The Nexus Between Air Pollution and Outward Foreign Direct Investment

Abstract

This study examines the causal relationship between air pollution (AP) and outward foreign direct investment (OFDI) in China by applying the bootstrap rolling-window full- and subsample Granger causality test in a sample from 2013 to 2022. We find that AP negatively influenced OFDI in 2016, while this influence became positive at the end of 2019. The results confirm the coexistence of the pollution haven hypothesis, factor endowment hypothesis, and Porter hypothesis. In turn, OFDI negatively influenced AP in 2019, which proves the “composition effect” and “technique effect,” implying that OFDI brings better air quality by optimizing the economic structure and promoting green technology. However, this influence became positive in 2020, which is consistent with the “scale effect,” indicating that OFDI worsens air quality by expanding production. This research provides insights for the government to coordinate OFDI growth and carbon neutralization to achieve sustainable development. It also has implications for firms to reduce environmental costs through OFDI.

Introduction

This study investigates the causal relationship between air pollution (AP) and outward foreign direct investment (OFDI) in China to determine whether the country can achieve sustainable development. The United Nations Conference on Trade and Development reported that global OFDI reached 1.7 trillion US dollars in 2021, with an increase of 119% over 2020. OFDI is an important driver of economic development (Razzaq et al., 2021), which AP affects. Serious AP may encourage multinational firms to relocate factories to other countries through OFDI to avert regulations. For example, horrible AP (e.g., the Los Angeles photochemical smog incident) promotes the US to strengthen environmental regulations, such as issuing the Clean Air Act Amendments, which causes regulated enterprises to accumulate their foreign assets through OFDI (Hanna, 2010). However, the promoting effect of AP on OFDI does not always hold. Serious AP and environmental regulations may bring financial constraints for firms, which reduces the funds for OFDI (Greenstone, 2002). Conversely, OFDI has a certain impact on AP. OFDI is conducive to economic growth but also leads to severe air pollution as many countries develop their economies (Cai et al., 2021), ultimately threatening public health (Tainio et al., 2021). Thus, the government should urgently coordinate AP and OFDI. In short, OFDI is conducive to economic growth, which is connected with AP. However, the relationship between AP and OFDI is still ambiguous. We examine the causal relationship between AP and OFDI, attempting to observe whether achieving improvement in both OFDI and environmental quality is possible.

China’s OFDI has continued to increase and has contributed to the country’s growing economy and employment in recent years (Liao et al., 2021). According to the Ministry of Commerce, China’s OFDI flows increased by 16.3% from the previous year in 2021, ranking among the world’s top three for the 10th consecutive year. OFDI creates a larger overseas market for Chinese enterprises; however, it also leads to more emissions of air pollutants as production increases (Zhao & Zhu, 2022). For example, in December 2016, 620,000 km² of the territory were affected by severe smog. Moreover, the Global Environmental Performance Index (GEPI) in 2022 reports that the country ranked 160 of 180 countries and regions, indicating severe environmental pollution. This event creates unprecedented challenges in coordinating AP and OFDI growth for the Chinese government. China has aimed to achieve carbon neutrality before 2060, which requires accelerated efforts to optimize the economic structure and facilitate green technology. In this process, firms must adopt clean technology from home and abroad. Alternatively, they consider moving factories through OFDI in other countries with laxer environmental regulations. In this way, air quality and OFDI may both increase. Considering that the country is undergoing economic transformation and emphasizing sustainable development (Yang et al., 2021), studying the relationship between AP and OFDI has special significance for China.

Our research has the following contributions. First, existing studies mainly discuss the impact of environmental regulations on OFDI (De Beule et al., 2022; Li & Liang, 2022), ignoring the direct impact of AP on OFDI. When AP exists, firms may change investment decisions quickly, expecting environmental costs to increase, while stricter environmental policies may have yet to be introduced immediately. Therefore, existing studies relating to environmental regulations cannot accurately reflect the impact of AP on OFDI. In addition, many studies focus on the relationship between AP and inward foreign direct investment (IFDI) (Ren et al., 2021; Yüksel et al., 2020) while neglecting the direct impact of AP on OFDI. This paper studies the two-way relationship between AP and OFDI in China, which can provide implications for the government to coordinate environmental protection and OFDI and help understand how firms adjust their OFDI decisions according to changes in AP.

Second, existing studies mainly focus on the relationship between AP and OFDI from a static point of view, failing to reveal the dynamics of the relationship (Zhou & Li, 2021). The current study extends the literature on the environment and investment by using the bootstrap rolling-window Granger causality test to examine the two-way relationship between AP and OFDI in different periods. This method considers the evolution of causality throughout subperiods (Balcilar et al., 2010; Su et al., 2020). In addition, this method can help analyze the influence of environmental and economic policies on the relationship between AP and OFDI.

The rest of the paper is arranged as follows. Section 2 reports the literature. Section 3 analyses the interaction mechanism of AP and OFDI in theory. The methodology is presented in Section 4. Section 5 describes the data. Section 6 shows the empirical results with a discussion. The last section concludes the study and provides suggestions.

Literature Review

AP is a significant concern worldwide and is influenced by economic growth (Jiang et al., 2022), public-private partnership investment (Kirikkaleli & Adebayo, 2021a), financial development (Kirikkaleli & Adebayo, 2021b; Rjoub et al., 2021), and foreign direct investment (Tan et al., 2021). The classic pollution haven hypothesis (PHH) states that firms relocate polluting factories to countries with laxer environmental policies, making the host countries that receive IFDI become a “pollution haven” (Copeland & Taylor, 1994; Walter & Ugelow, 1979). Guzel and Okumus (2020) test the PHH by investigating the effect of the IFDI on air quality for the Association of South East Asian Nations (ASEAN), concluding that the IFDI worsens air quality in ASEAN countries. Additionally, Janjua (2021) reports that IFDI had a long-run positive impact on AP. In another study, Ali et al. (2021) show that the IFDI enhanced environmental degradation in Pakistan. Moving on, Abid et al. (2022) confirm that IFDI brings more carbon emissions. In contrast to the above literature, Odugbesan and Adebayo (2020) argue that the IFDI reduces AP in Nigeria both in the short and long run. He et al. (2021) also prove that globalization helps improve the environment. Yasmeen et al. (2022) confirmed that IFDI positively impacted the environment. Using samples from 125 countries, Azam and Raza (2022) find that the IFDI has an insignificant influence on AP in Europe and Latin America. Moreover, Ren et al. (2021) find a two-way relationship between IFDI and carbon dioxide (CO₂). Yüksel et al. (2020) further highlight that carbon emissions have an impeding influence on the IFDI.

In addition to the research on IFDI and AP, several studies have discussed the relationship between OFDI and AP. Some studies investigate the impact of environmental policies on OFDI. Yu et al. (2021) evaluate the impact of the emissions trading system (ETS) on a firm’s OFDI decision, concluding that the ETS has accelerated the depth and breadth of OFDI. De Beule et al. (2022) also show that the European Union ETS plays an essential role in the OFDI location choice of firms regulated by environmental policy. However, the view that stringent environmental policy in home countries promotes OFDI cannot always be supported. Manderson and Kneller (2012) argue that “dirtier” enterprises are not more likely to invest in host countries with lax environmental policies than “cleaner” enterprises are. Naughton (2014) finds that OFDI decreases when the home country’s environmental policy gets stricter. Unlike the above research, Zhang et al. (2021) examine the influence of OFDI on AP, showing that OFDI reduces the carbon footprint in Turkey and Mexico. By using a global sample of 111 countries, Ashraf and Doytch (2022) point out that OFDI reduces the ecological footprints of the home county. Mohanty and Sethi (2022) also prove that OFDI helps to reduce pollution in BRICS countries (Brazil, Russia, India, China, and South Africa). Tanaka et al. (2022) further find that relocating battery recycling from the US to Mexico improves air quality in the US.

Furthermore, as the levels of air pollution and OFDI have increased in China in recent years, many researchers have studied AP and OFDI using Chinese data. Dong et al. (2022) imply that AP induces firms to relocate to other countries with laxer environmental policies through OFDI. Additionally, Gong et al. (2021) find that environmental regulations can promote OFDI. In addition, Liu, Zhao et al. (2021) point out that environmental regulations can amplify the role of OFDI in facilitating green technology. Moreover, Li and Liang (2022) show that environmental regulation has different impacts on OFDI in different regions of China. Liu et al. (2022) further point out that the promoting influence of environmental policies on OFDI is stronger in small and private firms. Regarding OFDI’s impact on AP, Kamal et al. (2022) find that the environmental quality of Belt and Road Initiative (BRI) countries is worsened by China’s OFDI. Zhuang et al. (2023) further prove that OFDI leads to a higher AP in the BRI sample, mainly by promoting tourism. In another study, Cai et al. (2021) show that OFDI increases CO₂ emission intensity. Yang et al. (2021) also reveal that OFDI increased CO₂ emissions in China. Zhao and Zhu (2022) draw a similar conclusion, suggesting that OFDI can increase carbon emissions. The studies mentioned above differ from that of Zhou and Li (2021), who concludes that an OFDI increase can reduce air pollutants. Similarly, Long et al. (2022) indicate that OFDI can improve Chinese firms’ environmental performance. Moreover, Hao et al. (2020) show that the OFDI increases domestic CO₂ emissions by expanding the production scale (scale effect), while OFDI reduces CO₂ by optimizing the domestic economic structure (composition effect) and bringing better technology (technique effect). Dai et al. (2021) and Fahad et al. (2022) further find that when environmental regulations exist, OFDI can promote green technological innovation, which implies that OFDI reduces AP through the technique effect.

The previous literature has mainly investigated the relationship between IFDI and AP (Ali et al., 2021; Azam & Raza, 2022) while neglecting the influence of the home country AP on OFDI. Although some studies have discussed the role of environmental regulations in OFDI (De Beule et al., 2022), they need to reflect the accurate impact of AP on OFDI. In addition, existing studies usually need to pay more attention to the structural breaks in the full sample and reveal the dynamics of the relationship between AP and OFDI. Considering that China has proposed several major environmental and economic reforms in recent years, which may produce structural changes in the association of AP and OFDI, it is necessary to examine the dynamic relationship between the two variables. To fill this gap, the present study uses the bootstrap rolling-window subsample causality test (Balcilar et al., 2010; Su, Qin et al., 2021) to investigate the time-varying causality of AP and OFDI in China.

AP and OFDI Interaction Mechanism

Three fundamental theories can explain the impact of AP on OFDI. The first theory is PHH, which states that firms from dirty industries will relocate to foreign countries with less stringent environmental regulations (Gollop & Roberts, 1983). When severe air pollution exists, the government usually issues environmental regulations requiring firms to use clean technology, pay taxes, and buy pollution discharge permits, which increases firms’ costs (Dong et al., 2022). Thus, AP leads to more OFDI.

The second theory is FEH, which states that firms tend to produce in regions with abundant resources (Mani & Wheeler, 1998). AP usually increases firms’ environmental costs, but firms will continue to produce as long as they can benefit from rich endowments (Leiter et al., 2011). Thus, firms may have funding constraints, which leads to a decrease in their total investment (Greenstone, 2002). Accordingly, we can infer that a higher level of AP leads to less OFDI. However, when the environmental cost is high enough to make domestic production unprofitable, firms will choose to produce in foreign countries with lower environmental standards, and OFDI increases (De Beule et al., 2022; Yu et al., 2021).

The third theory is PH, which states that environmental regulations encourage firms to adopt innovative technology (Dai et al., 2021; Fahad et al., 2022). Accordingly, we infer that AP positively influences OFDI, as firms have the incentive to obtain cleaner technology from home and abroad (i.e., through OFDI).

According to the preceding analysis, we can infer that OFDI decreases as AP increases because a higher AP increases environmental costs, which may lead to financial constraints for firms. However, as AP continues to grow, OFDI will increase because firms may obtain more advanced clean techniques through OFDI to reduce environmental costs or relocate factories to countries with laxer environmental regulations.

Conversely, OFDI also influences AP through three channels. The first channel is the “composition effect,” which states that OFDI reduces AP in the home country by changing the economic structure (Yang et al., 2021). When firms in the polluting industry relocate abroad through OFDI, the domestic economic structure will be upgraded, leading to less air pollution in the home country (Wu & Wang, 2022). The second channel is the “technique effect,” which proposes that firms can obtain advanced clean technology through OFDI, improving air quality in the home country (Fahad et al., 2022). Thus, we can infer from the technique effect that OFDI reduces AP. The third channel is the “scale effect,” which implies that the expansion of production has a negative impact on the environment (Hao et al., 2020). OFDI is often considered an effective way to expand production (Razzaq et al., 2021), leading to more fossil energy consumption and air pollutants emissions (Zhou et al., 2021). Thus, we can infer that OFDI worsens air quality.

In summary, AP and OFDI have a two-way relationship, but there needs to be a consensus on the direction of impacts.

Methodology

Bootstrap Full-Sample Causality Test

The standard Granger causality test is based on the vector autoregression (VAR) model, which assumes that statistics such as the Lagrange multiplier (LM) or likelihood ratio (LR) obey the standard asymptotic distribution in full samples (Sun et al., 2021). However, Sims et al. (1990) and Toda and Phillips (1993, 1994) pointed out that such statistics may not follow the standard asymptotic distribution because structural breaks may exist. Such breaks cause difficulties in estimating VAR models. Shukur and Mantalos (1997) found that estimation performance can be improved by using the residual-based bootstrap (RB) method. Furthermore, Shukur and Mantalos (2000) proved that RB-based corrected LR-statistics have the relatively better size and power properties, even in small samples.

Therefore, we use the RB-based modified LR-statistic to explore the causal relationship between AP and OFDI. The VAR model is shown in equation (1):

$${\mathbf{y}}_t={\mathbf{\Phi }}_0+{\mathbf{\Phi }}_1{\mathbf{y}}_{t-1}+\dots +{\mathbf{\Phi }}_p{\mathbf{y}}_{t-p}+{\mathbf{\epsilon }}_t,t=1,2\dots ,T$$ (1)

where ${\mathbf{y}}_t$ is a column vector of variables, ${\mathbf{\epsilon }}_t$ is the white-noise vector, T is the number of samples, ${\mathbf{\Phi }}_0$ ,… ${\mathbf{\Phi }}_p$ are matrixes of coefficients to be estimated, and p is the lag length. As economic growth can influence AP and OFDI (Ali et al., 2021; Grossman & Krueger, 1991), we choose an industrial added value (IAV) as a control variable. Then, equation (2) is expressed as follows:

$$\begin{array}{c}\left[\begin{array}{c}A{P}_t\hfill \\ \multicolumn{1}{c}{\mathrm{OFD}{I}_t}\end{array}\right]=\left[\begin{array}{c}{\varphi }_{10}\hfill \\ \multicolumn{1}{c}{{\varphi }_{20}}\end{array}\right]+\left[\begin{array}{ccc}{{\varphi }_{11}}^{(1)}\hfill & {{\varphi }_{12}}^{(1)}\hfill & {{\varphi }_{13}}^{(1)}\hfill \\ \multicolumn{1}{c}{{{\varphi }_{21}}^{(1)}}& \multicolumn{1}{c}{{{\varphi }_{22}}^{(1)}}& \multicolumn{1}{c}{{{\varphi }_{23}}^{(1)}}\end{array}\right]\hfill \\ \left[\begin{array}{c}A{P}_{t-1}\hfill \\ \multicolumn{1}{c}{\begin{array}{c}\mathrm{OFD}{I}_{t-1}\hfill \\ \multicolumn{1}{c}{\mathrm{IA}{V}_{t-1}}\end{array}}\end{array}\right]+\dots +\left[\begin{array}{ccc}{{\varphi }_{11}}^{(2)}\hfill & {{\varphi }_{12}}^{(2)}\hfill & {{\varphi }_{13}}^{(2)}\hfill \\ \multicolumn{1}{c}{{{\varphi }_{21}}^{(2)}}& \multicolumn{1}{c}{{{\varphi }_{22}}^{(2)}}& \multicolumn{1}{c}{{{\varphi }_{23}}^{(2)}}\end{array}\right]\hfill \\ \left[\begin{array}{c}A{P}_{t-2}\hfill \\ \multicolumn{1}{c}{\begin{array}{c}\mathrm{OFD}{I}_{t-2}\hfill \\ \multicolumn{1}{c}{\mathrm{IA}{V}_{t-2}}\end{array}}\end{array}\right]+\dots +\left[\begin{array}{ccc}{{\varphi }_{11}}^{(p)}\hfill & {{\varphi }_{12}}^{(p)}\hfill & {{\varphi }_{13}}^{(p)}\hfill \\ \multicolumn{1}{c}{{{\varphi }_{21}}^{(p)}}& \multicolumn{1}{c}{{{\varphi }_{22}}^{(p)}}& \multicolumn{1}{c}{{{\varphi }_{23}}^{(p)}}\end{array}\right]\hfill \\ \left[\begin{array}{c}A{P}_{t-p}\hfill \\ \multicolumn{1}{c}{\begin{array}{c}\mathrm{OFD}{I}_{t-p}\hfill \\ \multicolumn{1}{c}{\mathrm{IA}{V}_{t-p}}\end{array}}\end{array}\right]+\left[\begin{array}{c}{\epsilon }_{1t}\hfill \\ \multicolumn{1}{c}{{\epsilon }_{2t}}\end{array}\right].\hfill \end{array}$$ (2)

In equation (2), when ${{\varphi }_{21}}^{(q)}=0$ , (q = 1, 2, …, p), AP is not a Granger cause of OFDI. Likewise, when ${{\varphi }_{12}}^{(q)}=0$ , the null hypothesis that OFDI is not a Granger cause of AP can be tested.

Parameter Stability Test

The parameters of the VAR model are required to be constant in the full-sample causality test (Su, Cai et al., 2021). However, if structural breaks exist in the time series, then parameters in the VAR model may not be constant, which leads to an unreliable result of the full-sample test (Balcilar & Ozdemir, 2013). Therefore, the stability of parameters should be tested.

This study uses the Sup-F, Mean-F, and Exp-F tests proposed by Andrews (1993) and Andrews and Ploberger (1994) to check the parameters’ stability. In addition, we apply the Lc test suggested by Nyblom (1989) and Hansen (1992) to test the stability of all parameters in VAR jointly. These tests can be used to check the parameters’ stability to determine whether structural mutations exist at unknown time points.

Rolling-Window Subsample Causality Test

When structural breaks exist in the full-sample data, devices such as dividing the samples or using a dummy variable can be employed to solve this problem, but biases still exist. This study uses the rolling-window bootstrap (Balcilar et al., 2010), which allows us to observe the changing casualty of AP and OFDI in different subsamples and avoid biases (Su, Huang et al., 2021).

The rolling-window subsample causality test is proposed to divide the full sample into fixed-size subsamples for causality testing. Specifically, the full-sample length, F, can be converted to a sequence of F−W + 1 subsamples, where W is the length of the fixed-size rolling window. Then, the subsamples are η−W + 1, η−W + 2, …, η, where η = W, W + 1, …F. Then, we can test the causality for each subsample. The confidence interval is 90% (Balcilar et al., 2010).

Data

We use monthly data of AP and OFDI from January 2013 to September 2022. At the beginning of 2013, air pollution in China was severe, with 30 provinces in China experiencing four serious haze events. Confronted with the horrible air pollution, the policymakers proposed comprehensive measures to control AP and began releasing the Air Quality Index (AQI). This index was first reported in January 2013 and is widely used as a comprehensive index to measure the status of urban AP in China. The AQI is calculated based on six air pollutants, namely, sulfur dioxide (SO₂), respirable particulate matter (PM₁₀), nitrogen dioxide (NO₂), fine particulate matter (PM_2.5), carbon monoxide (CO), and ozone (O₃). Higher AP values indicate more serious air pollution. In addition, we use the nonfinancial outward direct investment to represent OFDI. The data for AP and OFDI are from the CEIC Data.

Figure 1 shows the trends of AP and OFDI. We can find that although changes in AP present seasonal characteristics, the overall annual trend of AP values has been declining, suggesting that the air quality has gradually improved. This situation has been caused mainly by China’s emphasis on environmental governance in recent years. For example, the revised Law on the Prevention and Control of Atmospheric Pollution (LPCAP) came into force in January 2016, which required polluting firms to reduce emissions of pollutants. Affected by this, AP dropped sharply. In addition, the changes in OFDI also show a fluctuating characteristic, and the positive value of monthly OFDI indicates increasing OFDI stock in China. Moreover, the trend of AP coincides with OFDI in many periods. For example, in the winter of 2019, AP showed a rising trend because of seasonal factors. During this time, OFDI values also increased, indicating that AP may lead to more OFDI. In the second half of 2021, the Ministry of Commerce issued the 14th Five-Year Plan (2021–2025) for business development, proposing to improve the level of OFDI and foreign economic cooperation. Encouraged by the policy, OFDI increased dramatically in December 2021. During this time, AP values also increased. However, the trends of AP and OFDI are sometimes different. In the winter of 2016, AP rose substantially, and simultaneously, OFDI declined sharply. In addition, the changes in AP and OFDI may be affected by economic growth (Ali et al., 2021; Grossman & Krueger, 1991). On the one hand, economic growth promotes OFDI, which has a certain impact on the emission of air pollutants. On the other hand, waste discharge differs in different economic development statuses. As the industry is the main sector contributing to economic growth and AP in China, we choose IAV as the control variable. In summary, AP and OFDI have a time-varying relationship, which is also influenced by economic growth.

Figure 1.

Trends of AP and OFDI.

The descriptive statistics of the variables are reported in Table 1. The mean AP is 4.747, and the average values of OFDI and IAV are 10.703 billion dollars and 22.045 trillion dollars, respectively. In addition, the positive skewness shows that AP, OFDI, and IAV follow a right-skewed distribution. The kurtoses of AP and OFDI are greater than 3, demonstrating a leptokurtic distribution. The kurtosis of IAV is less than 3, showing a platykurtic distribution. Furthermore, the Jarque–Bera test suggests that AP and OFDI obey a nonnormal distribution at the 1% level of significance. The subsequent analysis transforms AP, OFDI, and IAV by taking natural logarithms. All variables are adjusted to eliminate seasonal factors to avoid potential heteroscedasticity and possible instability.

Table 1.

Descriptive Statistics of the Sequence of AP, OFDI, and IAV.

Statistics	AP	OFDI	IAV
Observations	117	117	117
Mean	4.747	10.703	22.045
Median	4.453	9.26	21.861
Maximum	11.339	52.89	33.124
Minimum	2.615	4.31	13.391
Std. Dev.	1.437	6.177	4.373
Skewness	1.371	4.013	0.162
Kurtosis	6.166	24.275	2.332
Jarque–Bera	85.525***	2,520.572***	2.687

***

indicates that the statistics are significant at the 1% level.

Empirical Results and Discussions

Before examining the Granger causality, this paper performs the ADF (Dickey & Fuller, 1981), PP (Phillips & Perron, 1988), and KPSS (Kwiatkowski et al., 1992) unit root tests. Table 2 shows that all sequences are stationary.

Table 2.

Unit Root Tests.

Series	ADF	PP	KPSS
AP	−4.521(0)***	−5.155[6]***	0.109[8]
OFDI	−7.866(0)***	−8.466[6]***	0.244[7]
IAV	−7.551(0)***	−7.5841[1]***	0.057[4]

Note. The number in parentheses indicates the lag order, which is selected based on the SIC.

The number in the brackets refers to the bandwidth, which uses the Bartlett Kernel as suggested by the Newey–West test (1987). The null hypothesis for KPSS is that the time series is stationary.

***

Denotes significance at the 1% level.

Then, we construct the VAR models and conduct the full-sample causality test. According to the final prediction error (FPE) and Akaike information criterion (AIC), the lag order is 3. Table 3 shows the full-sample causality results. As the p-value is .43, which shows that AP is not a Granger cause of OFDI. Similarly, the hypothesis that OFDI is not a Granger cause of AP cannot be rejected at the significance level of 10%. Therefore, AP and OFDI do not influence each other, which is inconsistent with the literature (Dong et al., 2022; Mohanty & Sethi, 2022).

Table 3.

Full-Sample Granger Causality tests.

	H0: AP is not a Granger cause of OFDI		H0: OFDI is not a Granger cause of AP
Tests	Statistics	p-Values	Statistics	p-Values
Bootstrap LR test	2.555	.43	2.142	.525

Note. The null hypothesis is that no-causal relationship exists between the variables. p-values are calculated using 10,000 bootstrap repetitions.

Considering that when structural changes exist, the causality of variables may change over time, we proceed to test the parameters’ stability and determine whether structural changes exist. Table 4 shows the results.

Table 4.

Parameter Stability Tests.

	OFDI equation		AP equation		VAR(3) system
	Statistics	Bootstrap p-value	Statistics	Bootstrap p-value	Statistics	Bootstrap p-value
Sup-F	29.269***	.003	27.019***	.008	33.010*	.056
Mean-F	18.467***	.000	12.233**	.036	20.230*	.053
Exp-F	12.006	.001	9.930***	.009	13.082*	.053
Lc					2.856	.167

Note. We calculate p-values using 10,000 bootstrap repetitions. *, **, and *** indicate significance at the 1%, 5%, and 10% levels, respectively. Lc shows the results of the Hansen–Nyblom parameter stability test for all parameters in the VAR jointly.

The results of Sup-F show that sudden structural breaks exist in the OFDI equation and AP equation at the 1% level of significance and the VAR system at the 10% level of significance. Mean-F and Exp-F tests are applied to test the null hypothesis that parameters follow a martingale process. The results show that the null hypothesis is rejected, indicating that the OFDI equation, AP equation, and VAR system will evolve with time. In summary, the results above show that the parameters are unstable when using the full-sample data, and structural changes exist. To solve this problem, we use RB-based subsample modified-LR causality tests to check the causal relationship between AP and OFDI. The rolling subsample data include 24 months of observations.

Figure 2 shows the rolling bootstrap of the p-values of the LR-statistics using OFDI as the dependent variable. The null hypothesis, namely, AP is not a Granger cause of OFDI, is significantly rejected in 2016:M3–2016:M10 and 2019:M11–2020:M1. Figure 3 depicts AP’s influence on OFDI. Specifically, AP exerts a negative effect on OFDI from 2016:M3 to 2016:M10, which implies that AP impedes OFDI. However, AP promotes OFDI from 2019:M11 to 2020:M1.

Figure 2.

Bootstrap p-value of the statistics (the null hypothesis is that AP is not a Granger cause of OFDI).

Figure 3.

Sum of rolling-window coefficients of AP’s influence on OFDI.

In January 2016, the revised LPCAP came into force, requiring firms to reduce air pollutants emissions. This prompts firms to invest more money to meet environmental regulations, increasing financing constraints (Gong et al., 2021). In addition, as investors tend to expect a higher risk for firms (Liu et al., 2018), it is more difficult for such firms to raise funds. Thus, the lack of funds impedes OFDI in 2016:M3–2016:M10. This result also supports FEH, which implies that firms will continue to produce in the home country when confronted with environmental regulations as long as the benefits from rich endowments outweigh the costs from regulation (Leiter et al., 2011).

However, AP exerts a positive effect on OFDI during 2019:M11-2020:M1. The LPCAP was revised again in 2018 and came into force in 2019, further increasing the cost of pollution control for firms. When it is unprofitable to produce in the home country, firms will consider relocating to foreign countries with laxer environmental policies (De Beule et al., 2022). In addition, confronted with stricter regulations, some firms may obtain advanced clean technology through OFDI to meet environmental standards (Fahad et al., 2022). Moreover, China has recently established a new system for the higher-level open economy, promoting an increase in OFDI. Hence, we observe a positive influence of AP on OFDI during 2019:M11-2020:M1. This finding is consistent with PHH and PH, implying that serious air pollution and the corresponding regulations encourage polluting firms to relocate to other countries.

Figure 4 depicts the p-values of the LR-statistic, testing the null hypothesis that OFDI is not a Granger cause of AP. The p-value is less than .1 in 2019:M3–2019:M8 and 2020:M10–2021:M4. Figure 5 shows the sum of the rolling-window coefficients for the impact of OFDI on AP. Combining Figures 4 and 5, we can conclude that OFDI reduces AP in 2019:M3–2019:M8 while increasing AP in 2020:M10–M10–2021

Figure 4.

Bootstrap p-value of the statistics (the null hypothesis is that OFDI is not a Granger cause of AP).

Figure 5.

Sum of rolling-window coefficients of OFDI’s influence on AP.

The negative influence of OFDI on AP proves the “composition effect,” which indicates that OFDI improves air quality in the home country by changing the economic structure (Yang et al., 2021). Polluting firms relocate abroad through OFDI, which is helpful for the upgrade of the domestic economic structure, leading to less pollution in China (Hao et al., 2020). In the Government Work Report of 2019, the Chinese government proposed upgrading the economic structure and pushing the “supply-side structural reform,” aiming to change the mode of economic development from one that is a factor- and investment-driven to an innovation-driven one (Tao et al., 2022). This promotes polluting firms with difficulty achieving industrial upgrades to relocate factories to less developed countries. In this way, air pollution can also be reduced. In addition, this result also proves the “technique effect,” which suggests that firms can obtain advanced green technology through OFDI, leading to less air pollution.

However, OFDI leads to more AP in 2020:M10–2021:M4, which is consistent with the “scale effect,” indicating that OFDI pollutes the air by expanding production. At the end of 2020 and early 2021, Chinese high-energy-consuming industries received many overseas orders because they recovered more quickly from COVID-19 than competitors, resulting in substantial fossil energy consumption. Although the government encourages the use of renewable energy, coal still accounts for over 50% of energy consumption in the country (Guo et al., 2016; Zhang, 2020). Hence, OFDI leads to a rising AP during 2020:M10–2021:M4.

In summary, we apply RB-based modified-LR causality tests to examine the time-varying causal relationship between AP and OFDI and draw three main conclusions. First, the influence of AP on OFDI was negative in 2016:M3–2016:M10. This finding is consistent with the FEH (Leiter et al., 2011). When the newly revised LPCAP came into force in 2016, firms needed to bear the cost of controlling pollution and faced financing constraints (Gong et al., 2021). Nevertheless, firms could still benefit from rich endowments in China; hence, they continued to produce in the country. Thus, they had fewer funds for further OFDI. This finding supports Naughton (2014), who finds that OFDI decreases when environmental regulation becomes stricter in the home country. Second, AP promoted OFDI in 2019:M11–2020:M1. This result confirms the coexistence of PHH and PH. As environmental regulations become stricter, firms might obtain more green techniques from other countries through OFDI to reduce costs or relocate factories to countries with laxer environmental regulations. Therefore, AP in the home country may promote OFDI. This finding supports other studies conducted in this area (Dong et al., 2022; Gong et al., 2021; Liu, Zhao et al., 2021), but they mainly focus on the influence of environmental regulations on OFDI, neglecting the direct impact of AP. For example, Gong et al. (2021) show that environmental regulations significantly impact OFDI growth. Dong et al. (2022) find that strict environmental regulations induce firms to relocate factories to other countries through OFDI. Third, different from several other studies that OFDI leads to more AP (Zhao & Zhu, 2022) or OFDI reduces AP (Mohanty & Sethi, 2022; Tanaka et al., 2022; Zhou & Li, 2021), this paper finds that OFDI has both positive and negative influences on AP in different periods. On the one hand, the OFDI of polluting firms reduces the AP of the home country by optimizing the economic structure or by obtaining green technology. On the other hand, OFDI exacerbates AP by increasing fossil fuel consumption, which increases air pollution emissions.

Conclusion

This paper investigates the causal relationship between AP and OFDI in China. The rolling window results display a two-way causality between AP and OFDI across various subsamples. First, in 2016, when environmental regulations were revised, AP inhibited OFDI mainly by increasing the environmental cost of firms. This result proves the FEH, indicating that environmental regulations raise costs and reduce OFDI. Still, firms will continue to produce in the country if they can benefit from rich endowments. Second, AP promoted OFDI in 2019, when environmental regulations and opening-up policies were further strengthened. This result supports the PHH and PH, which suggest that when confronted with AP, firms tend to circumvent environmental regulations through OFDI. Third, a negative causal relationship running from OFDI to AP in 2019 is observed. This result is consistent with the “composition effect,” implying that OFDI can reduce air pollution by improving the economic structure of the home country. It also complies with the “technique effect,” showing that firms can obtain green technology through OFDI, thereby reducing the AP of the home country. Fourth, OFDI contributed to a higher level of AP from the end of 2020 to the beginning of 2021, which is consistent with the “scale effect,” indicating that OFDI pollutes the air by expanding production. Therefore, we confirm that AP and OFDI can influence each other and that environmental and economic policies can change such a relationship.

Understanding the interaction between AP and OFDI provides implications for the government. First, when AP exists, firms may obtain green technology through OFDI to meet regulatory requirements, ultimately benefiting the upgrading of the industrial structure. Hence, appropriate environmental regulation is essential. Second, policymakers should realize that the strictness of environmental regulation will affect firms’ investment decisions, further influencing economic growth. Hence, policymakers should coordinate environmental regulation with domestic economic growth to achieve sustainable development. Third, as OFDI can improve air quality, the government should adhere to the opening-up policy to achieve continuous OFDI growth and carbon neutralization. Fourth, policymakers should realize that OFDI is conducive to expanding production, which may worsen air quality by increasing the consumption of fossil fuels. Hence, the government should encourage renewable energy and promote green technological innovation when developing the economy. The interaction between AP and OFDI can also provide suggestions for firms. AP increases the cost of firms, thereby impeding OFDI. Hence, firms should encourage green technological innovation to reduce environmental costs. In addition, confronted with strict environmental regulations, although firms can relocate factories to foreign countries through OFDI to reduce the cost of environmental governance, they still need to consider other risks, such as geopolitical conflicts.

This paper has some limitations that can be considered recommendations for future studies. First, this paper investigates the relationship between AP and OFDI in China while not considering other countries. AP has different impacts on OFDI in countries with varying economic development levels. Future studies can examine whether AP is always a push for OFDI in other countries and compare the results with ours. Second, the period is relatively limited. Using the sample from January 2013 to September 2022, we find that AP promoted OFDI in 2019. This indicates that stricter environmental regulations force firms to relocate factories to other countries with laxer environmental standards. However, we have yet to determine whether this phenomenon will be a long-term trend. Hence, future studies could extend the sample period to see whether this conclusion still holds. However, the current period is enough to draw a meaningful conclusion that AP has both positive and negative effects on OFDI.