateral), the level of economic development and the stringency of environmental regulation. Control for the possible presence of endogeneity among the explanatory factors appears to improve the reliability of the estimates to significant extent. However, no significant contribution of factor endowment hypothesis, the type of data and specification of pollution measures is observed. Keywords: Trade and Environment, Meta-analysis, Pollution Haven Hypothesis, LOGIT Regression, Marginal Effects; JEL Classifications: Q50, Q53, Q56, Q58. 1 1. Introduction The Pollution Haven Hypothesis (PHH) posits that with opening up of the countries to international trade, the polluting industries will relocate to jurisdictions with weak environmental regulations known as ‘pollution havens’ which are generally the low income countries. On the other hand, the high income countries having stringent environmental regulations will specialize in clean goods. The intuition behind such hypothesis is as follows. Environmental regulations raises the compliance cost 1 resulting in the rise in cost of key inputs required to produce the pollution intensive goods which reduce the country’s comparative advantage. The PHH signifies the impact of trade on environment and most of the discussions in the trade-environment area is centered on PHH. International Trade has been a very important issue when it comes to the context of environment. The effect of trade on environment can be captured through three channels – the scale effect, the composition effect and the technique effect. The scale effect basically captures the effect of growth (occurring through trade) on environment. According to this effect with increase in liberalization, a country will witness a higher growth trajectory which in turn can deteriorate the environment through scale effect due to increased economic activity. The composition effect on the other hand captures the change in the composition of the trade baskets of various trade participating countries. The composition effect can have either a positive or negative impact on the environment of a country depending on the pattern of specialization of a particular country. For example, according to the PHH, the developing countries should specialize in dirty goods since they 1 Compliance cost is the expenditure of time or money in conforming with government requirements such as legislation or regulation. 2 have less stringent environmental standards. So the composition effect for a developing country should have a negative impact on environment. Finally, the technique effect will always contribute favorably to environmental quality by introducing greener technology. This effect is caused by changes in input mix, sector energy intensity, fuel mix and carbon coefficients (Zhang, 2011). So, the ultimate effect of trade on environment will depend on the relative strength of these three effects, which in their turn depend on the specific policy driven incentives by the country’s governments. Whether openness leads to degradation of environmental quality is an important query. Basically the environmental divergence brought in by free trade has been analyzed. Thus testing the PHH is nothing but testing for environmental divergence. 2 If the PHH holds true then we have environmental divergence since with trade the environmental policies of the developing countries may become laxer to attract the polluting industries. Till date several theoretical and empirical literatures evolved around this area mostly dealing with the testing of PHH. The inference drawn from the studies are mixed. Some found significant evidence for the PHH while some found negligible or no evidence for the hypothesis. Since already a good amount of work has been conducted in this area, there is a little scope for new avenues to open up. Testing PHH in different frameworks has been witnessed in the previous literature. There is presence of heterogeneity among the studies in the form of different methods used, different set of independent and dependent variables considered, different estimation techniques used, etc. So, conducting another test for PHH cannot make any substantive contribution to the pollution haven debate. In this case, meta- analysis can be really a good method which should help us in understanding the heterogeneity 2 By environmental divergence we mean both environmental policy divergence and quality divergence since policy divergence will lead to environmental quality divergence and both of them are interrelated. Since we are interested in capturing the impact of trade, it is irrelevant which divergence we consider first. 3 present in the literature as well as isolating the factors which are increasing the chance of PHH holding true as well as separating out two factors – specific factors and structural factors. 3 Meta-analysis is a statistical technique for combining the findings from independent studies. Previously this method was used in medical science by doctors to assess the clinical effectiveness of healthcare interventions. It does this by combining data from two or more randomized control trials. Meta-analysis has been recently used in social sciences and in economics this method is also called as meta-regression analysis. This study attempted here is similar to the other meta-analyses which have been conducted in various topics such as Ricardian equivalence (Stanley, 1998), environmental Kuznets curves (Cavlovic et al., 2000, Li et al., 2007), export growth hypothesis (R. Mookerjee, 2006) and many other topics in environmental economics. These studies revealed that a meta-analysis is a much superior technique of summarizing several empirical evidences by isolating similar relationships and removing biases present in the literature. The bias in the literature is generally created when a debate regarding a particular phenomenon or theory evolves and some studies find any evidence for such theories. Then it is observed that most of the studies try to adapt to the new theory and focuses on finding evidence for that theory even if it is not witnessed in reality. Similarly in the PHH literature, most of the studies got involved in finding significant evidence for the hypothesis just after the claim of such migration of dirty industries by Lawrence Summers (who was then the chief economist of the World Bank) in in the Summers Memo in 1991. Summers in that memo on trade liberalization mentioned that “Dirty Industries: Just between you and me, shouldn't the World Bank be encouraging 3 Specific factors are those factors which is influencing the chance of getting PHH in this case for some specific papers which depends on the time period of analysis, the region or industries studied, etc. The Structural factors or the systematic factors are those which influence the chances of getting a valid PHH in general for all studies. 4 MORE migration of the dirty industries to the LDCs”. Such statements led to heated criticism and that lead to the birth of the PHH debate. To remove such biases meta-analysis is required. However, this is the first meta-analysis done for PHH using 43 studies with 59 observations to explore the impacts of several methodological choices and other control variables on the evidence for PHH. Since meta-analysis deals with a quantitative review of the literature, so for conducting this method for PHH only the empirical literature is considered in this study for analysis. Thus an extensive review of the empirical literature for PHH has been done in section 2. In order to set up a strong base for understanding the meta-analysis performed on PHH in this paper, the concept of meta-analysis has been introduced in section 3, while the whole of section 4 is dedicated to discuss the details of how the meta-analysis of the pollution haven hypothesis has been conducted along with the results and its interpretations. Finally, the last section concludes the paper by providing an overall assessment and directions for future research. Further, the references have been subdivided into two sets of bibliography – categorized and non-categorized bibliography. The ‘Categorized Bibliography’ consists of all the literatures which are there in our meta-regression analysis data set and also includes those literatures related to the concept of meta-analysis. The other part is the ‘Non-categorized Bibliography’ where the remaining references are included which was needed for understanding the trade- environment debate. 2. Empirical Evidence In the literature a distinction has been made between Pollution Haven Hypothesis (PHH) and Pollution Haven Effect (PHE). However this distinction was first pointed out by Copeland and Taylor (2003). According to them, there is a general tendency to put PHH and PHE in the same bracket. But the two are different. PHE is witnessed when environmental regulation 5 affect plant location decisions and trade flows, i.e. a strict environmental policy should decrease the net exports of dirty goods. The PHH on the contrary, is a stronger version of the PHE which predicts the relocation of pollution intensive goods from developed countries with stricter environmental regulation to the developing countries with a laxer regulation. In short PHH captures the effect of trade on environment while the PHE captures the effect of changing environmental policy on a country’s trade pattern. But the PHE is so strong that it more than offsets other motives for trade in dirty goods. On the other hand relocation of dirty industries to less developing countries can be not necessarily driven by environmental costs which cannot be captured by the PHH. But the PHH and PHE are almost overlapping concepts. The PHE actually lead to the PHH. All the papers before 2003 did not consider the difference. Even most of the papers post 2003 considered both as same. Since meta-analysis require sufficient number of observations, doing separate meta-analyses for PHH and PHE will reduce the sample size and thus result in inconsistent results. Also it is very hard to distinguish between the two effects since both them are leading to a higher pollution in the developing countries and a lower pollution in a developed countries. So both the PHE and PHH studies will be considered under the same bracket in this paper. Econometrics Studies of the PHH or PHE generally focused on the reduced form regressions where a measure of economic activity is regressed on some measure of regulatory stringency and other covariates. The reduced form regression equation is as follows: Yi = αRi + X i' β + ε i Where Y denotes economic activity, R is regulatory stringency, X is other characteristics that affect Y and ε is an error term. The empirical literature on PHH contains a wide variety of implementations of the above reduced form equation. Some studies focus on international 6 trade and considered Yi as net exports from country I while the right hand side contains the country characteristics. While the other set of studies foreign direct investment or new manufacturing plant births as their preferred proxy for economic activity Yi. Some studies used country level data to estimate the above equation while some preferred to examine the validity of the PHH at a disaggregated industry level data to get a better picture. To understand such heterogeneous empirical studies in a better way, the whole of the literature review will be presented in an analytical way rather than depicting it in a descriptive way. This analytical and structured survey would be also appropriate for the purpose of meta-analysis. The review has been divided into several subsections depending on the type of empirical studies conducted by different authors in PHH. The sections will try to compartmentalize the papers based on the methodologies used, the type of data analyzed, the specified model, stringency measures and instruments, the trade type considered and the level of disaggregation considered in the data. 2.1 Based on PHH measures: PHH has been verified by different studies through two broad indicators – one set considered FDI being a pollution transmission channel 4 while the other set of studies dealt with net exports acting as a proxy for PHH. The papers which used FDI inflow or outflow as an instrument to verify PHH, considered the fact that with trade liberalization the FDI will flow out of the country imposing stricter environmental regulation. With strict regulation, the cost of production of the pollution intensive goods increases and since developed countries are the one imposing such strict regulation then it is expected that the developing countries will 4 FDI as a pollution transmission mechanism should not be confused with the capital flight hypothesis. According to the capital flight hypothesis, the capital flows out of a country due to many reasons such as returns from investment of other countries, future prospects, political instability, etc. But we cannot consider it to prove PHH since PHH deals with the impact of environmental regulation on FDI inflows or outflows. The capital flight is a much broader concept and it includes many factors other than the environmental policies. 7 enjoy a cost advantage in these dirty goods. So, according to the PHH, the developed countries will use FDI as their transmission channel to shift the pollution intensive activities to their developing counterparts. On the other hand, the second set of literatures was focused on finding the impact of environmental regulation on the exports and imports of a trading country. If it is observed that the countries having a stricter environmental regulation export cleaner goods and import dirtier products then the papers concluded that there is significant evidence for a PHH. Unlike FDI it is the trade in dirty goods and not a shift of the industries. In this case pollution flows through goods while in case of the FDI measure, pollution flows through capital outflows. McConnell and Schwab (1990) studied the impact of environmental regulation on the location decisions in Motor vehicle industry. Since vehicle industry is a major contributor to ozone problem, this industry is a dirty industry. He observed the location choice of the firms and found that the regulation proxied by attainment status is not significantly affecting the location decision of the motor vehicle industries. However some evidence was there showing that firms are deterred from locating in regions where there is a serious ozone problem. Similar set of studies have been done by Levinson (1996), Becker and Henderson (2000) and others. Although different studies were done for different countries and different time frames. Although some papers got evidence for PHH, still ambiguity remains for many others (Eskeland and Harrison 2003; Manderson and Kneller 2012; and Elliot and Shimamoto 2008). Eskeland and Harrison (2003) did not get any evidence for PHH since their analysis found that the US outbound FDI increased in areas where there are high abatement costs. So it is very difficult to infer that whether PHH holds true in general. 8 2.2 Stringency measures and instruments: To find an appropriate measure of regulatory stringency (R) is not simple. It is not about only collecting the data on “R” but also the measures has to be conceived which is pretty difficult. The environmental compliance costs can take many forms like environmental fees or taxes, permitting costs, emission standards, product or process redesign, forgone output and so on. Some papers have attempted to measure these costs by creating indices by giving weights to various country or state characteristics such as the environmental agencies budgets, public awareness of the environmental problems, the country’s involvement in several international environmental agreements, etc. For example the stringency measure used by Tobey (1990) in his analysis is a qualitative index based on an ordinal ranking of countries (scale from 1 to 7, where 1 denotes the most lax regulation and 7 denote the strictest regulation). Van beers and Van den Bergh (1997) also used strictness of environmental policy same as Tobey since at that time other indicator was not available. They also mentioned the difference between input based and output based indicators. Input oriented indicators consist of the input efforts devoted to environmental protection (e.g. R&D activities, PAC which is used as proxy for stringency of environmental regulation since high value of these denotes the country has a stricter environmental regulation). On the other hand, according to Van beers and Van den Bergh, “output oriented indicators reflect the concrete results of environmental regulations and can be considered as a better proxy for environmental policy strictness than input-oriented indicators under the assumption that better environmental performance is due to stricter environmental regulations.” The output oriented measure allow for assessing the effectiveness of policies ex-post If a government would impose an energy tax on domestic producers and at the same time directly subsidies energy use. This will be translated in a smaller reduction in energy use 9 than without subsidies and therefore a less strict environmental policy. This will be picked up by the relevant output-oriented environmental regulations strictness indicator. Dinda (2006) used output oriented stringency where he considered per capita carbon dioxide emissions as a proxy for stringency and found that for OECD countries, openness reduced the emissions. Cole and Elliot (2005) used energy-GDP ratio as a proxy for stringency which is an output oriented indicator. But they did not find any evidence for PHH. Other studies used different measures of pollution directly as a proxy for regulatory stringency. For example, if for a country, a high carbon dioxide emissions is witnessed then it is believed to reflect the laxity of environmental regulation in that country. But studies which are based on US data have generally used Pollution Abatement Cost Expenditures (PACE or PAC) as a measure of regulatory stringency. For example papers by Kalt (1988), Birdsall and Wheeler (1993), Ederington, Levinson and Minier (2003), Manderson and Kneller (2012) used PAC or PAOC (Pollution Abatement Operating Costs) as their regulatory stringency variable. Due to the presence of such varied degree of heterogeneity present in the usage of different stringency measures we can club different papers under different heads in the meta-analysis section of this paper. 2.3 Type of data used: While enumerating the empirical literature, it can be found that the first generation of empirical work on the PHH used cross sections of data making no attempt to control for unobserved heterogeneity. Most of them found small or insignificant evidence for PHH and none of them found any robust significant support for PHH. The problem of the cross sections studies is that they failed to take into account the correlation between the country or 10 industry characteristics with both economic activity and the regulatory stringency. For example a country with a comparative advantage in dirty goods is most likely to both export that good and impose strict environmental regulation. This may lead to a less robust result. In fact many studies found that countries with strict environmental policies specializes in dirty goods which is the opposite of PHH. While in recent years economists have begun to use panel data with fixed effect models to control for unobserved heterogeneity. In contrast to the earlier cross section studies, these newer ones have found some statistically significant evidence for PHH. In one of the earliest papers, Kalt (1988) used a cross sectional Heckscher Ohlin (HO) model to analyze whether domestic environmental policy affects the competitiveness of the US industries. Results show that pollution abatement costs have a positive impact on net exports (i.e. with increase in PAC, the net exports of dirty goods by the developing countries increased). But the effects became negative as some industries were removed from the data set and the regressions are run. This counterintuitive result may be due to presence of heterogeneity across industries which the cross sectional studies cannot capture. Tobey (1990) also conducted a similar kind of study where he used a cross sectional HO model to study trade patterns in five most polluting sectors. He found that if the factor endowment differences are controlled for then there is no significant effect of stringency on the trade patterns in these polluting industries. But later on literatures used panel data which also controlled for heterogeneity and this helped to get some results in favor of PHH. For example, Cole and Elliot (2001) used panel regression (both fixed effects and random effects) and found evidence for PHH. Similarly Cole, Elliot and Okubo (2010) also found evidence for PHH by using panel data for 41 Japanese manufacturing sectors. They analyzed the impact of environmental costs on net 11 imports of dirty goods and found that regulation positively affected net imports. Ederington, Levinson and Minier (2003) also used panel data to study the impact of environmental regulation (proxied by US PAOC) on net imports of polluting industries of the US; in their case PHH turned out invalid as the more polluting industries are also least mobile (footloose). This study was somehow a replication of Birdsall and Wheeler (1993), which they carried out for the Latin American countries and got similar results. One drawback of the cross sectional studies is that the estimates are based on observed between-region variations in the model and ignored the unobserved regional heterogeneity. Henderson (1996) recognized this limitation and used panel data to study the impact of air quality regulations on the number of plants from five polluting industries in the US. They found evidence for PHH. 2.4 Endogeneity Problem: The problem of Endogeneity or simultaneity is a serious econometric issue which has been dealt by many studies to improve upon their previous flaws. Although the PHH asserts that the environmental regulation affects exports and FDI flows, it may happen that exports or FDI flows may affect environmental regulations. The logic goes like this. Opening up to trade increases the income and environmental quality being a normal good its demand increases, thus putting pressure on a country’s government for imposing of strict environmental regulations. Trade may also lead to innovation in the process of production and can help countries produce the erstwhile dirty goods with less pollution generating processes. So, to control for this Endogeneity in the data, instrumental variables has been used by several papers. 12 Xing and Kolstad (2001) tried to make an improvement over the previous studies by using instrumental variables and analyzed the impact of environmental regulation on the capital movement of polluting industries for the US. They found significant evidence for PHH since it was observed from the regression results that laxer environmental regulation attracts FDI in dirty industries. The others who also incorporated the control for endogeneity are Cole (2003), Cole and Elliot (2005), Elliot and Shimamoto (2008), Levinson and Taylor (2008), et al. Most of these papers are found to get a significant evidence for PHH. 2.5 Level of Disaggregation: Different studies have used different levels of data either aggregated, i.e country level or state level data or considering an industry level disaggregated data for their analysis. It can be observed from the past empirical literature that generally papers which based their analysis on a higher level of disaggregation witnessed a much improved results in favour of PHH. For example Levinson and Taylor (2003) considered a sector level data, where each sector comprises of several heterogeneous industries. They got a valid PHH as the overcome the aggregation bias present in other literature using aggregate exports. Similarly Cole (2004) also used an industry level data for US and found that the imports of dirty products increase with increase in PAOC which validates the PHH. But Mani, et al. (1996) who used an aggregate industry data state wise for India found no evidence for PHH. So the higher the level of disaggregation considered, the chance of getting a valid PHH increases. 2.6 Type of Trade: Based on the type of trade considered by the studies for their analysis, they can be broadly divided into two sets – studies which considered a bilateral trade framework and studies which considered multi-lateral trade framework. A bilateral trade occurs between two 13 countries and the studies which considered the trade pattern between two countries then it is a bilateral framework. On the other hand papers considering data across several countries are in a multi-lateral trade set up. But the results were mixed. For example Alief rezza (2013) used a multi-lateral trade framework but did not find any evidence for PHH but on the other hand Cole, et al. (2010) also used a multi-lateral framework but they found positive evidence for PHH. Same is the case for studies which considered analysis at bilateral level. 2.7 Model Specification: Different studies have considered different theoretical models to test the presence of PHH. The different theoretical frameworks which have been used are the Heckscher Ohlin Samuelson (HOS) model and the Gravity model. The HOS model followed the comparative advantage principle and is built on David Ricardo’s comparative advantage principle by predicting patterns of specialization based on factor endowments. According to this theory, a labour abundant country will specialize in labour intensive goods while a capital abundant country will specialize in capital intensive goods. Many papers used this framework to find out whether factor endowment differences affect the export and import basket of countries. Here comes the testing of another hypothesis, the factor endowment hypothesis (FEH) which asserts that it is not the differences in pollution policy, but the differences in endowments or technology that determine trade. It predicts that the capital abundant country will specialize in capital intensive goods while the labour abundant country will specialize in labour intensive goods. Since the pollution intensive goods are generally capital intensive, then by the FEH it is the developed countries which will specialize in the dirty goods. The FEH predicts just the opposite to the PHH. In the literature also these two contrasting hypotheses have been tested. Whenever the PHH was found to be invalid, it was concluded that the FEH to be valid. Again, which hypothesis explains the impact of trade on environment better is an empirical 14 question. Kalt (1988), Tobey (1990), and many other studies used the HOS framework to find evidence for PHH. The results were mixed. Some found evidence for PHH while others failed to get any such evidence and concluded that PHH is false. Among the other models used by papers is the gravity model. The gravity model postulates that the bilateral trade flows depends on the economic sizes of the two countries (proxied by GDP) and the distance between the two countries. One modification made by the papers is that they incorporated another factor in the gravity model which is the environmental costs proxied by regulations and taxes. While some used the input output framework to get a detailed picture of the pollution flow across industries. Based on the above segregation of the empirical literature, the variables of the meta-analysis can be constructed in section 4. 3. Meta-regression analysis Meta-analysis was first proposed by Gene Glass (1976) and popularized its name. It is a method of systematic quantitative summary of evidence across empirical studies. It consists of a statistical analysis of a collection of results from individual studies for the purpose of integrating the findings. This type of analysis will improve the precision of an estimate by using all available data. Initially it was used in clinical trials to measure the effectiveness of a particular drug but now it currently enjoys a widespread use in health sciences, psychology, education, marketing and social sciences. Application of meta-analysis in economics began in 1989-90 starting from Stanley and Jarrell (1989) to Jarrell and Stanley (1990), Smith and Kaoru (1990a, 1990b), Walsh et al. (1989, 1990), and Weitzman and Kruse (1990). Several hundred analyses have been prepared in economics, with at least one-third in the area of environmental and resource economics. 15 “Meta-analysis is a body of statistical methods that have been found useful in reviewing and evaluating empirical research results. If a number of independent studies have been conducted on a particular subject, using different data sets and methods, then combining their results can furnish more insight and greater explanatory power than the mere listing of the individual results” (Stanley, 2001). In economics, a technique referred to as meta-regression analysis is employed, which is a form of meta-analysis especially designed to investigate empirical research in economics. Technically speaking a meta-regression analysis consists of a dependent variable and some independent variables. The dependent variable is a summary statistic or “effect size” which is a regression parameter drawn from each study while the independent variables include characteristics of the data, study design, sample size, model specification, econometric methods, and other quality variables. Most of the variables are specified as binary dummies. The objective of meta-analysis is to identify the extent to which the particular choice of methods, design and data affects the reported results of different studies. In short meta- analysis tries to explain the study to study variation or the heterogeneity in effect sizes. This process will also eliminate bias present in the literature thus leading to a reliable conclusion. Also, the aggregation from different independent studies increases statistical power of generalization by isolating the specific effects from the general ones. 3.1 Steps involved in a Meta-analysis: Since Meta-analysis is systematic quantitative review of the literature, this analysis will consist of several steps as Stanley (2001) postulated: (a) Precisely define a question you want to answer, i.e. the objective of doing such analysis in your area. 16 (b) We must include all relevant studies from a standard data base so as to exhaust the literature and make the analysis more reliable. (c) The most important part of the meta-analysis is the choice of a summary statistic and variables determination. After that, reduce the summary statistic to a common and comparable metric. The transformation to a common metric is needed in order to compare, summarize and analyze. (d) The multiple estimates studies are averaged and several independent variables also known as the moderator variables are chosen. (e) Identify individual observations in each study and transcribe the data collected from them, by assigning values and codes into the data set. (f) Conduct a meta-regression analysis by running a logistic regression. (g) Finally interpret the results derived from the regression. (h) A list of complete references should be provided in order to ensure the exhaustiveness of the analysis. 3.2 Limitations of the Method: Although meta-analysis has been very useful to quantitatively summarize the vast empirical literature in various fields, such analyses have some limitations. “Economists reluctant to trust a meta-analysis, which mixes what they might consider to be ‘good’ and ‘bad’ studies, as much as they trust the individual study that they prefer” (Stanley, 2001). The framing up of the summary statistic is difficult. A researcher who publishes a comprehensive single paper on a subject can be underweighted to a researcher who publishes a number of closely related 17 papers. Also in such analyses there is publication bias 5 which arises due to not considering the unpublished works in journals. The publication bias can be considered as an issue for future research. Data heterogeneity, heteroscedasticity, and non-independence were recognized as problems by several economists who prepared early meta-analyses (e.g., Stanley and Jarrell 1989; Smith and Kaoru 1990a, 1990b). The simplest method of using only one estimate from each study might result in a very small sample for the meta-analysis. 6 Another problem faced by the analyst is that of using all available information. Since most of studies in economics provide more than one econometric estimates for a particular effect, estimates of the same study are likely to be correlated. References to literature reviews, bibliographies, relevant journals, electronic databases and search engines are important aspects of a meta-analysis, but few analyses in economics make a serious effort to provide sufficient details on it. It is very difficult to judge the thoroughness of analysis without them. Many a times, the analysts selectively delete observations which make it impossible to tell that how many studies are incorporated in the regressions. So it is necessary to report the data employed at each stage of the analysis. Many analyses miss out in showing the preliminary stage where graphs and summary statistics are required to give an overview of the variables of interest. 4. Meta-analysis of Pollution Haven Hypothesis There has to be a pre-defined question in mind before conducting a meta-analysis. The objective of this paper is to find out the factors which are significantly influencing the outcomes of each study, i.e. to find out the reasons behind different studies getting different 5 Publication bias or “file-drawer problem”, is a form of sample selection bias that arises if studies with statistically weak, insignificant or unusual results tend not to be submitted for publication or less likely to be published. 6 The use of a single estimate is a form of “best evidence synthesis.” 18 outcomes. In the PHH literature heterogeneity is present. This paper make an attempt to figure out the factors which are responsible for increasing or decreasing the chance of getting a valid PHH when considered in the studies. The meta-regression analysis of PHH has been undertaken in the following steps: 4.1 Quantitative Sorting and Data Construction: The most important step in a meta-analysis is to construct data out of the selected relevant empirical papers. This step has to be carefully conducted since any errors committed while framing the data set, will lead to biased or inconclusive results. Each and every variable constructed must be paid full attention and which variables to be constructed are also crucially important. If the construction is done without any specific objective or rationale then problems might be faced while running regressions and also interpreting the results. As has already been discussed, when a meta-regression analysis is conducted, most of the variables would be dummy variables. The dummy variables should be constructed in such a way that the ‘dummy variable trap’ is avoided (i.e., the number of dummies has to be one less than the number of categorical variables). To start with, first the data is sorted into different categories and economic logic behind those categories is also provided. After sorting the data, the whole data set is generated. The descriptive statistics for each variable has been provided. Finally a logistic regression is run to isolate the marginal effect of each factor on the final effect variable. Then the interpretation of the results is provided. First of all the entire data has been classified into two sets of variables: Dependent variable (Summary statistic) and the meta-independent variables (Moderator variables). Since the PHH is measured by two different ways in the literature, it is difficult to consider the t-values as the summary statistic. So, the meta-dependent variable (summary statistics) is constructed as a categorical variable which will be a dummy variable (binary in nature, where 1 denotes 19 that a study get evidence for PHH while 0 denotes that the study failed to get any significant evidence for PHH. Before constructing the moderator variables, the study characteristics are segregated into several categories (Table 1). Table 1: Categorization of study characteristics CATEGORIES CHARACTERISTICS Data Type Whether studies used Cross-section, Time series, or Panel data. Disaggregation Whether studies are based on the overall economy or industry specific. Whether studies considered Bilateral Trade Flows or Multilateral Trade Type Trade Flows. Model Different studies using different models such as Gravity model, Specification Heckscher Ohlin Samuelson model, Input-Output model, and others. Some studies considered per capita or lagged GDP term as their Development explanatory variables to control for the level of development. Some studies controlled for the endogeneity of regulatory stringency by Endogeneity constructing instrumental variables. Stringency Some of the studies using quantitative measures of stringency while Measures some used qualitative measures of stringency. 7 Whether the studies used input based (includes abatement costs) or Stringency output based (includes emission fees or pollution emissions) measures Instruments of stringency. Sample Size This consists of the number of observations in each study. After sorting the data into different categories, the data set is generated by constructing ten dummy variables (including the summary statistics) and one regular variable (sample size). The data set is created from the different categories framed up in the previous table. The variables are taken from the bold categories. For example from the first category of data type, 7 Quantitative measures includes abatement costs, pollution levels, etc which can be measured in value terms while qualitative measures includes some index of stringency or ambient air quality, etc. 20 the panel data has been considered as one of the variables which is depicted in bold. The meta-independent variables are listed in Table 2. Table 2: List of Dummies ASPECTS DUMMIES Dummy variable 1 indicates the study got evidence in favour for PHH, 0 PHH otherwise; PANEL 1 indicates the study used Panel data, 0 otherwise; DISGG 1 indicates the study used disaggregated data set, 0 otherwise; MULTI 1 indicates the study used multilateral trade framework, 0 otherwise; HOS 1 indicates the study used Heckscher Ohlin Samuelson framework, 0 otherwise; 1 indicates the study used Instrumental variable to control endogeneity, 0 IV otherwise; DEV 1 indicates the study controlled for a country’s development, 0 otherwise; QTY 1 indicates the study used quantitative measure of stringency, 0 otherwise; 1 indicates the study used pollution abatement as a measure of environmental PAC regulation, 0 otherwise; FDI 1 indicates the study used FDI as pollution transmission channel, 0 otherwise; SIZE The sample size of a study; Here each paper is considered as one observation. But if a particular paper verified PHH using two different techniques or methodology, then we can have multiple observations from that paper. Total number of relevant papers = 43, Total number of observations = 59. PHH is 21 valid for 29 observations and invalid for 30 observations. In Table 3 an overall description of the data is provided, depicting number of observations available for each variable. Table 3: Data Description 8 VARIABLES OBSERVATIONS PHH [Y/N] Y = 29, N = 30 PANEL [Y/N] Y = 26, N = 33 DISAGG [Y/N] Y = 34, N = 25 MULTI [Y/N] Y = 39, N = 20 HOS [Y/N] Y = 24, N = 35 IV [Y/N] Y = 14, N = 45 DEV [Y/N] Y = 21, N = 38 QTY [Y/N] Y = 38, N = 21 PAC [Y/N] Y = 32. N = 27 FDI [Y/N] Y = 20, N = 39 8 In general, Y denotes ‘Yes’ which means the paper used that methodology, while N denotes ‘No’. However, in case of PHH variable, Y denotes that the paper got evidence for PHH while N denotes that the paper failed to get any evidence for PHH. 22 4.2 Modeling Considerations and Regression Results: Since the dependent variable is a categorical binary variable as well as the independent variables are dummy variables, OLS (Ordinary Least Squares) method of estimation is not applicable in this context. A Logit regression is conducted where the regression coefficients does not have their usual meanings. The regression coefficients does not represent the marginal effects. Marginal effects measure the change in Pi (the probability of success) due to a marginal change in Xi, i.e. how do predicted probabilities change as the independent variable changes. Thus for a categorical variable the marginal effect is given by the difference in the probabilities due to the change of the value of a meta-independent variable from 0 to 1 and is expressed as δPi = β j [Pi (1 − Pi )]. The main objective of this paper is to find out which factors when δX ij introduced in a study of PHH are responsible for increasing the chances of getting a valid PHH. So, the logistic regression will help to achieve this objective. In total four set of regressions are run. The first regression consists of all the meta-independent variables introduced in the data construction. The other three regressions are run by gradually dropping some insignificant variables whose absolute values of z are smaller than one. For example, the first regression is conducted on all variables of the data set. The second regression has three omitted variables – PANEL, HOS, QTY, since these variables have a very small z-value (absolute value of z is less than 1). After dropping these three variables, the results slightly improve. As both MULTI and IV are now significant at 1% level of significant whereas they were initially significant at 5% level. In the third regression FDI is dropped but the results are more or less unchanged. In the final regression the SIZE variable 23 is dropped. It was found that PAC becoming significant and thus there is improvement in the results. The regression results for all the four regressions along with all the marginal effects are reported in Table 4 9. Along with the marginal effects, the z values, sample size, Pseudo R2, LR-χ2 are also provided. (Note that the z values are reported within brackets in table 4). Table 4: LOGIT regression results (Dependent Variable: PHH) Independent Estimation 1 Estimation 2 Estimation 3 Estimation 4 Variables PANEL - 0.04 (-0.16) -- -- -- DISAGG 0.74 *** (5.95) 0.75*** (6.81) 0.73*** (6.53) 0.72*** (6.35) MULTI 0.55 ** (3.10) 0.53*** (3.25) 0.46*** (3.01) 0.48*** (3.21) HOS 0.05 (0.23) -- -- -- IV 0.53** (3.16) 0.54*** (3.62) 0.51*** (3.32) 0.48*** (3.17) DEV -0.29 (- 1.38) -0.32 (- 1.64) -0.32 (- 1.61) -0.36* (- 1.98) QTY 0.19 (0.76) -- -- -- PAC -0.36 (-1.65) -0.29 (- 1.40) -0.31 (- 1.52) -0.34* (- 1.77) FDI -0.24 (-1.05) -0.19 (- 0.92) -- -- SIZE 0.000024 (1.27) 0.000021 (1.15) 0.000017 (0.98) -- Sample size 59 59 59 59 LR χ2 28.43 27.77 26.95 25.74 Pseudo R2 0.3477 0.3396 0.3296 0.3148 * Significant at 10% level ** Significant at 5% level *** Significant at 1% level 9 The results from alternative regressions are reported in Table 4 by following the format adopted in James & Murty (1999). 24 The last regression provides the best result, so only regression 4 is interpreted. Among the insignificant variables, the insignificance of FDI variable implies that measuring PHH using either FDI or Net Exports should not influence the outcome of the study. So getting evidence for PHH or not does not depends on the choice of dependent variables by the studies. Among the other insignificant variables are the HOS, QTY and SIZE. All these results imply that the choice of models (HOS, Gravity model or input-output model) or the choice of stringency measures (quantitative or qualitative measure) or the sample size by a particular study should not alter the paper’s evidence. Regarding the significant variables, there are five such variables. Among them, DISAGG, MULTI and IV has positive marginal effects while DEV and PAC have negative coefficients. The positive sign of the DISAGG variable signifies that studies using a disaggregated dataset are expected to increase their chances of getting evidence for PHH because analyzing an industry level data will help us to capture the heterogeneity within an economy. The pollution trading may be industry specific and at the aggregate level there would be intersectoral smoothing through a bias cancellation process. The MULTI variable also has a positive sign which can be interpreted in the following way. If a paper considers a multilateral trade set up then it is expected to get evidence for PHH rather than using a bilateral framework. This may be because in a bilateral framework when two countries trading between themselves, one country becoming a pollution haven is not always possible since the countries can revolt after some period of time. But in a multilateral set up, many countries participating in trade will have huge trade flows and this will lead one country to become a pollution haven. Moreover, the pollution can be traded more amorphously when the total effect spreads over a larger commodity bundle and a bigger size of trading partners. The IV also has a positive sign because studies which controlled for the pollution regulation endogeneity are expected to be 25 able to isolate the effect of regulatory stringency on pollution transfer more clearly generating results in favour of PHH. While the negative signs implies that papers which controlled for the level of development and used PAC as a proxy for stringency measure is expected to decrease the chances of getting a valid PHH. This is depicting a valid result, since if we increase the PAC by one unit, it is expected that the pollution havens will not be witnessed in that country. Again, higher the level of development, the higher is the demand for clean environment and thus lesser is the chance of getting a valid PHH. But there is a problem when considering the marginal effects as probabilities. In regression models we generally want a measure of the unique effect of the independent variables on the dependent variable. In the logistic regression the marginal effects measures the effect of the independent variables on the likelihood on the categorical dependent variable having a specific value through probability. But the effect is not constant since the effect of X (independent variable) on the probability of Y (dependent variable) has different values depending on the value of X. So, to measure the effect in a single number, the odds ratio is the appropriate measure. Odds ratio is the measure of effect size or the strength of association between two binary variable. The odds ratio gets a statistical advantage compared to the probabilities since it represents the constant effect of X on the likelihood of occurrence of Y. The odds ratios of all the five significant variables are reported in table 5. 26  Table 5: Reporting ODDS Ratio X VARIABLES ODDS RATIO Z VALUE P VALUE DISAGG 44.01*** 3.67 0.00 MULTI 9.02 *** 2.56 0.01 IV 9.13*** 2.47 0.01 DEV 0.21* -1.73 0.08 PAC 0.24* -1.63 0.09 * Significant at 10% level ** Significant at 5% level *** Significant at 1% level From the above table, it can be seen that the DISAGG variable has an odd ratio of 44.01 which signifies that by using a disaggregate data set the chances of getting PHH valid will increase by 44 times which is very high. As well as using a multilateral framework and controlling for policy endogeneity (denoted by IV) will increase the chances of getting a valid PHH by almost 9 times. On the other hand for DEV and PAC the odds ratio is very less which signifies that if either of the variables are used as control variables then the chance of not getting PHH will decrease by approximately 5 times. 10 So, the results of the odds ratio provides similar interpretation as the marginal effects in table 4. 10 If odds ratio = 0.21 then odds against is 1/0.21 which is approximately 5. 27 5. Conclusion From the meta-analysis, it can be concluded that though the empirical literature have been heterogeneous in terms of different variables and methods used, some of the factors have been irrelevant in influencing the outcomes of the studies. It is observed that a particular study outcome does not change irrespective of the study using panel or cross section data, or using a HOS model or gravity model or an input output model, or the sample size considered by each study. Most importantly, the study outcome is unaffected by the choice of dependent variable, i.e. whether they use FDI as a pollution transmission channel or use Net exports/imports it is insignificant. The choice of methodological variables is crucially important (e.g. use of an industry level data, use of instrumental variables, use of multilateral trade framework, use of development indicators and use of PAC as an instrument of stringency). So, the variables can be categorized into two sets of factors – specific factors and systematic factors. The specific factors include those factors which are specific to any context of study. For example the specific factors in this study are the PANEL, HOS, QTY, FDI and SIZE. These factors are observed to be effective in getting a significant PHH effect only for some specific countries or for some particular period of time. While the systematic factors include the significant variables which is expected to influence the outcome of the studies testing PHH. The systematic factors are DISAGG, MULTI, IV, DEV and PAC. Intuitively speaking, if these factors are considered by any particular study testing PHH, then it is expected that the chances of getting PHH valid will be much higher than ignoring these factors. The above results are consistent with the PHH and the findings of the previous studies. The negative sign before PAC and DEV indicates that the countries with higher value of PAC and high levels of development are not pollution havens, which is consistent with the pollution 28 haven debate. But since this is the very first attempt to conduct a meta-analysis on Pollution Haven Hypothesis, there is an ample scope for further improvement. The literature can be more exhausted and in future an effort can be made to expand the data set by including the other papers which this paper has excluded. For that the researcher must try to set up a new structure to incorporate the omitted studies. Another aspect the future researcher can look for is the impact of the journal (in which the paper has been published) on the summary statistics. This can be analyzed by using the journal impact factors provided in the link: http://ideas.repec.org/top/top.journals.simple.html. Since the literature contains varied degree of methodologies and frameworks it is quite difficult to include all the empirical literature in this analysis. So, in future an attempt can be made to consider the complex vast literature in a more detailed data set and draw some more useful conclusions. 29 CATEGORIZED BIBLIOGRAPHY (references used in the Meta-analysis): Abimanyu, A. (1996): Impact of Free Trade on Industrial Pollution: Do Pollution Havens Exist? ASSEAN Economic Bulletin 13 (1), 39-51. Aliyu, and Aminu M. (2005): Foreign Direct Investment and the Environment: Pollution Haven Hypothesis Revisited, In Eighth Annual Conference on Global Economic Analysis, Lüeck, Germany. Antweiler, W., Copeland, B. R., and Taylor, M. S. (2001): Is Free Trade Good for the Environment? American Economic Review 91, 877-908. Azhar, and Elliot (2007): Trade and Specialization in Pollution Intensive Industries: North– South Evidence, International Economic Journal 21 (3), 361-380. Busse, M. (2004): Trade, Environmental Regulations and the World Trade Organization: New Empirical Evidence, World Bank Policy Research, Working Paper No. 3361. Cavlovic, et al. (2000): A Meta-analysis of Environmental Kuznets Curve Studies, Agricultural and Resource Economics Review 29, 32-42. Chung, S. (2014): Environmental regulation and foreign direct investment: Evidence from South Korea, Journal of Development Economics 108, 222-236. Cole, M. A. and R. J. R. Elliot (2005): FDI and the Capital Intensity of “Dirty” Sectors: A Missing Piece of the Pollution Haven Puzzle, Review of Development Economics 9(4), 530- 548. Cole, M. A. and R.J.R. Elliot (2003): Do Environmental Regulations Influence Trade Patterns? Testing Old and New Trade Theories, The World Economy, 1163-1186. 30 Cole, M. A. and T. Okubo (2010): Trade, Environmental Regulations and Industrial Mobility: An Industry-Level Study of Japan, Ecological Economics 69, 1995-2002. Cole, M. A. (2004): US environmental load displacement: examining consumption, regulations and the role of NAFTA, Ecological Economics 48, 439-450. Dam, L. and B. Scholtens (2008): Environmental regulation and MNEs location: Does CSR matter? Ecological Economics 67, 55-65. Dinda, S. (2006): Globalization and Environment: Can Pollution Haven Hypothesis alone explain the impact of Globalization on Environment? mimeo, S.R. Fatepuria College, Beldanga, Murshidabad, WB, India. Ederington, J. and J. Minier (2001): Is Environmental Policy a Secondary Trade Barrier? An Empirical Analysis, Canadian Journal of Economics 36 (2003), 137-154. Ederington, J., A. Levinson, and J. Minier (2005): Footloose and Pollution-Free, Review of Economics and Statistics 87 (1), 92–9. Ederington, J., A. Levinson, and J. Minier (2004): Trade Liberalization and Pollution Havens, Advances in Economic Policy and Analysis 4 (2), Article 6. Elliot, R. J. R. and K. Shimamoto (2008): Are ASEAN Countries Havens for Japanese Pollution-Intensive Industry? The World Economy 31 (2), 236–254. Eskeland, G. S. and A. E. Harrison (2003): Moving To Greener Pastures? Multinationals and the Pollution Haven Hypothesis, Journal of Development Economics 70, 1-23. Frankel, J. A. and A. K. Rose (2005): Is Trade Good or Bad for the Environment? Sorting out the Causality, Review of Economics and Statistics 87 (1), 85-91. 31 Fredriksson, P. G., J. A. List and D. A. Millimet (2003): Bureaucratic corruption, environmental policy and inbound US FDI: theory and evidence, Journal of Public Economics 87, 1407–1430. Harris, M. N. et al. (2002): Modelling the Impact of Environmental Regulations on Bilateral Trade Flows: OECD, 1990–1996, World Economy 25 (3), 387–405. Jug, J and D. Mirza (2005): Environmental regulations in gravity equations: evidence from Europe, World Economics 28 (11), 1591–1615. Kellenberg, D. K. (2009): An empirical investigation of the pollution haven effect with strategic environment and trade policy, Journal of International Economics 78, 242–255. Keller, W. and A. Levinson (2002): Pollution Abatement Costs and Foreign Direct Investment Inflows to U.S. States, The Review of Economics and Statistics 84 (4), 691-703. Levinson, A. (1996): Environmental Regulations and Manufacturers’ Location Choices: Evidence from the Census of Manufactures, Journal of Public Economics 62, 5-29. Levinson, A. and M. S. Taylor (2008): Unmasking the Pollution Haven Effect, International Economic Review 49, 223-254. Li, H. et al. (2007): Economic growth and environmental quality: a meta-analysis of environmental Kuznets curve studies, Economics Bulletin 17 (5), 1-11. List, J. A. and C. Y. Co (2000): The Effects of Environmental Regulations on Foreign Direct Investment, Journal of Environmental Economics and Management 40, 1-20. 32 Manderson, E. and R. Kneller (2012): Environmental Regulations, Outward FDI and Heterogeneous Firms: Are Countries Used as Pollution Havens? Environmental Resource Economics 51, 317-352. Mani, M. et al. (1996): Does environmental regulation matter: Determinants of the location of new manufacturing plants in India in 1994, World Bank Policy Research, Working Paper No. 1718. Marconi, D. (2010): Environmental regulation and revealed comparative advantages in Europe: is China a pollution haven? Banca d’Italia Occasional Paper No. 67. McConnell, V. D. and R. M. Schwab (1990): The Impact of Environmental Regulation on Industry Location Decisions: The Motor Vehicle Industry, Land Economics 66 (1), 67-81. Millimet, D. L. and J. Roy (2011): Three New Empirical Tests of the Pollution Haven Hypothesis When Environmental Regulation is Endogenous, The Institute for the Study of Labor Discussion Paper No. 5911. Mookerjee (2005): A Meta-analysis of the Export Growth Hypothesis, Economics Letters 91, 395-401. Mulatu, A., et al. (2004): Environmental Regulation and International Trade, Tinbergen Institute Discussion Paper, 20/3. Mulatu, A. et al. (2010): Environmental Regulation and Industry Location in Europe, Environmental and Resource Economics 45, 459-479. Rezza, A.A. (2013): FDI and pollution havens: Evidence from the Norwegian manufacturing sector, Ecological Economics 90, 140-149. 33 Santis, R. D. (2011): Impact of Environmental Regulations on Trade in the main EU Countries: Conflict or Synergy? European Network of Economic Policy Research Institutes Working Paper No. 56. Sawhney A. and R. Rastogi (2014): Is India Specializing in Polluting Industries? Evidence from US-India Bilateral Trade, The World Economy, 10.1111/twec.12164. Stanley, T. D. (2001): Wheat From Chaff: Meta-Analysis As Quantitative Literature Review, Journal of Economic Perspectives 15 (3), 131–150. Kirkpatrick, C and K. Shimamoto (2008): The Effect of Environmental Regulation on the Locational Choice of Japanese Foreign Direct Investment, Applied Economics 40, 1399–409. Tang, J. P. (2010): Pollution Havens and the Trade in Toxic Chemicals: Evidence From U.S. Trade Flows, Discussion Papers, U.S. Census Bureau, Center for Economic Studies. Van Beers, C. and Jeroen C. J. M. Van Den Bergh (1997): An Empirical Multi-Country Analysis of the Impact of Environmental Regulations on Foreign Trade Flows, Kyklos 50 (1), 29–46. Van Beers, C. and Jeroen C. J. M. Van Den Bergh (2010): The Impact of Environmental Policy on Foreign Trade, Tinbergen Institute Discussion Paper, 69/3. Waldkirch, A. and M. Gopinath (2008): Pollution Control and Foreign Direct Investment in Mexico: An Industry-Level Analysis, Environmental Resource Economics 41, 289-313. Xing, Y. and C. D. Kolstad (2002): Do Lax Environmental Regulations Attract Foreign Investment? Environmental and Resources Economics 21, 1-22. 34 Xu, X. (2000): International Trade and Environmental Regulation: Time Series Evidence and Cross Section Test, Environmental and Resource Economics 17, 233–257. NON-CATEGORIZED BIBLIOGRAPHY (Additional references used in the paper): Cole, M. A. (2004): Trade, the pollution haven hypothesis and the environmental Kuznets curve: examining the linkages, Ecological Economics 48, 71– 81. Cole M. A. and R. J. R. Elliot (2005): FDI and the Capital Intensity of “Dirty” Sectors: A Missing Piece of the Pollution Haven Puzzle, Review of Development Economics 9 (4), 530- 548. Copeland Brian R., and M. Scott Taylor (1994): North-South Trade and the Environment, The Quarterly Journal of Economics, August (1994). Dinda, S. (2004): Environmental Kuznets Curve Hypothesis: A Survey, Ecological Economics 49, 431– 455. Grossman, G. M. and A. B. Krueger (1993): Environmental impact of a North-American free trade agreement, in The U.s.-Mexico Free Trade Agreement, P. Garber, Ed. (MIT Press, Cambridge, Mass). James, A.J. and M.N. Murty (1999): Water Pollution Abatement: A Taxes-Standards Approach for Indian Industry, in Economics of Water Pollution: The Indian Experience, M.N. Murty, A.J. James and S. Misra (eds.) (OUP, India). Levinson, A. and M. S. Taylor (2008): Unmasking the Pollution Haven Effect, International Economic Review 49, 223-254. 35 Tobey, J. A. (1990): The Effects of Domestic Environmental Policies on Patterns of World Trade: An Empirical Test, Kyklos 43, 191-209. Zhang, Y. (2011): Scale, Technique and Composition Effects in Trade – Related Carbon Emissions in China, Environ Resource Econ 51, 371-389. 36