Endogenous Pollution Havens: Does FDI Influence Environmental Regulations?* 

Authors


  • * 

    We would like to thank three helpful referees, Sangeeta Bansal, Jayasri Dutta, Angeliki Kourelis, John List, Daniel Millimet and participants at presentations at Rice University, SMU, and the EAERE meetings in Budapest for helpful comments and discussions, and Jakob Svensson for some of the data. This paper was started while Fredriksson visited the Department of Economics, University of Gothenburg, whose hospitality is greatly appreciated. Cole and Elliott gratefully acknowledge the support of ESRC grant number RES-000-22-0016 and Leverhulme Trust grant number F/00094/AG. Fredriksson gratefully acknowledges support from the Malmsten Foundation and the University of Gothenburg Jubilee Fund (Elof Hansson's Foundation Gift), and travel funds from SMU. The usual disclaimers apply.

Abstract

We suggest a novel perspective on the relationship between the stringency of environmental policies and foreign direct investment (FDI). We develop a political economy model with imperfect product market competition where local and foreign firms jointly lobby the local government for a favorable pollution tax. FDI is found to affect environmental policy, and the effect is conditional on the local government's degree of corruptibility. If the degree of corruptibility is sufficiently high (low), FDI leads to less (more) stringent environmental policy, and FDI thus contributes to (mitigates) the creation of a pollution haven. Our empirical results using panel data from 33 countries support the predictions of the model.

I. Introduction

Opponents to international trade and investment flows frequently argue that the presence of foreign-owned firms (multinationals) causes local environmental regulations to become suboptimally weak, and “pollution havens” to emerge; see Newell (2001). According to UNCTAD (2001), 60,000 multinationals have 800,000 foreign affiliates worldwide, yet the literature contains surprisingly little formal work on investigating their impact on local environmental policies; see Vogel (2000) for some anecdotal evidence. Instead, the related theoretical and empirical literature has focused on the effects of local environmental regulations on investment flows; see, e.g., List and Co (2000) and Keller and Levinson (2002). In this paper, we seek to fill this gap in the literature. We ask the following question: what are the effects on local environmental policy, and quality, of foreign direct investment?

In order to address this question, we begin by setting up a simple model of government environmental policymaking that sheds light on the political forces of interest. The theory applies to all cases of direct investment by a parent firm located outside a jurisdiction, where the subsidiary repatriates the profits to its home jurisdiction. Local politicians are assumed to value bribes (political contributions) and aggregate social welfare; see, for example, Grossman and Helpman (1994) and Aidt (1998). We view the government's weight on campaign contributions relative to social welfare as a useful measure of the corruptibility of local policymakers.1,2 The imperfectly competitive local goods market contains both locally owned firms and foreign subsidiaries; see Grossman and Helpman (1996). All firms produce exclusively for the local market.3

In a three-stage game, all domestic and foreign firms (with local production facilities) exogenously form a lobby group, which first offers a prospective bribe (political contribution) schedule to the domestic government.4 We ignore free-riding problems among firms, as in Olson (1965), and lobby group formation is assumed to occur exogenously, as in most of the relevant literature. In the second stage, the local government sets its optimal policy, and collects the associated bribe. In the third stage, the firms set output and abatement levels.

We find that the establishment of an additional foreign plant (given the number of domestic firms) has two main effects on local environmental policymaking. First, foreign direct investment leads to a greater output level produced for the local market. Thus, more is at stake in the policy outcome because the pollution tax applies to a greater output level. This increases the size of the bribe offered by the lobby for a lower pollution tax. This “bribery effect” of foreign investment leads to a lower pollution tax. Second, in an imperfectly competitive market, the government has an incentive to lower the pollution tax below the first-best level (equal to marginal damage) in order to stimulate output and raise consumer surplus; see Katsoulacos and Xepapadeas (1995), henceforth KX. An increase in the number of firms increases the level of competition and therefore reduces the government's incentive to lower the pollution tax. This “welfare effect” of foreign direct investment leads to a higher pollution tax. The net effect of an additional foreign subsidiary is conditional on the government's weight on the “bribery effect” relative to the “welfare effect”, i.e., the degree of corruptibility. We find that foreign direct investment raises (reduces) local environmental policy stringency when the degree of government corruptibility is relatively low (high).5

The empirical analysis lends support to our theoretical predictions. Using a panel of 33 developed and developing countries for the years 1982–1992, we find that inward FDI has a positive impact on the stringency of environmental regulations when the level of corruptibility is low. At higher levels of corruptibility this impact is lessened and eventually becomes negative. This is consistent with the “bribery effect” dominating (being dominated by) the “welfare effect” of FDI for high (low) levels of corruptibility. The sensitivity analysis reveals that our empirical findings are robust across a range of different specifications.

We believe our results may have implications for future empirical investigations of the effects of environmental policies on new plant locations by foreign firms. Many such studies to date have found few robust negative effects on foreign direct investment; see, e.g., the survey by Jaffe, Peterson, Portney and Stavins (1995). This may be attributable to the fact that most empirical studies have treated environmental policy as exogenous. If foreign firms’ rent-seeking activities affect environmental policy, any regression model trying to discern the impact of environmental policy on foreign investment has to take into account that both variables are endogenous. Only recently has the literature begun to recognize this problem. In particular, the effect of foreign direct investment on environmental policies has been ignored. Our study provides a new determinant of environmental policy. The main value added of the current paper is thus to recognize and show that FDI affects the environmental policy formation process. In particular, the paper details how the effect of FDI is conditional on the degree of government corruptibility. The previous literature linking economic liberalization, corruption and environmental policy has either studied the effects of trade liberalization (openness) on environmental policies, as in Damania et al. (2003), or the effects of environmental regulations on FDI flows, as in Fredriksson, List and Millimet (2003). Damania et al. (2003) find that greater trade openness leads to stricter (weaker) environmental policies where government corruptibility is relatively high (low).6Fredriksson et al. (2003) provide an empirical study of the effect of environmental policy on FDI into the U.S. that treats environmental policy endogenously, and show that stricter state environmental policies negatively affect the inflow of FDI in several industry sectors.7

Our results also indicate that warnings about the negative effects of foreign direct investment should be taken seriously, in particular when the degree of government corruptibility enables officials to sell policy favors to polluting firms. The model suggests that in such countries, the feared pollution havens may be more likely to emerge as a result of foreign direct investment. On the other hand, the results are encouraging for less corrupt countries. Our results reinforce the need to reduce the level of government corruption (corruptibility) in many countries.

The paper is organized as follows. The model is set up in Section II, where the theoretical predictions are also derived. Section III reports the empirical analysis. Section IV concludes.

II. The Model

Consider a small economy where production causes local pollution damage, s. A continuum of N D domestic firms and N F foreign subsidiaries are producing and competing (in quantities) in an imperfectly competitive local market, where N D + N F = N; see Grossman and Helpman (1996). We take the number of active firms in the market as given, and assume that the market is supplied exclusively by (for simplicity) identical firms located within the jurisdiction's borders. The producers are shielded from the world market by, e.g., high transportation costs, and for simplicity no trade takes place.8 There are four types of agents: consumers, domestic producers, foreign producers and the government. The population of consumers is normalized to 1. The utility of the representative consumer equals U = u(Q) − s, where Q is consumption of the polluting good with price p = a − Q, where a > 0 reflects the size of the local market. u(Q) is a concave, twice differentiable subutility function.

Local production results in local pollution, and the government controls emissions by levying an emissions tax, tTR+, per unit of pollution. Output of firm i is given by qi. Following KX, the gross profit function of firm i is given by

image(1)

where p(Q) is the inverse demand function, Q = Σiqi = Nqi, ei(qi, wi) = cqi +gwi is the cost function, where wi is abatement expenses, and the parameter g represents the marginal abatement cost.9 The pollution damage function equals inline image, which is increasing (and linear) in q and decreasing and concave in w(as in KX), and where si ≥ 0 is required. Thus, pollution abatement exhibits diminishing returns. In our view, this functional form has the advantage of being tractable. It implies that marginal damage from pollution equals unity (since the population of consumers is normalized to unity), and that firms may reduce pollution damage by both abatement and output reduction. F is the fixed cost of production, and c, v, β and γ are positive parameters. Applying the implicit function theorem to the FOCs of (1) yields dqi/dt < 0 and dwi/dt > 0. In the Nash equilibrium, firm i's output and abatement levels, given the pollution tax, equal qi =(a − c − vt)/(1 + N ) and wi = (βγt/g)1/1+γ.

Foreign firms’ profits are assumed fully repatriated to the firms’ home jurisdictions, and are consequently not part of the government's social welfare function. The consumers’ aggregate welfare equals:

image

i.e., the sum of consumer surplus, pollution tax revenues (redistributed equally to all consumers), less the damage from pollution, respectively (and where x is the integrating variable).

In the first stage (out of three), all domestic and foreign firms active in the local economy join the firm lobby. The lobby offers a prospective bribe schedule C(t) to the government. Thus, the promised bribe is contingent on the government's choice of pollution tax policy. In the second stage, the government selects its environmental policy and collects the corresponding bribes from the lobbies. In the third stage, the firms set output and abatement expenditure levels.

Since the organized lobby contains a negligible number of individuals, it ignores consumer surplus and revenues, and thus has a utility function simply given by V(t) ≡ (N D + N F)π. We assume that the government maximizes a weighted sum of bribes (political contributions) received and aggregate (gross-of-bribes) social welfare, given by G(t) = C(t) + αW A(t), where α > 0 is the weight given by the government to aggregate social welfare relative to bribes. The weight α is commonly used as a measure of the level of government corruptibility; see, e.g., Damania et al. (2003).10 Only domestic firms’ profits are included in aggregate social welfare, W A(t) = W CO(t) + N Dπ.

Following Bernheim and Whinston (1986), Grossman and Helpman (1994) and Fredriksson (1997), the subgame-perfect Nash equilibrium pollution tax, t*, can be shown to satisfy the equilibrium characterization, ∂V(t*)/∂t + α(∂W A(t*)/∂t) = 0. By taking the partial derivatives of the expressions for the lobby groups’ welfare as well as aggregate social welfare, and substituting the result into the equilibrium characterization, we find:

image(2)

where term A is the effect of the lobby group on the equilibrium tax, and term B is the government's consideration of consumers’ welfare. In this model, the pollution tax is subject to several downward pressures. These contribute to reducing the tax below the pollution tax set by a welfare-maximizing government when the output market is perfectly competitive (the first-best pollution tax). First, the lobby group bids for a lower pollution tax. Second, with imperfect competition in the output market, the government lowers the pollution tax to raise consumer surplus; see Barnett (1980) and KX. Since term A in (2) is unambiguously negative, term B is positive. We make the following assumption on the equilibrium pollution tax:

Assumption 1.  In equilibrium, t* < 1.

Assumption 1 simplifies the discussion below, but does not drive the results. It implies that t* is set below marginal social damage from pollution.11 We can now show:

Proposition 1. In the political equilibrium, the pollution tax increases(decreases)with the number of foreign firms if the degree of corruptibility is sufficiently low(high).

Proof:  Differentiation of (2) yields

image(3)

where D is the SOC of (2). We assume D < 0. Term A is negative, and term B is positive (Assumption 1). Thus,

image

The effect of the (exogenous) establishment of an additional foreign subsidiary (given the number of domestic firms) on environmental policy depends on the level of corruptibility of the local policymakers. The net impact is determined by two main effects. First, an increase in the number of firms active in the domestic (output and political) markets raises the political pressure for a lower pollution tax. This is because the foreign firms’ output level increases, so that the stakes involved regarding pollution taxation increase for the lobby group. This results in a less stringent policy (term A). We denote this a “bribery effect” of foreign direct investment. Second, an increase in the number of active firms raises the level of product market competition. This reduces the government's incentive to lower the stringency of the pollution tax in order to keep output and consumer surplus high, a “welfare effect” of foreign direct investment. This causes the policy stringency to increase (term B).

When the degree of corruptibility is high (α is low), term A dominates and the additional foreign firm causes a decrease in environmental policy stringency. When the degree of corruptibility is low (α is high), term B dominates and the additional foreign firm causes an increase in the tax policy stringency.12

III. Empirical Analysis

Specification and Methodology

We now seek to test the main implication generated by our model in order to shed new light on the debate surrounding the pollution haven hypothesis. In particular, we analyze the relationship between environmental policy, foreign direct investment (FDI) and corruptibility, formulated in Proposition 1. We do this by estimating the following equation:

image(4)

where REGSit denotes environmental regulatory stringency in country i in year t, αi is a time-invariant country fixed effect, γt is a location-invariant time fixed effect, X is our vector of independent variables, and ɛit is the error term. Equation (4) is estimated using both fixed and random effects specifications.

Our model predicts that the impact of foreign direct investment on environmental regulations in country i is conditional on country i's level of corruptibility. In particular, FDI should have a positive (negative) effect on the stringency of environmental regulations when corruptibility is low (high)(Proposition 1). In order to capture this effect, vector X contains a continuous measure of inward FDI (FDI ), a measure of the degree of corruptibility (CORRUPT ), and the interaction term FDI × CORRUPT. We therefore expect the estimated coefficient on FDI to be positive, while the coefficient on FDI × CORRUPT is expected to be negative.

We also include a number of control variables. We expect the demand for environmental quality to increase with per capita income (GDP). The urban population share (URBANPOPsh) controls for the greater exposure to industrial pollution suffered by citizens in more urbanized countries, and should have a positive effect on regulatory stringency as a result of greater political pressure. However, the marginal effect of both GDP and URBANPOPsh may be diminishing, consistent with the literature on the environmental Kuznets curve; see, for example, Millimet, List and Stengos (2003). Quadratic terms are therefore also included. Conversely, MANUFsh represents the pressure from workers in the manufacturing sector for lower regulations to protect jobs in the face of increased competition, and is expected to have a negative impact on the stringency of environmental regulations. Potters and Sloof (1996) suggest that the effect of lobby group size may be non-monotonic; we therefore include a quadratic term for MANUFsh. It may alternatively be the case that MANUFsh captures the degree to which an economy consists of pollution-intensive manufacturing, in which case MANUFsh may be positively signed. Since the impact of these explanatory variables on environmental regulations is unlikely to be immediate, all explanatory variables are lagged by one year.13

While equation (4) expresses environmental regulations as a function of FDI, it is assumed in the pollution haven literature that the causality between these two variables runs in the opposite direction. To allow for this potential endogeneity, we instrument FDI using two-stage least squares (2SLS). To be suitable for use as an instrument, a variable must be correlated with FDI, yet exogenous with regard to REGS—requirements which considerably limit the choice of variables. In the first instance we use three such instruments. Two instruments capture the degree of public infrastructure within the host country, a factor likely to influence a potential investor's decision regarding where to invest; see Wheeler and Mody (1992). The two measures of public infrastructure that we use are the number of telephone mainlines (per 1,000 people) and the number of television sets (per 1,000 people), where the latter variable captures the extent of a country's telecommunications network, electricity supplies, and so on. The third instrument is the economically active population, a variable which captures the size of the host country. It is a well-known empirical observation that small (large) countries tend to be more (less) open to both international trade and investment, and hence typically have a greater (smaller) share of trade and FDI in GDP; see, for example, Streeten (1993). Finally, the exogenous variables from equation (4) are also included as instruments. To minimize any possible causality moving from FDI to our measures of public infrastructure, the latter variables enter the first-stage regression in lagged form. Since the explanatory variables in our second-stage regression (equation (4)) are in lagged form, these also enter the first-stage regressions in lagged form, thereby again minimizing any possible causal link from FDI. We use a Sargan test of over-identifying restrictions to assess the validity of our instruments and also report F-tests of joint instrument significance. Note that our sensitivity analysis examines the robustness of our results to the choice of instruments by testing two alternative instruments.

Variable Definitions and Data

We have data for 33 countries for the period 1982–1992; 13 are OECD and 20 are developing countries. Table A1 in the Appendix provides all data sources and lists the countries in the sample.

Our measure of environmental regulations is grams of lead content per gallon of gasoline, previously used by, for example, Hilton and Levinson (1998) and Damania et al. (2003).14 This variable has a number of features that make it desirable as a measure of industry environmental regulations, even though it applies primarily to the transportation sector. In particular, lead content in gasoline has both cross-section (developed and developing countries) and time-series coverage, thereby making it arguably unique among suitable measures of environmental regulations. The other attractive features include: (i) the content of lead in gasoline is (almost) entirely a policy decision and is unlikely to be influenced by other factors; (ii) lead emissions are a particularly damaging local air pollutant with significant health implications. As a result, the control of such emissions is often an early environmental objective during a country's development.15

Damania et al. (2003) report that the lead content of fuel has a statistically significant negative correlation with three other measures of industry environmental regulations. The first such measure was constructed using forecasting techniques from Eliste and Fredriksson's (2002) single-year measure of environmental stringency and has a correlation of −0.78 with lead content in gasoline. This measure is derived from the United Nations Conference on Environment and Development (UNCED, 1992) country reports which cover numerous aspects of the environmental regulatory framework such as legislation, control mechanisms and enforcement. The second measure of environmental stringency considered by Damania et al. (2003) is public expenditure on environmental R&D as a proportion of GDP. This variable is available for 1982–1992, although for OECD countries only, and has a negative, statistically significant correlation of −0.38 with lead content. Third, Damania et al. (2003) consider per capita membership of environmental organizations. This variable may be loosely correlated with the stringency of a country's environmental regulations, but is only available for nine countries. It has a statistically significant correlation of −0.45 with lead content.16 We believe these measures to be inferior to lead content in gasoline as a measure of national environmental regulations due to their limited cross-country coverage, their potentially weak relationship with national environmental stringency and, in the case of the measure calculated using forecasting techniques, the reliance on a single-year measure of regulations. However, the fact that all three measures possess statistically significant correlations with lead content suggests that the latter variable captures national environmental regulations. It should also be noted that we create our REGS variable by multiplying the lead content in gasoline variable by −1. Thus, an increase in REGS represents an increase in the stringency of regulations (i.e., a decrease in lead content).

While our theory discusses the effect of the number of foreign firms, empirical measures of this variable are unavailable (to our knowledge). We use two different measures of inward FDI: (i) FDI stocks and (ii) FDI flows; see UNCTAD (2001). Both measures are scaled by aggregate GDP.17 In our view, these two measures adequately capture the political effects discussed in the theory. While the FDI flow variable captures the political economy effects of new investments on REGS, the FDI stock variable may better capture the overall effect of foreign investment. The stock variable will also partially capture the effects of FDI flows if there is a lag between the investment made and its political effects although, as already mentioned, explanatory variables are lagged to capture this political inertia.

Our measure of corruptibility is the “government honesty” variable reported by the International Country Risk Guide; see Knack and Keefer (1995). This variable measures the extent to which “high government officials are likely to demand special payments” and takes the form of an index between 0 and 6, where 0 represents the least government honesty, and 6 the most. For ease of interpretation, we subtract this index from 6 so that it forms CORRUPT, a measure of the lack of honesty, i.e., the degree of corruptibility (corruption). Thus, a higher value of CORRUPT represents a higher level of corruptibility. The control variables are defined in the Appendix. Table 1 provides summary statistics.

Table 1. Summary statistics
 MeanS.D.MinimumMaximum
CORRUPT2.551.570.006.00
REGS−1.780.98−3.980.00
FDI Stock8.647.250.3133.20
FDI Flow0.820.950.017.92
GDP7.839.650.0941.35
URBANPOPsh(%)55.2825.3011.0097.00
MANUFsh(%)19.617.153.6134.56
TOTAL POP. (mn)67.48134.408.41882.30
ECON. ACTIVE POP. (mn)40.0979.044.52525.00
PHONE(per 1,000 people)148.09168.021.10564.95
TV(per 1,000 people)196.36184.651.02627.89
INFLATION57.44275.24−9.813,079.81

Results

Table 2 reports our fixed effects (FE) estimates using both stock and flow measures of FDI. While Models (1) and (3) report the OLS results, Models (2) and (4) report the IV (2SLS) estimates. A Hausman specification test indicated that the effects are correlated with the independent variables, implying that the random effects results are inconsistent. For this reason, we do not report the random effects results here (they are available from the authors on request). We found no evidence of first-order autocorrelation. A DWH test indicates that OLS yields inconsistent estimates. Thus, our focus will be on the IV results, but we report OLS estimates for completeness. A Sargan test of over-identifying restrictions fails to reject the null that our IV equations are properly specified for the FE results. This would suggest that our instruments are valid. An F-test also indicates that the instruments are jointly significant in the first-stage estimations. Table A2 in the Appendix reports these first-stage results.

Table 2. Fixed effects results for FDI stocks and flows; dependent variable: REGS (environmental regulations)
 (1)(2)(3)(4)
 FDI stock
FE, OLS
FDI stock
FE, IV
FDI flow
FE, OLS
FDI flow
FE, IV
  1. Notes:Absolute value of t-statistics in parentheses. *Significant at the 10% level;**significant at the 5% level;***significant at the 1% level. Marginal effects are calculated at the sample means of CORRUPT; s.d. = standard deviation.

FDIt−10.049***0.26***0.0970.65***
(4.2)(3.5)(1.1)(3.8)
CORRUPTt−10.21***0.16**0.12***0.093**
(3.8)(2.5)(2.7)(2.0)
(FDICORRUPT)t−1−0.017***−0.028***−0.031−0.11**
(4.6)(4.5)(1.2)(2.4)
GDPt−1−0.0380.11*0.0210.056
(0.5)(1.6)(0.3)(0.8)
inline image−0.0012−0.0034***−0.0019**−0.0020**
(1.3)(3.6)(2.1)(2.2)
URBANPOPsht−1−0.100.035−0.14*−0.20***
(1.4)(0.4)(1.9)(4.5)
inline image0.00097*0.000490.0012**0.0015***
(1.8)(0.08)(2.2)(3.8)
MANUFsht−10.0320.17**0.017−0.0062
(0.7)(2.2)(0.4)(0.2)
inline image−0.0017−0.0061***−0.0170.00016
(1.5)(2.8)(1.1)(0.2)
Observations319319319319
R20.420.430.390.36
F-test FDI (p-value)21.823.21.3315.0
(0.000)(0.000)(0.51)(0.000)
F-test on IVs (p-value) 60.3 69.7
 (0.000) (0.000)
Sargan test (p-value) 4.05 2.91
 (0.13) (0.23)
DWH endogeneity test (p-value)94.8 63.6 
(0.001) (0.02) 
REGS/∂FDI0.00570.190.0170.36
(∂REGS/∂FDI)s.d.0.0411.370.0160.34

The empirical findings in Table 2 lend support to our theory's predictions. FDI has a positive effect on REGS in all models, with statistical significance at conventional levels in three of them. Moreover, the interaction variable FDI × CORRUPT is negative in all models and significant in three, suggesting that the effect of FDI on REGS is conditional on CORRUPT. The F-test on the two FDI terms (which restricts their coefficients to zero) is highly significant in three models, including both of the favored 2SLS models. In order to study the economic and environmental significance of FDI, Table 2 also reports the estimated marginal effect of FDI on REGS at the mean level of CORRUPT for all four models. In particular, estimates are provided for (i) a one unit increase in the FDI stock or flow, and (ii) a one standard deviation increase in FDI. These indicate that the effect of FDI on REGS is consistently positive at the mean of CORRUPT, i.e., the “welfare effect” dominates the “bribery effect”. For instance, Model (2) indicates that a unit increase in FDI results in a decline of 0.19 (= 0.26 − (0.028 × 2.55)) grams of lead per gallon, evaluated at the mean of CORRUPT.18

Figure 1 illustrates these declining marginal effects for the four models in Table 2. Models (1), (3) and (4) all suggest that the marginal effect of FDI on REGS becomes negative at within-sample levels of CORRUPT of between 2.9 and 5.6. Thus, three out of the four models produce a within-sample reversal of signs, as predicted by Proposition 1. To put these values in context, the average level of CORRUPT over the period 1982–1992 was 2.1 in Italy, 3.2 in Venezuela, 4.2 in Nigeria and 5.7 in Bangladesh. Model (2) estimates that the marginal effect becomes negative at an out-of-sample value of 8.9. Thus, Model (2) suggests that FDI raises environmental policy stringency in all countries. In sum, all four models in Table 2 suggest that the effect of FDI is lower, the greater the level of CORRUPT.

Figure 1.

The marginal effect of FDI on REGS conditional on CORRUPT a

aMarginal effects are calculated using Models (1)–(4), Table 2.

The positive marginal effects reported in Table 2 imply a decline in lead content in gasoline as a result of FDI. Model (2) suggests that a one standard deviation increase in the FDI stock raises REGS by 1.37, equivalent to a reduction in the lead content in gasoline of 1.37 grams per gallon. This reduction in lead content is equivalent to a decline of 1.40 of a standard deviation. Model (4) suggests that a one standard deviation increase in FDI flows raises REGS by 0.34, equivalent to 0.35 of a standard deviation.

The estimated overall effect of CORRUPT on REGS is negative at the mean level of FDI stock in Model (2), as may be expected; see, e.g., Damania et al. (2003). At the mean level of FDI, the effect of a one unit change in CORRUPT equals −0.082 (= 0.16 − (0.028 × 8.64)), representing an increase in the lead content per gallon of gasoline of 0.082 grams. Using Model (4), at the mean level of FDI flows, the impact of CORRUPT on REGS is positive. However, since in all models the impact of CORRUPT on REGS declines with the level of FDI, in Model (4) the impact of CORRUPT on REGS becomes negative at only one-twentieth of a standard deviation above the mean level of FDI flows (−0.0024 =0.093 − (0.11 × (0.82 + (0.95/20)))).

With regard to our control variables, in three of the four models GDP yields a positive relationship with REGS, although decreasing at the margin.19 However, GDP and GDP2 are not consistently significant. In three of the four models URBANPOPsh displays a negative relationship with REGS, albeit decreasing at the margin. A closer examination reveals that the (minimum) turning point levels of URBANPOPsh are between 51% and 67%, where the latter is roughly equivalent to South Korea or Colombia. Thus, for many countries in our sample, our estimates suggest that REGS increases with the urban population share in accordance with our prior expectations. MANUFsh exhibits a generally positive relationship with REGS, which is decreasing at the margin. Across the four models, the maximum turning-point level of MANUFsh is 13.9%(from Model (2), while the turning point from Models (1) and (3) is even smaller), broadly equivalent to Kenya or Bangladesh. Thus, for many countries in our sample, an increase in MANUFsh is associated with a decrease in REGS, in accordance with expectations.

Sensitivity Analysis

To assess the robustness of our results, Table 3 reports eight alternative specifications of equation (4), with Models (5)–(8) relating to FDI stocks and Models (9)–(12) to FDI flows. Models (5) and (9) drop the URBANPOPsh and MANUFsh variables to ensure that they are not unduly influencing the results. Models (6) and (10) assess the sensitivity of our results to our chosen instruments. In these models we drop our two public infrastructure variables and replace them with the rate of inflation, a variable shown to be a deterrent to inward FDI; see, for example, Schneider and Frey (1985), Singh and Jun (1995) and Chakrabarti (2001). Alongside the rate of inflation we use a new measure of the size of a country: the total population.20 Again, a Sargan test fails to reject the null that our equation is properly specified for both models, while an F-test confirms that the instruments are jointly significant in the first-stage regressions. The sign and significance of our results reveal little sensitivity to this change of instruments and are similar to the results in Table 2.

Table 3. Sensitivity analysis (all using instrumental variables); dependent variable REGS (environmental regulations)
 (5)(6)(7)(8)(9)(10)(11)(12)
 FDI
stock FE
FDI stock
FE Alt. IV
FDI stock FE
Pop. scale
FDI stock
FE No scale
FDI
flow FE
FDI flow
FE Alt. IV
FDI flow FE
Pop. scale
FDI flow FE
No scale
  1. Notes: Absolute value of t-statistics in parentheses. *Significant at the 10% level;**significant at the 5% level;***significant at the 1% level. Marginal effects are calculated at the sample means of CORRUPT; s.d.  =  standard deviation. Models (5) and (9) include only minimal control variables, Models (6) and (10) use alternative instruments, Models (7) and (11) scale FDI by population rather than GDP, and Models (8) and (12) do not scale FDI at all.

FDIt−10.20***0.16*0.60***0.13***0.85***0.76*0.32***0.066***
(4.7)(1.8)(4.1)(3.0)(4.6)(1.7)(4.4)(3.9)
CORRUPTt−10.17***0.19***0.10*0.0660.11**0.14***0.14***0.12**
(3.1)(3.0)(1.9)(1.4)(2.4)(2.7)(2.9)(2.5)
FDICORRUPTt−1−0.023***−0.026***−0.098***−0.014*−0.089**−0.11**−0.062**−0.0091
(4.3)(4.1)(3.4)(1.9)(2.0)(2.4)(2.0)(1.5)
GDPt−10.15***0.0360.120.110.066−0.00820.150.16*
(2.7)(0.5)(1.2)(1.2)(1.1)(0.1)(1.6)(1.8)
inline image−0.0038***−0.0022**−0.0047***−0.0039***−0.0019***−0.0014−0.0038***−0.0035***
(5.2)(2.4)(4.2)(3.8)(2.6)(1.6)(3.2)(3.1)
URBANPOPsht−1 −0.050−0.11−0.095 −0.24***−0.058−0.045
 (0.5)(1.4)(1.2) (3.8)(0.7)(0.6)
inline image 0.000630.000760.00068 0.0019***0.000420.00028
 (1.0)(1.2)(1.2) (3.7)(0.7)(0.5)
MANUFsht−1 0.095−0.043−0.062 0.051−0.013−0.022
 (1.0)(1.0)(1.2) (0.9)(0.3)(0.5)
inline image −0.00380.00140.0019 −0.00200.000590.0012
 (1.5)(1.2)(1.3) (1.4)(0.5)(0.9)
Observations319319319319319319319319
R20.400.420.350.350.370.390.360.36
F-test FDI vars. (p-value)31.6817.6422.9611.1421.186.6419.6815.28
(0.000)(0.000)(0.000)(0.003)(0.000)(0.03)(0.000)(0.000)
F-test on IVs (p-value)57.5845.768.965.9747.5123.6627.7413.12
(0.000)(0.000)(0.03)(0.04)(0.000)(0.000)(0.000)(0.004)
Sargan test (p-value)2.070.0020.691.581.960.0761.452.38
(0.15)(0.96)(0.71)(0.46)(0.16)(0.78)(0.48)(0.30)
(∂REGS/∂FDI)s.d.1.000.760.771.270.590.450.480.80

Models (7), (8), (10) and (12) assess the extent to which the scaling of FDI by GDP influences the results in Table 2. Thus, Models (7) and (11) scale FDI by the level of population rather than GDP, while Models (8) and (12) do not scale FDI at all.21 In each case the sign and significance of our variables of interest remain very similar to those in Table 2.

Each model in Table 3 also reports the marginal effect of a one standard deviation increase in FDI. These also appear to be reasonably stable across models. Including the IV models in Table 2, the marginal effect of a one standard deviation increase in FDI stocks ranges from 0.76 to 1.37 across the five different models, while the equivalent figures for FDI flows range from 0.34 to 0.80.

As a final check on the robustness of our results, we estimated dfbetas, which measure the difference between each regression coefficient when the ith observation is included and excluded, with the difference scaled by the estimated standard error of the coefficient. Bollen and Jackman (1990) argue that an observation is deserving of special attention if ∣dfbeta∣ > 1, implying that the observation shifted the estimated coefficient by at least one standard error. Across all independent variables, including those within our first-stage regressions, we find no dfbetas that exceed 1. In fact, no dfbetas exceed 0.5. We therefore have no evidence to suggest that outliers are exerting undue influence on our estimated coefficients.

IV. Conclusion

Whereas the theoretical and empirical literature investigates the effects of variations in the stringency of local environmental policies on foreign direct investment, the effects of foreign investment on environmental policy have largely been ignored. In this paper we take a first step toward remedying these deficiencies.

We apply a political economy model of local environmental policymaking. The environmental policy effects of foreign direct investment are found to be conditional on the government's degree of corruptibility. Foreign direct investment leads to a higher (lower) stringency of environmental policy when the degree of local government corruptibility is low (high). Our empirical findings are fully consistent with the predictions of the model.

The results reported here raise some concerns about the previous empirical literature which seeks to uncover “pollution haven” effects, i.e., that foreign firms locate where environmental policies are relatively weak. This literature has largely ignored the fact that environmental policies are endogenously determined and, in particular, has not incorporated the environmental policy effects of foreign direct investment discussed in this paper. Our analysis may consequently provide some guidance for future empirical efforts in this area.

The policy implications that emerge are that warnings of negative effects of foreign direct investment should be taken seriously, in particular where the degree of government corruptibility (corruption) among policymakers is high. In such countries, foreign direct investment contributes to the creation of the feared pollution havens. On the other hand, the results are encouraging for countries with relatively low degrees of corruptibility among policymakers. Foreign direct investment may even result in improved environmental quality in such countries. Our results further reinforce the need for reforms that reduce the level of government corruption (corruptibility) in many countries.

Appendix

Table A1. Data definitions and sources
VariableDefinitionSource
  1. Countries in sample: Argentina, Australia, Bangladesh, Belgium, Brazil, Canada, Chile, Colombia, Ecuador, Egypt, Ethiopia, France, Germany, Ghana, Greece, India, Italy, Japan, Kenya, South Korea, Mexico, Morocco, Mozambique, Netherlands, Nigeria, Pakistan, Philippines, Portugal, South Africa, Spain, Tanzania, Thailand, Venezuela.

REGSLead content of gasoline, multiplied by −1 to
form an index of environmental regulations
Octel's Worldwide
Gasoline Survey
(various years)
CORRUPT“Government honesty” subtracted from 6 to
form an index of corruption from 0 to 6
International Country
Risk Guide
FDIInward FDI stocks and flows, divided
by aggregate GDP
UNCTAD FDI
Database (2001)
GDPPer capita incomeWorld Development
Indicators (2004)
URBANPOPshShare of the population living in urban areasWorld Development
Indicators (2004)
MANUFshManufacturing value added as a share of GDPWorld Development
Indicators (2004)
PHONETelephone mainlines (per 1,000 people)World Development
Indicators (2004)
TVTelevision sets (per 1,000 people)World Development
Indicators (2004)
INFLATIONInflation rateWorld Development
Indicators (2004)
ECON.ACT.POP.Economically active populationWorld Development
Indicators (2004)
POPTotal populationWorld Development Indicators (2004)
Table A2. First-stage FDI equations; dependent variable: FDI stock or flow
 (A1) FDI stock(A2) FDI stock(A3) FDI flow(A4) FDI flow
 used in Model (2)used in Model (6)used in Model (4)used in Model (10)
  1. Notes: Absolute value of t-statistics in parentheses. *Significant at the 10% level;**significant at the 5% level;***significant at the 1% level. These first-stage equations were used to estimate the main instrumental variables models in Table 2(Models (2) and (4)) together with the alternative instruments models in Table 3(Models (6) and (10)). The first-stage equations used to estimate the remaining sensitivity models in Table 3, which use the same specification as models (A1) and (A3) above, are omitted.

PHONEt−1−0.014** −0.0018* 
(2.3) (1.9) 
TVt−10.0077 0.0069*** 
(1.3) (6.7) 
ECON.ACT.POP.−0.45*** −0.68*** 
(7.5) (4.0) 
INFLATION −0.00040 −0.00083
 (1.4) (0.8)
TOT.POP. −0.29*** −0.59***
 (6.6) (4.8)
CORRUPTt−10.39**0.31*0.0180.065*
(2.2)(1.8)(0.5)(1.8)
GDPt−10.11−0.094−0.054***−0.043**
(0.9)(1.1)(2.8)(2.3)
inline image−0.0000300.0000260.000015***0.000012**
(0.9)(1.1)(2.8)(2.3)
URBANPOPsht−1−0.70***−0.66***0.11**0.11**
(4.1)(3.7)(2.3)(2.3)
inline image0.0047***0.0041***−0.00061−0.00081**
(3.5)(2.9)(1.6)(2.2)
MANUFsht−1−0.89***−0.92***−0.55−0.10**
(4.9)(5.1)(1.3)(2.4)
inline image0.026***0.027***0.000920.0025***
(6.2)(6.7)(0.9)(2.6)
Observations319319319319
R20.360.310.300.28
F-test on IVs (p-value)60.2745.7669.6923.66
(0.000)(0.000)(0.000)(0.000)
Sargan test ( p-value)4.050.0022.910.076
(0.13)(0.96)(0.23)(0.78)

Footnotes

  • 1

    Schulze and Ursprung (2001) argue that the model in Grossman and Helpman (1994) can be seen as a model of corruption, in particular since campaign contributions (bribes) are given in order to affect policy, not election outcomes; see also Damania, Fredriksson and List (2003).

  • 2

    Corruption of local government officials is a relevant issue also in industrial countries, for example in the U.S. (see http://www.chicago.fbi.gov/silvershovel/silvershovel.htm). Hall (1999) reports that executives of a French firm (Général des-Eaux) were convicted of bribing the Mayor of St-Denis in order to obtain a water concession. Examples of high-level politicians who have been charged with corruption and fund-raising violations include German Chancellor Kohl, Italian Prime Minister Berlusconi, and in Israel both President Weizman and Prime Minister Barak.

  • 3

    Thus, we abstract from modeling the decisions by foreign multinationals to invest abroad and where to locate production. The decision to invest abroad has been examined by, for example, Head, Ries and Swenson (1999). Foreign firms may prefer to produce locally rather than exporting to the market due to high transportation costs or trade barriers (or the threat of such barriers).

  • 4

    Even firms located outside the U.S. influence the U.S. government. For example, see Gawande, Krishna and Robbins (2002) for empirical evidence on (successful) foreign lobbying for reductions of U.S. trade barriers. Moreover, Transparency International (2003) reports an index ranking the propensity of companies from 21 different countries to pay bribes when they do business abroad (0 = high bribery; 10 = low bribery). The scores included Australia (8.5), the U.S. (5.3), Japan (5.3), Italy (4.1) and Russia (3.2). The business sectors in which bribery most commonly occurs were also identified. Among heavily polluting sectors, the scores included heavy manufacturing (4.5), mining (4.0), and oil and gas (2.7).

  • 5

    Thus, when the degree of corruptibility is relatively high, foreign direct investment may create pollution havens (i.e., increased pollution levels). However, when corruptibility is low, it may result in a decline of the pollution damage, despite an increase in total output produced and sold in the domestic economy.

  • 6

    Note that this is a reverse relationship between increased economic liberalization (greater trade openness or FDI flows) compared to the effect discovered in our paper.

  • 7

    Levinson and Taylor (2003) account for endogeneity in a study of the effect of environmental regulations on trade flows, but ignore corruption.

  • 8

    The introduction of an additional market complicates the analysis without adding further insights; we ignore this aspect of the problem.

  • 9

    Since our focus is on the political economy effects of FDI on environmental policy, we abstract from possible differences in marginal abatement costs between domestic and foreign firms. Thus, we (implicitly) assume that capital vintage, technology and know-how are identical across firms.

  • 10

    Schulze and Ursprung (2001, p. 68) argue that “the portrayed interaction between the organized interest groups and the government meets the circumstances of corruption”, which is consistent with Shleifer and Vishny's (1993, p. 599) view of corruption as “the sale by government officials of government property for personal gain”, where government property refers to government policies; see also López and Mitra (2000).

  • 11

    Assumption 1 appears relatively weak. Note that if we assumed that all firms’ profits were included in aggregate social welfare, it would hold automatically.

  • 12

    Using Proposition 1 we may find the effect of FDI on total pollution levels. In particular, when ∂t/∂N F < 0 (corruptibility is relatively high), FDI is more likely to create a pollution haven (results available on request).

  • 13

    The use of two- and three-year lags did not substantially change our results. Equation (4) would ideally include variables capturing the marginal cost of regulations as well as the marginal benefit. Suitable variables would include a measure of compliance costs or the level of pollution. Compliance cost data are not available, but this variable may be picked up by the level of worker pressure (which is a function of the costs of environmental regulations in terms of job and wage losses) as captured by MANUFsh. Including the level of pollution in equation (4) raises endogeneity concerns since pollution is obviously a function of REGS. However, lagged sulfur dioxide was tested as a determinant of REGS but was not consistently significant and did not influence the sign and significance, or the marginal effects, of our variables of interest.

  • 14

    Lead content in gasoline is reported biannually. To obtain our reported results we assumed that lead content remained constant over each two-year period. However, in order to test the sensitivity of our results to this assumption, we also ran regressions in which each missing year of lead content data was assumed to be the average of the previous and the following year's lead content (e.g. 1985 lead content data is equal to the average of 1984 and 1986 lead content). Furthermore, we also ran regressions in which we only included the first reporting year (so the panel is effectively halved). The results were very similar across these three different samples and the key conclusions that stem from our empirical analysis were the same.

  • 15

    Since high lead content is considered by some to improve certain aspects of engine performance, it is possible that the demand for lead content in fuel could increase over time as consumers’ income rose. However, in light of the known toxicity of lead it seems unlikely that such an effect would outweigh the environmental and health considerations.

  • 16

    See Appendix A of Damania et al. (2003) for details on the construction of these three alternative proxies for the stringency of environmental regulations. See the Octel World Gasoline Survey for detailed information on the construction of the lead content variable.

  • 17

    Our sensitivity analysis tests the sensitivity of our results to the scaling of FDI by GDP and includes a model in which we scale FDI by population and a model in which FDI is not scaled.

  • 18

    Model (1) indicates that a unit increase in FDI results in a decline of 0.0057 (= 0.049 − (0.017 × 2.55)) grams of lead per gallon, evaluated at the mean of CORRUPT. The effect differs sharply at high levels of CORRUPT. At one standard deviation above the mean of CORRUPT, the effect equals an increase of 0.021 [= 0.049 − (0.017 × (2.55 + 1.57)] grams of lead per gallon of gasoline. Thus, for sufficiently high levels of corruptibility, the “bribery effect” dominates the “welfare effect”, as is consistent with our theory.

  • 19

    Both country fixed effects and time effects are consistent with prior expectations, with the latter indicating that REGS has increased over time (i.e., lead content has decreased).

  • 20

    Models (A2) and (A4) in Table A2 in the Appendix provide the first-stage results for Models (6) and (10), respectively.

  • 21

    Most empirical FDI papers scale FDI, most typically by GDP or population; see, for example, Singh and Jun (1995), Chakrabarti (2001) and Portes and Rey (2005).

Ancillary