The impact of foreign direct investments on the CO2 emissions

Vuong Duc Hoang Quan et al | The impact of foreign direct investments on the CO2 emissions in Southeast Asian countries VUONG DUC HOANG QUAN Ho Chi Minh City Institute for Development Studies – quanv.biz@gmail.com BUI HOANG NGOC University of Labour Social Affairs HOANG THI BICH DIEN University of Labour Social Affairs Abstract The pressure to increase the national income per capita sometimes puts the Government of developing countries into a dilemma between economic growth and environmental protection On the basis of the original model developed by Kuznets (1955) regarding the relationship between a country’s CO2 emissions to the environment and its average income, an empirical test applying the System Generalized Method of Moments (S-GMM) and Pooled Mean Group (PMG) regression to data from Southeast Asian countries during a period of twenty years from 1995 to 2014 is conducted The results find strong statistical evidence for the existence of an inverted- U effect as the theory assumes Moreover, the impact of the factor of FDI attraction on the environment which is added to the Kuznets’ original model is also confirmed by the study Keywords: FDI; Environmental Kuznets Curve (EKC); CO2 emission; Southeast Asian countries Introduction The capital needed for investment in infrastructure, social security, education, health care, national defense, and so on is always great in the process of development Economic theory has shown that capital for developing countries or developing territories is of particular important significance in the early stages of development, which satisfies immediate needs, also helps to promote the efficiency of other sources of capital such as resource capital, human capital, scientific and technical capital, etc So then, foreign direct investment (FDI) is considered as a priority capital source In addition, the positive contribution of economic growth based on the overall assessment of the World Economic Forum (WEF) 2010/2011, the FDI capital has also had negative impacts on the host countries The typical negatives are: Natural resources being over-exploited; causing the difficulties for the development of domestic enterprises; a national identity being abrasive; the risk reducing the quality of life caused by excessive environmental pollution The Association of SouthEast Asian Nations (ASEAN) which is abbreviated as Asean was established in 1967 The five original members have now expanded to eleven countries including Brunei, Cambodia, Indonesia, Laos, Malaysia, Myanmar, Philippines, Singapore, Thailand, Timor-Leste and Vietnam According to the International Monetary Fund (IMF), the amount of FDI being poured into ASEAN countries is still increasing Asean is considered a fairly attractive destination for foreign investors with annual FDI capital exceeding $ 100 billion which not many regions in the world are able to If only considering per capita income, the economic development model of Singapore and Malaysia are the models worth learning However, considering the quality of life is another matter because the per capita CO2 emissions of Singapore and Malaysia are more than twice the per capita CO2 emissions of the world There have been empirical studies on the inverted-U letter effect in the Asean region, but some challenges need still to be studied further: specifically finding the answer to the bending point of the U figure (ie, the point transferring gradually from the polluted environment to improved environment) This study is divided into sections After the introduction, section presents the summary theoretical background and overview of previous studies on the topic of this study Section introduces research models, data sources, and methods of analysis and data processing The results of the study, discussion of the results and some management policy implications will be presented in section The final section will be the conclusions and limitations of the study and some suggestions for the next research 2 Theoretical background and overview of previous studies 2.1 Theoretical background Describing the relationship between economic growth and environmental quality (represented by the level of environmental pollution), Kuznets (1995) [1] has given the idea of an inverted U curve (EKC - Environmental Kuznets Curve) He said that in the first stage of economic growth, the Government has tended to loosen environmental regulations to attract FDI because of the high growth pressure and the capital accumulation scale of the whole economy limited Thanks to FDI, the average income has improved; however, with the increase in average income, the environmental pollution has also increased At this stage, the large fuel consumption, has resulted in huge amounts of CO2 emissions into the environment because countries have mainly exploited natural resources in raw form, production technology has been backward, management level has been weak As the average income increases to some extent and life is improved, people begin to perceive the importance of quality of life and the quality of the surrounding ecological environment In addition, along with the improved economic conditions, the economic integration and the advantage of the later developed countries will help nations, businesses and citizens to be able to choose green, clean technologies and friendly with environment Environmental pollution will decelerate, reverse and then reduce, so then environmental quality will be enhanced 2.2 Overview of previous studies The relationship between economic growth and environmental quality has been considered and studied in many countries/ regions around the world Pham Xuan Hoan et al., (2014) [2] have used balance panel data with two variables including average GDP and average GDP2, regression estimation using the Fixed Effect Model (FEM) and the Random Effect Model (REM) to test the Kuznets curve of environment for the ten Asean countries during the period 1985-2010 confirmed the existence of an inverted-U effect However, this study has ignored one important factor that is the relationship between the CO2 emissions of the current year and the CO2 emissions of previous years Tran Thi Tuan Anh (2016) [3] extended the model to include four variables under the study: average GDP, average GDP2, trade openness, and density population Applying the Spatial Regression to Dynamic Panel Data (DPD) to the data of eight Asean countries during the period 1994-2011, the author found the statistical evidence of the environmental existence of the inverted-U shape effect in the Southeast Asian countries, and the impact of CO2 emissions last year to the present year was credible However, both the studies ignored the impact of FDI and the industrial share of GDP, according to Anis Omri et al (2014) [4] and Huiming Zhu et al., (2016) [5], it is really unfortunate that the sector that emits the most of CO2 is the industry, and in the early stages of economic growth, the industry is mainly dominated by foreign invested enterprises In other regions of the world, the study of Shafik & Bondyupadhya (1997) [6] was performed for 149 countries during the period 1960-1990, with time-series data and crosssectional data, these authors found that the deterioration of the environmental quality as rising average income, and the trends of the environmental quality to be better when the countries are richer Galeotti & Lanza (1999) [7] used panel data for 110 countries from 1970 to 1996 finding the existence of an invertedU shape effect for the global environment They asserted that global pollution was still rising by the causes from the pressures of the rapidly growing per capita income of developing countries, and two factors of income and population played a decisive role in the amount of CO2 emissions The study by Anis Omri et al (2014) used panel data for fiftyfour countries around the world, and the study by Shenggang Ren et al., (2011) [8] for the Chinesse economy proved that CO2 emissions would be increased by FDI which was the second strongest impact factor after the previous year's CO2 emissions The study by Huiming Zhu et al., (2016) for five Asean countries including Indonesia, Malaysia, Philippines, Singapore, Thailand, and the study by Pao & Tsai (2010) [9] for the four BRIC countries including Brazil, Russian Federation, India and China which both of these studies confirmed to have the relationship between CO2 emissions, FDI and economic growth However, Dijkagraff & Vollebergh (2005) using data of OECD countries during the period 1960- 1997 found evidences to deny the existence of the inverted-U shape effect of EKC “A new look” was the expression of Jungho Baek (2015) when testing the effect of EKC for five Asean countries including Indonesia, Malaysia, Philippines, Singapore and Thailand during the period 1981-2010 He did not find the inverted-U shape effect, but found the U shape effect in the long-term cointegration relationship between GDP, GDP2, energy consumption and FDI and CO2 emissions into the environment This difference in the results of the above studies on the existence of the inverted-U shape effect and the impact of FDI on CO2 emissions in EKC curve testing has explained the necessity for having more experimental evidence of this relationship Research methodology 3.1 Research models To test the inverted-U shape effect or the relationship between CO2 emissions and per capita income, Ang (2008), Sharma (2011), and Anis Omri (2013) using the Cobb- Douglas production function has been suitable Especially, Anwar & Nguyen (2010) [14], and Anwar & Sun (2011) also added the FDI factor to the production function So, the Cobb-Douglas function is written as: Y ε α = e AK λ (Equation 1) β E L Where: Y is total production (the real value of all goods produced in a year (GDP)), A is total factor productivity, E is total Energy consumption, K is the capital input of the economy (domestic capital and FDI), L is labor input (the total number of personhours worked in a year) α, β, λ is the contribution proportion of factors to the actual output When further research on the production function, Pereira & Pereire (2010) [15] proposed E = b.CO2, while Anis Omri et al (2014) suggested that K = c.FDI, so that Equation can be rewritten into: λ α ε λ α β Y = b c e A(CO2 ) (FDI ) L (Equation 2) Assumed that the economy is constant in scale (ie α + β + λ = 1), then divide both sides of the equation for L to find the per capita income, Equation is written: Y λ α ε = b c e A( λ FDI α ( ) L L CO2 ) L Take the logarithm both sides of Equation as follows: (Equation 3) Y log( ) = log(b λ α c A) + λ log( + ε L CO2 )+ L λ α α log( FDI ) L Set a = log(b c A) and move the side, Equation is represented in the form of panel data as follows: (Equation 4) CO2 g( ) = β0 g(GDP ) + g( FDI it + L β2 L + β1i )L εit i it it (Equation 5) To test the inverted-U shape effect, variable (GDP/L) squared have to be added, and the control variables are increased into Equation Tran Thi Tuan Anh (2016), using the control variable as the openness of the economy and population density Huiming Zhu et al., (2016), beside the openness variable, population density is also used the variable of the industry ratio in GDP and total capitalization of the stock market Inheriting the above studies, this paper proposes the model for this study is as follows: LnCO2 capita +β it =0 (βi + 1v ) + β LnGDPcapita it it LnGDPcapita it it +δ LnControl + e (Model A) According to Bond (2002), data of CO2 emissions is often a persistent data series, meaning that the current year's emissions are strongly correlated with previous years’ emissions Ignoring this effect may cause model A to be endogenous by omitting the variable, so the paper uses Model B to study the inverted-U shape effect as follows: LnCO =0 (βi + v ) + γ2 LnCO capita it i,t−1 capita + β LnGDPcapita +δ LnControl + e +it β LnGDPcapita it it (Model B) Where: i = 1,2,.,7 denotes the country: Cambodia, Indonesia, Malaysia, Philippines, Singapore, Thailand, and Vietnam vi: denotes individual characteristics of each country It means εit = vi + eit t: denotes the time period (from 1995 to 2014) Controlit: denotes control variables respectively other factors impacting on CO2 emissions (including FDI variable, Hour variable, Open variable and Industry variable) The paper uses more the Hour variable (average number of working hours) because in addition to the mainly factor emitting CO2 is the industrial sector and the FDI sector, increasing the number of working hours will also create emissions higher than normal Moreover, due to the characteristics of sectors, the industry sector will often have to increase the number of working hours than other sectors, so it is reasonable to add more the average working hour variable to be a control variable Table Conventions of Variables Variable Sign CO2bq Variable Content GDP bq Per capita income (calculated by the fixed price in 2011) USD/ a person GDP2bq Per capita income squared Million people CO2 emissions per capita Unit Metric tons Data Source IEA1 WB2 WB Number of FDI FDI Ho ur Open Industry 3.2 per capita (calculated by FDI inflow) The average number of working hours by an employee a year USD/ a person UNCTAD3 hour/ a year Openness of the economy Proportion of industry in GDP % % WB UNCTAD WB Research methodology and data Most studies on EKC before 2000 used time-series and crosssectional data This is only suitable at that time because timedata requires a long observation, and cross- sectional data is not a reflection of the continuity of observations by having to cut at certain points Jodson (1995) argues that if a study does not utilize all of the time and spatial aspect of data, the study wastes a lot of information possibly provided by data4 Panel data that later evolved to address those disadvantages, but the lumped impact estimate (Pooled) did not account for the differences in special characteristics Estimated by fixed effects (FE), random effects (RE) will be dictated when the model has a short time series t and large space i (Judson et al., 1996) According to Bond (2002) [10], data of FDI in terms of CO2 emissions are usually persistent time series, ie, the amount of the attracted FDI, CO2 emissions in the following years are often the very powerful relationship with data from previous years, so the synchronism of this factor should not be neglected in this study model After Hansen [11] published the Generalized Method of Moments (GMM) in 1982, Arellano & Bond http://www.iea.org/statistics/ http://www.conference-board.org/data/economydatabase/ http://unctadstat.unctad.org Tran Tho Dat (2011), The Role of Human Capital in Growth Models, Economic Research No 393 (1991) [12] applied GMM to the dynamic table model to improve the firmness and the effectiveness of the DPD (Dynamic Panel Data) model However, the GMM method also has limitations5: (i) the slope coefficients vary by panel unit; (ii) the short-term dynamic characteristics and long-term coherence are not shown Therefore, in this study, the authors used both the Generalized Method of Moments (GMM) and the Pooled Mean Group Regression (PMG) estimation method based on balanced panel data for seven countries including Cambodia, Indonesia, Malaysia, Philippines, Singapore, Thailand and Vietnam for 20 years from 1995 to 2014 to determine the inverted-U shape effect of the Kuznets curve that actually exists in the long term ? And in the short term, what factors impact on the CO2 emissions of the seven Southeast Asia countries? Data collected from three reliable sources are the International Energy Agency (IEA) and the World Bank (WB), the United Nations Conference on Trade and Development (UNCTAD) Myanmar, Laos, Brunei, and Timor- Leste have to be eliminated because the data of these four countries is lacking, especially the lack of data on working time Laos and Timor-Leste also lack import and export data, so it is impossible to calculate the openness of the economy Results and discussion 4.1 Descriptive statistics According to the International Energy Agency (IEA), the annual emissions of the world are relatively stable and rising slightly However, in Southeast Asia, the average level of CO2 emissions of the countries varies considerably Singapore and Malaysia have been two times higher than the world average of CO2 emissions, in the last 20 years, Singapore has been trying to decrease greenhouse gases, but CO2 emissions of Malaysia has been still rising In Cambodia, Indonesia, Philippines, Thailand and Vietnam, though CO2 emissions have tended to rise, but remain below the world average However, this does not mean that these five countries are not threatened by air pollution, which has occurred in some concentrated industrial parks and large urban areas where average statistics can not reflect the details Nguyen Minh Tien (2015), DGMM and PMG Regression with Panel Data, Foreign Economic Relations Journal, No 11, 40-48 Figure Average CO2 Emissions of the World and ASEAN Countries Average CO2 Emissions of the World Average CO2 Emissions of ASEAN Countries 15.00 10.00 5.00 - 5.00 4.50 4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 - Campodia Malaysia Singapore Indonesia Vietnam Philippines Thailand 1971 1974 Table describes the average value of variables in the model by each country, and Singapore is the country with the highest GDP per capita at $ 62,500 per year and also the country that attracts the highest average FDI per capita reaching $ 6,000 per year Cambodia and Vietnam are the two countries with the lowest average income, but also the two countries in which labor have to work with the most of the average hours per year Table Average Values by Country Variable s CO2bq Indonesi a Cambodi a Malaysi a Philippine s GDPbq 7,732 2,051 4,986 Hour 2,002 2,485 20,16 02,321 FDI 27.99 47.47 22.11 Open 0.32 0.64 233.5 91.37 Industry 44.71 23.05 42.85 1.39 0.23 5.72 0.85 2,105 Singapor e 9.25 Thailan d 2.94 62,521 12,40 2,339 92,327 Vietna m 0.94 3,617 2,352 107.5 60.83 50.17 0.49 6,087.0 2.39 32.91 30.54 37.80 35.55 0.76 4.2 Research results The time series properties of the variables in Model B are checked through five types of panel unit root test: LLC, Breitung, IPS, ADF and PP tests Both LLC and Breitung test assume that there is a common unit root process across the cross-sections For these tests, the null hypothesis is that there is a unit root, while the alternative hypothesis is 110 | ICUEH2017 that there is no unit root The other tests assume that there are individual unit root process across the cross-sections The null hypothesis is that there is a unit root On balance, the results of table demonstrade that all of the series in Model B appear to contain a panel unit root in their levels but are stationary in their first differences, indicating that they are integrated at order one, i.e., 1st Table Unit Test Results of Each Variable Common Unit Root Varia ble Names CO2bp@ GDPbp@ @ GDPbq2@ FDI@@ Hour@ Open@@ Industry @ LLC Level 1.13 4.598 7.734 2.178* 1.860* 1.359 1.660* Breitung 1st diff Leve 1st diff l 1.39 -0.405 2.77** 3.30** 0.51 3.71** 6.41 8.71** 1.459* 0.54 0.114 4.13** 2 1.49 2.72** 0.461 0.66 3.61** 2.44** 1.66 1.77** 3.41** Individual Unit Root IPS ADF Level 1st diff Level 3.00 7.69 9.15 1.437 0.467 0.77 0.29 4.84*** 2.70*** 6.57** 5.77** 4.22*** 3.95*** 2.77** 4.08 0.13 0.04 23.22 * 19.4 16.3 17.4 1st diff 49.97** * 32.11* ** 21.29* 57.75** * 45.69** * 40.30** * 30.18** * PP Level 1st diff 3.463 994.71* ** 0.035 45.80** * 0.011 36.32** * 29.77** 107.67* * ** 21.75 143.63* * ** 19.63 97.61** * 17.80 86.07** * Note: @ not trending, @@ trending *,**,*** respectively with the significance level of 10%, 5%, 1% However, Anis Omri et al (2014), testing the U-curve shape inverted effect of the EKC for 54 countries in the world during the period 1990-2011 found that the stop test was only important for regression result, and it was difficult to ensure the stop of the actual data series because of large geographic distribution In order to make this study reliable, the paper uses three models simultaneously Model uses the original data, and it can be estimated using the OLS method (Pooled), fixed-effect (FE), and randomeffect (RE) with balance panel data Model uses the original data; the regression results are estimated by using the System Generalized Method of Moments (S-GMM) Model uses 1st differential data, using PMG (Pooled Mean Group Regression) to estimate The paper uses the software Stata 14 to process the data For the Model 1, the paper estimated by methods including POOLED, FEM, REM, the test results were compared FEM with POOLED, FEM with REM showed that the FEM model was most suitable for the data sample Further treatment of the phenomenon of the error variance change and self-correlation of series, regression results of the Model are shown in the revised FEM column of the Table As a result, the GDPbq variable is positive, the GDPbq2 variable is negative, FDI and Industry variable are positive, and all four variables are statistical significance that showed the existence of the inverted-U shape effect in the environmental EKC of Southeast Asia, average FDI and industrial share of GDP contributed to increase CO2 emissions However, this calculation result neglected the relationship of CO2 emissions from the previous years to the current year, and this relationship was controlled by using the Model for validation and the Model for analysis Table Regression Results of Research Variables The Revised FEM Coefficient β Prob CO2bq(t-1) S-GMM PMG Coefficient β Prob 0.8861976 0.000 Coefficient β Prob GDPbq GDP 0.000319 0.000 0.0000369 0.027 0.0002896 0.000 -2.72E-09 0.000 -4.98E-10 0.008 -3.12E-09 0.000 Hour -0.0001707 0.257 0.00026 0.147 0.0001565 0.292 FDI 0.0000709 0.039 0.000081 0.000 0.0004266 Open 0.0804988 0.219 0.1211371 0.286 -0.3973444 0.000 Industry 0.0093469 0.024 0.0084525 0.086 -0.004495 Intercept -0.1325144 0.726 -0.969967 0.088 bq No Obs 140 F - test 33.64** * 39.43** * Hausman test AR(1) test 133 0.002 0.16 133 1540000 *** 0.109 AR(2) test 0.542 Sargan test 0.267 Log 245.88 Likelihood Notes: ***, ** and * respectively showed for the significance level 5of 1%; 5% va 10% 16 | ICUEH2017 Source: According to the calculation of the authors The paper continued to test the inverted-U shape effect using the System-GMM regression method proposed by Arellano & Bond (1991) The test results of AR (1) = 0.109, AR (2) = 0.542, and the Sargan test = 0.267 were all satisfactory, so the regression result was reliable enough for analysis According to the results in the S-GMM column of the Table 4, the statistical significance variables included the CO2bq(t-1) variable is positive, the GDPbq variable is positive, the GDPbq2 variable is negative, and the FDI variable is positive This proved that the inverted-U shape effect still existed, and the CO2 emissions of the previous year as well as the average FDI had the impact on increasing the CO2 emissions of the current year According to this result6, the bending point of the inverted U for Southeast Asia would be at the threshold of $ 37,000 per year For the Model 3, the results of table demonstrade that all of the series in Model B appear to contain a panel unit root in their levels but are stationary in their first differences, the article continued to test the cointegration by Fisher test based on the Augmented Dickey Fuller and Philips Perron with the lags = According to the results in the Table 5, the CO2bq variable and the Industry variable stop at the difference, so as McCoskey & Cao (1998) recommended that it was necessary to additionally test the cointegration with Westerlund test7 ( 2007) Table The Fisher and Perron Stop Test Result with the Lags = Variables CO2bq GDPbq GDP2 bq Hour FDI Open ADF Test (Prob>chi2) No Trending Trending No Trending Trending 2.54 0.106 0.04 28.97** 18.269 3.18 0.07 0.02 24.32** 15.3 5.17 2.45 20.8 35.89*** 1.76 4.49 110.75*** 26.88** 15.8 93 11.1 67 10.892 PP Test (Prob>chi2) 16.1 25 15.0 21.53* 08 The bending point = -(0.0000369/(2x(-4.98E-10))) = 37.048 Nguyen Minh Tien (2015), DGMM and PMG Regression with Panel Data, Foreign Economic Relations Journal, No 11, 40-48 Industry 7.673 19.18 20.595 6.817 33.31* 23.06 129.96** 107.04*** ** ** 131.73*** 151.79*** 21.22* 12.60 Note: ***, ** and * respectively showed for the significance level of 1%; 5% va 10% ∆CO2 ∆Industry The Westerlund test results in the Table show that at least tests reject the null hypothesis H0 (no cointegration) between the average CO2 emissions with the average income, average working hours, average hours worked, Average FDI, openness of the economy and industrial proportion of GDP According to Anshasy (2012) this is achieved, that is, all independent variables are cointegrated to the dependent variable, so the application of the PMG model is suitable Table 6: Testing Cointegration by Westerlund Test Dependent Variable: CO2 Independent Gt Variables GDPbq GDP2 Gα Pt Pα -2.972** -15.46* -8.31*** -11.70 -8.35*** -25.49*** -10.37*** -19.12*** FDI 2.54 2.57 -3.20*** 9.89 -11.42 -28.75*** -8.14*** -17.83*** Open -2.935*** -32.79*** -9.49*** -21.17*** bq Hour Industry -25.88** -7.88*** -16.31*** 2.28 Note: ***, ** and * respectively showed for the significance level of 1%; 5% va 10% The estimated results using Pooled Mean Group regression method (PMG) in short term are shown in the Table 7, in the long term shown in the PMG column of the Table Table 7: The Estimated Results Using Pooled Mean Group Regression Method (PMG) in Short Term Dependent variable: ∆ CO2 Average Revised coefficient ∆GDPbq Coefficie nt β -0.375 -0.0001 Error Prob 0.125 0.00 0.39 0.0001 Dependent variable: ∆ CO2 Average ∆GDP bq ∆Hour ∆FDI Coefficie nt β Error 2.78E-08 0.0014 2.41E08 0.0011 -0.0002 0.0001 ∆Open 0.140 0.193 ∆Industry 0.007 0.025 Intercept 0.089 0.112 Prob 0.24 0.15 0.10 0.46 0.76 0.42 According to this result, in the short term, beside of the factors that contribute to the research model are still the other factors that impact on CO2 emissions The revised coefficient is -0.375, confirms the existence a long-term cointegration relationship in at least one of the studied countries The absolute value of the revised coefficient (0.375